ABSTRACT. In this work, we propose an integrated pipeline for generating synthetic transthoracic echocardiographic (TTE) images from computed tomography (CT) volumes, together with spatially aligned segmentation masks. The framework combines multi-label anatomical segmentation (Total Segmentator), tissue-classbased echogenicity modelling, geometry-aware plane extraction, scatterer-based ultrasound simulation (MUST simulator), and CycleGAN-based image refinement within a unified workflow. Unlike approaches based on simplified anatomical models, the proposed method exploits patient-specific CT data to preserve anatomical variability and maintain spatial consistency across all generated outputs. The simulation stage produces physically grounded B-mode images, while the subsequent domain adaptation step reduces the visual gap between synthetic and real echocardiographic data. In addition to qualitative assessment, preliminary proof-ofconcept experiments were conducted to evaluate the use of the generated data for training a semantic segmentation network in a low-data regime. The progressive addition of synthetic images to a limited set of real annotated samples improved segmentation performance, supporting the feasibility of the proposed framework as a source of annotated synthetic data. Although the present implementation focuses on TTE image generation, the modular structure of the pipeline could be extended to other anatomical regions, enabling the generation of synthetic ultrasound images from CT-derived anatomical information in different body districts. Overall, the proposed framework provides a flexible and controllable approach for generating synthetic echocardiographic datasets, with potential applications in training, validation, and benchmarking of AI-based image analysis algorithms, while reducing the dependence on manually annotated clinical datasets.

ABSTRACT. Antimicrobial resistance has intensified the need for alternative anti-infective strategies, particularly against clinically relevant bacterial pathogens. In this study, three short synthetic peptides from the L-series—L1, L2, and L3—were rationally selected based on high predicted antimicrobial peptide probability and evaluated against Staphylococcus aureus, Escherichia coli, Pseudomonas aeruginosa, and Enterococcus faecalis. Peptides were synthesized by solid-phase peptide synthesis, purified, and tested using a broth microdilution-associated alamarBlue assay and minimum inhibitory concentration (MIC) determination. In parallel, computational docking was used to explore peptide interaction with the transglycosylase domain of E. coli penicillin-binding protein 1b (PBP1b), a relevant target in bacterial cell-wall biosynthesis. Experimental results showed distinct strain-dependent antibacterial profiles for the three peptides despite their closely related amino acid composition. L1 exhibited the broadest and most balanced activity, with its strongest effect against E. faecalis (MIC 4 µg/mL), whereas L2 showed preferential activity against E. coli (MIC 8 µg/mL) and L3 was most active against P. aeruginosa (MIC 8 µg/mL). Docking analysis supported the experimental findings by indicating sequence-dependent differences in predicted target recognition. Overall, the study demonstrates that subtle residue rearrangements within short cationic peptides can substantially influence antibacterial spectrum and potency, supporting the integration of computational prioritization and experimental screening as an efficient strategy for early-stage antimicrobial peptide discovery.

ABSTRACT. Mammography-based breast cancer screening, while clinically effective, remains highly resource-intensive due to the low prevalence of malignancy among screened populations and the widespread reliance on independent double reading to ensure diagnostic accuracy. Deep learning systems have shown considerable promise in providing support to radiologists; nevertheless, several obstacles persist, including the inherent class imbalance arising from low cancer rates, the computational demands of handling very high-resolution imaging data, and the complexity of integrating complementary information across the multiple views acquired in a standard examination. In this paper, we develop and evaluate a transformer-based AI triage system for screening mammography, designed to reduce clinical workload while preserving high sensitivity. The model was trained on a large dataset comprising more than 2.5 million images and assessed on two independent test sets including out-of-distribution vendors and women of different nationalities, to evaluate its generalization capabilities across heterogeneous clinical settings. The proposed architecture builds on an improved TinyViT backbone to efficiently process the four standard mammographic views at high resolution, subsequently combining view-level representations into a single exam-level prediction. The model achieved AUROC values ~97%, with standalone sensitivities of ~98%. When combined with human readers, the model reduced the workload of the second reader by approximately 75%, while simultaneously improving positive predictive value and lowering recall rate relative to standard double reading. Benchmarked against two reference multi-view architectures, the proposed model demonstrated superior overall discrimination, supporting transformer-based AI triage as a promising tool for workflow optimization in mammography screening programs operating across diverse clinical environments.

ABSTRACT. The functions of PPARGC1A, FTO, and MC4R as metabolic genes that connect training adaptation, athletic performance, and metabolic health are examined in this narrative review. The rs8192678 polymorphism of PPARGC1A, which codes for PGC-1α, a master regulator of mitochondrial biogenesis and oxidative metabolism, has been linked to aerobic performance and endurance capacity, even though, the effects vary depending on the sport and ethnicity. In turn, MC4R and FTO are essential for controlling adiposity and energy balance. MC4R helps regulate hunger and activity-behavior, while FTO variations affect how the body responds to exercise. Exercise and training load can alter their phenotypic expression, and evidence suggests that polygenic models integrating these and related loci increase prediction of obesity and metabolic characteristics beyond single-gene effects. The need for more detailed, longitudinal, and genome-integrated research is highlighted by methodological limitations, such as limited sample numbers, heterogeneous phenotyping, and a lack of standardized training-intervention studies in athletes. PPARGC1A, FTO, and MC4R together demonstrate how genetic diversity interacts with lifestyle, nutrition, and training to shape athletic potential and metabolic resilience. This suggests a future in precision sports medicine that combines functional and metabolic phenotyping with genetic profiling.

ABSTRACT. Osteogenic commitment is a complex and tightly regulated process that involves coordinated changes at both the morphological and transcriptional levels. As cells progressively acquire an osteoblastic phenotype, they undergo structural and transcriptional rearrangements that reflect the activation of lineage-specific programs, which are critical for proper bone formation and tissue homeostasis. The aim of this study is to determine whether nuclear eccentricity correlates with transcriptional activation during osteogenic commitment, providing a potential quantitative marker to monitor early differentiation events. Among the most evident morphological alterations is the variation in nuclear shape and spatial organization, which mirrors chromatin reorganization associated with differentiation and may reflect underlying cytoskeletal remodeling and mechanical adaptations. Nuclear architecture is increasingly recognized as a sensitive indicator of cell fate transitions, as it integrates mechanical cues, chromatin dynamics, and transcriptional activity within a dynamic three-dimensional nuclear environment. In this project, nuclear eccentricity will be analyzed in MC3T3 cells, a well-established preosteoblastic cell line widely used as an in vitro model of osteogenesis. These cells represent a reliable system to investigate both early and late events associated with osteogenic differentiation, providing insights into cellular behavior during lineage commitment. Cells will be cultured for 21 days under osteogenic induction conditions using a differentiation medium containing specific osteoinductive supplements such as ascorbic acid, β-glycerophosphate, and melatonin, and will be compared with undifferentiated control cells maintained in standard growth medium. This experimental design will allow a comprehensive assessment of both morphological and molecular changes associated with the progression of osteogenic commitment. Nuclear morphology will be evaluated through DAPI staining followed by fluorescence microscopy imaging, enabling clear visualization of nuclear contours and overall nuclear organization. A quantitative morphometric analysis will be performed to determine nuclear eccentricity as a descriptive parameter of nuclear remodeling. Acquired images will be analyzed using ImageJ software to calculate eccentricity values, allowing objective, reproducible, and precise quantification of nuclear shape. This approach will allow discrimination between more rounded nuclei (characterized by low eccentricity values) and more elongated nuclei (characterized by high eccentricity values), providing a measurable indicator of structural adaptation and cellular remodeling. In parallel, gene expression analysis of major osteogenic markers, including both early and late differentiation genes (RUNX2, Alkaline phosphatase, Osteocalcin, Osteopontin, Osteonectin, Osterix, Collagen type I) will be performed using qRT-PCR. This combined evaluation of morphological and molecular parameters will provide insight into the relationship between nuclear remodeling and osteogenic transcriptional activation. Overall, this project aims to establish nuclear eccentricity, together with osteogenic gene expression, as complementary and reliable parameters for monitoring osteogenic commitment and assessing whether cellular differentiation is effectively progressing toward the correct lineage fate in a controlled in vitro system.

ABSTRACT. Accurate assessment of dental apices in pediatric panoramic radiographs is a fundamental pillar for proper endodontic treatment planning. The integration of artificial intelligence is transforming radiographic analysis, offering crucial diagnostic support tools even in pediatric dentistry. This study analyzes the effectiveness of a transfer learning approach using the YOLOv8 model for identifying mandibular first permanent molars and their root apices. The primary objective is to test the model’s accuracy in identifying root apices, a critical step for subsequent apical classification and a discriminating parameter between apexogenesis and apexification processes during clinical assessments. The system was validated using spatial accuracy, segmentation, and image quality metrics, including the Dice coefficient, mean precision (mAP), and structural similarity index (SSIM). The obtained data confirm the high performance of YOLOv8, which achieved a Dice score of 0.93 ± 0.06 for molars and 0.81 ± 0.08 for apices, and a mAP@50 of 0.93 and 0.90 for molar and apical detection tasks, respectively. The SSIM results also highlight a remarkable ability to preserve the finest anatomical details. Overall, the study demonstrates the validity of single-stage detection models in processing thin, low-contrast structures, typical of apical regions.

ABSTRACT. Non-invasive glucose monitoring remains a key challenge in Type 1 diabetes mellitus. Among emerging alternatives to subcutaneous continuous glucose monitoring (CGM), electrocardiogram (ECG)-based approaches have shown potential, although existing studies are largely limited to fully subject-specific models. This study evaluates ECG-based continuous glucose estimation with particular focus on the role of personalization, leveraging a newly collected paediatric T1DM dataset acquired in real-life conditions. A two-stage feature extraction framework was applied to 6-second ECG windows preceding each CGM measurement, generating 160 statistical descriptors derived from beat-level morphological features. Three modelling strategies based on multilayer perceptron networks were compared: subject-specific (SS), subject-independent (SI), and subject-independent with fine-tuning (SI+FT). Performance was assessed using Clarke Error Grid analysis, distinguishing between subjects monitored with CGM devices sampling every 5 minutes and those sampling every 15 minutes. In the 5-minute sampling group, SS and SI+FT achieved comparable clinical accuracy (97.33% ± 3.16 vs 97.01% ± 4.27 in Zones A+B), while SI showed lower performance (90.63% ± 11.86). In the 15-minute sampling group, performance decreased across all strategies. These results indicate that fine-tuning enables performance comparable to fully personalized models while supporting scalable ECG-based glucose estimation.

ABSTRACT. Wearable devices enable large-scale and non-invasive sleep monitoring by continuously capturing physiological signals in real-world settings. However, the relative contribution of different sensing modalities and the impact of apnoea-related variability on model performance remain insufficiently understood in wearable sleep/wake classification.

In this work, we present a controlled experimental study to investigate the role of multimodal wearable signals and apnoea-related variability. We evaluate multiple modality configurations, ranging from accelerometry alone to combinations with additional physiological signals, and assess robustness under different training and testing conditions defined by the presence or absence of apnoea events. A lightweight 1D convolutional neural network is adopted and kept fixed across experiments to isolate the effect of input configurations.

Experiments on the DREAMT dataset show that combining accelerometry with heart rate consistently improves performance over both single-modality and full multimodal configurations, reaching an F1-score of 0.766, outperforming both single-modality and full multimodal configurations. In addition, results reveal a significant impact of apnoea-related variability, with performance degradation under mismatched training/testing conditions, where apnoea events are present in only one of the two datasets, indicating the presence of distribution shifts.

These findings highlight the importance of modality selection and physiological variability in wearable sleep analysis, and show that simple models can achieve competitive performance when paired with appropriate input representations.

ABSTRACT. In the context of rapid technological advancement and the emergence of Industry 4.0, education systems are undergoing a profound transformation, particularly within vocational education and training (VET). The increasing demand for a workforce equipped with both theoretical knowledge and practical, technology-driven skills has accelerated the adoption of Education 4.0 approaches, especially in interdisciplinary fields such as biomedical engineering, where engineering principles are directly applied to healthcare challenges. This paper explores the integration of key enabling technologies, including artificial intelligence (AI), virtual reality (VR), and 3D printing, alongside Massive Open Online Courses (MOOCs) and Open Educational Resources (OER), as tools for enhancing learning outcomes and strengthening digital competencies in vocational education. The study highlights how these technologies facilitate experiential, flexible, and personalized learning environments, enabling students to acquire hands-on skills aligned with real-world industry requirements. In the context of biomedical engineering, such approaches are particularly relevant for the development of competencies related to medical device design, simulation of clinical procedures, and analysis of biomedical data. Special attention is given to the role of 3D printing and immersive VR environments in bridging the gap between theoretical instruction and practical application, while AI-driven tools support adaptive learning and data-informed educational processes. Furthermore, MOOCs and OER contribute to increased accessibility and scalability of high-quality educational content, including specialized topics in healthcare technologies. Despite the significant potential of these technologies, the paper also identifies key challenges related to infrastructure, institutional readiness, and pedagogical adaptation, which continue to limit their widespread implementation. The findings suggest that a strategic and integrated approach to Education 4.0 is essential for modernizing vocational education systems and ensuring their alignment with the evolving demands of the digital economy, particularly within the rapidly advancing biomedical engineering sector.

ABSTRACT. Photoplethysmography (PPG) has become a foundational sensing modality in both clinical monitoring and wearable health technologies. Beyond pulse oximetry, PPG waveforms are increasingly utilised for pulse wave analysis (PWA) and morphology-based feature extraction in applications such as cuffless blood pressure estimation, vascular ageing assessment, arterial stiffness evaluation, and cardiovascular risk modelling. As reliance on PPG-derived biomarkers expands, an important question arises: are waveform-derived features consistent across individuals with diverse skin tones? Emerging evidence suggests that device performance may vary systematically with skin pigmentation, raising concerns regarding potential bias in PPG-based technologies. However, whether skin tone is associated with differences in waveform morphology itself remains insufficiently understood.

In this study, we extracted features from signals observed from the Open Oximetry dataset and analysed them to explore potential differences across skin tone groups. Initial statistical analysis suggested differences in several temporal features, indicating possible variation in waveform morphology. When accounting for multiple comparisons, these differences did not survive statistical robustness. Effect size analysis nevertheless revealed small to moderate trends across a subset of features, suggesting that any underlying variation may be subtle and distributed rather than strongly pronounced. These findings emphasize the need for careful statistical treatment and appropriate units of analysis when studying physiological signals and highlight the importance of further investigation using larger and more diverse datasets.

ABSTRACT. Ovarian cancer is the most lethal gynecologic malignancy: around 60% of patients are diagnosed at an advanced stage, with an associated 5-year survival rate of about 30%. Early identification of non-responders to neoadjuvant chemotherapy remains a key unmet need, as it could prevent ineffective therapy and avoid delays in optimal surgical management. This work proposes a non-invasive deep learning framework to predict neoadjuvant chemotherapy response from pre-treatment contrast-enhanced CT by leveraging automatically derived 3D lesion masks. The approach encodes axial slices with a partially fine-tuned pretrained image encoder and aggregates slice-level representations into a volumetric embedding through an attention-based module. Training combines classification loss with supervised contrastive regularization and hard-negative mining to improve separation between ambiguous responders and non-responders. The method was developed on a retrospective single-center cohort from the European Institute of Oncology (Milan, IT), including 280 eligible patients (147 responder, 133 non-responder). On the test cohort, the model achieved a ROC-AUC of 0.73 (95% CI: 0.58–0.86) and an F1-score of 0.70 (95% CI: 0.56–0.82). Overall, these results suggest that the proposed architecture learns clinically relevant predictive patterns and provides a robust foundation for an imaging-based stratification tool.

ABSTRACT. Continuous Respiratory Rate monitoring is an important early indicator of clinical deterioration. Although non-invasive sensors—such as wearable Holters and smartwatches—can acquire electrocardiogram and photoplethysmogram signals, reliably extracting respiratory modulations remains challenging due to artifact corruption and intrinsic morphological variability. To bridge the gap between clinical accuracy and edge-device constraints, we propose UT-RespNet, a lightweight multimodal deep learning architecture. Combining 1D U-Net feature extractors with a Transformer-based late-fusion mechanism, UT-RespNet processes 32-second multimodal windows. The U-Net encoders capture high-resolution local morphologies via skip connections, while the Transformer attention mechanism learns long-term global temporal dependencies, dynamically weighting the most reliable modality. To ensure unbiased clinical validation, UT-RespNet was evaluated on the public BIDMC dataset (53 critically ill patients) using Leave-One-Patient-Out CrossValidation. Setting a new State-of-the-Art performance, our approach achieved a Mean Absolute Error (MAE) of 1.16 ± 2.59 Breaths Per Minute (BPM) on clean data, maintaining a MAE of 1.24 ± 1.88 BPM under out-of-distribution noise on the test sets. UT-RespNet achieves this accuracy while requiring 1.2M parameters and 2.21 ms of inference time, demonstrating its viability for continuous Respiratory Rate monitoring on resource-constrained ambulatory devices.

ABSTRACT. Chronic tinnitus is commonly accompanied by sleep disturbance, emotional distress, and impaired attentional regulation, yet effective non-pharmacological interventions remain limited. We developed a wearable closed-loop EEG-guided mindfulness system for home-based tinnitus intervention and evaluated its feasibility and preliminary efficacy in a randomized controlled study. Sixty patients with chronic tinnitus were enrolled, and 57 completed the protocol. Participants were assigned to an experimental group receiving real-time EEG-contingent feedback or to a control group receiving sham feedback during daily 20-min mindfulness training over 1 month. Linear mixed-effects modeling showed a significant group-by-time interaction for Tinnitus Handicap Inventory (THI), indicating a greater reduction in tinnitus-related handicap in the experimental group than in the control group. Within the experimental group, tinnitus severity and anxiety decreased, mindfulness level increased, and subjective sleep improved. EEG measures in the experimental group further showed a reduced theta/beta ratio and an increased decoded mindfulness score after intervention, suggesting improved attentional regulation. These findings support the feasibility of wearable closed-loop EEG-guided mindfulness training as a home-based, non-pharmacological intervention for chronic tinnitus and suggest potential benefits for tinnitus-related burden, sleep, and brain-state regulation.

ABSTRACT. Pressure injuries represent one of the main complications in individuals forced to remain in the same position for prolonged peri- ods. Frequent repositioning has been shown to reduce the incidence of pressure ulcers. This work evaluated how five different supine postures influenced pressure distribution using the public PoPu dataset. For each configuration, median pressure and contact area were computed on sen- sors exceeding a 5 mmHg threshold. The analysis was performed by com- paring global (full-body) metrics with a regional evaluation focused on the torso (Region of Interest (ROI)), an area particularly exposed to overload risk. The results showed that global metrics provided an over- all characterization of pressure distribution, while the ROI analysis was significantly more discriminative. Notably, the posture with arms behind the head exhibited the highest pressure concentration in the torso area (21.06 mmHg) despite a large global contact area. Sensitivity analysis (3, 5, 10 mmHg), conducted on the full–body configuration, showed that increasing the threshold reduced noise but also led to the loss of a rel- evant portion of the pressure distribution. These results suggested that regional pressure monitoring was crucial for effective pressure injury pre- vention and that approaches of this type could support existing clinical guidelines for patient repositioning.

ABSTRACT. Effective calibration of medical equipment is critical for patient safety and clinical accuracy. However, hospitals in low-resource settings often face challenges such as limited technical personnel, inconsistent calibration schedules, and inadequate documentation systems, which compromise equipment performance and regulatory compliance. This paper proposes a digital framework for full-cycle medical equipment calibration management, encompassing planning, execution, verification, and reporting. The framework integrates digital tracking tools, automated alerts, and dashboards to monitor calibration status, prioritize high-risk devices, and ensure adherence to international standards. By leveraging digital systems, the framework enhances operational efficiency, reduces human error, and supports sustainable resource utilization. Its design considers scalability, accessibility, and adaptability to resource-constrained healthcare environments, promoting equitable and reliable management of biomedical assets. The proposed solution intersects clinical engineering, digital health technologies, and regulatory compliance, addressing Topics 2, 3, and 6 of the IFMBE conference. The framework is presented conceptually with illustrative workflows, demonstrating its potential to improve equipment reliability, patient safety, and compliance monitoring in diverse healthcare settings.

ABSTRACT. Ambient Assisted Living (AAL) technologies can support independent living through unobtrusive monitoring of daily behavior, yet real-world deployment remains constrained by limited robustness across heterogeneous home environments and the lack of transparent, caregiver-interpretable decision support. This study aims to develop an explainable hybrid symbolic-learning framework in which a symbolic module provides interpretable behavioral states, while a deep-learning-based one characterizes resident behavior through sleep-activity analysis, used here as the prototypical behavioral dimension to capture long-range temporal dependencies. The system was developed across eight independent real-world residential environments involving older adults, each spanning one-to-two months of continuous operation. Integrating symbolic behavioral representations with deep learning consistently outperforms raw-sensor learning across all evaluated model families. The Transformer-based configuration achieves the strongest performance on held-out test scenarios, reaching F1-score of 0.96, MCC of 0.94, and PR–AUC of 0.99 —an approximate 10% relative improvement over its raw-sensor counterpart. By reframing sleep monitoring from a dichotomous rule-based decision to a probabilistic estimation problem, the proposed approach reduces the false-negative rate for sleep detection from 9.2% to 3.9%, enabling more reliable real-time monitoring. The proposed system supports real-time, edge-compatible deployment while preserving traceability through explanation-aware alarms, providing a practical and auditable solution for scalable AAL deployments.

ABSTRACT. Diabetic neuropathy (DN) is one of the most prevalent and disabling complications of diabetes mellitus, affecting up to half of patients over their lifetime and contributing substantially to ulcers, amputations, and mortality. Despite this burden, existing DN screening methods remain resource‑intensive, often invasive, and poorly suited for large‑scale or community‑based deployment. Recent advances in digital health and artificial intelligence (AI) offer opportunities to improve early detection, monitoring, and accessibility through mobile and point‑of‑care (POC) solutions. This systematic literature review aims to identify and evaluate DN screening methods that have been integrated with digital health technologies, assess their diagnostic performance, usability, and feasibility for non‑specialist use, and identify gaps to inform the design of an AI‑enabled mHealth screening toolbox. A structured search was conducted across four databases, namely PubMed, Scopus, Web of Science, and Embase, covering publications from 2011 to the present. Eligible studies involved digital or mobile health solutions applied to DN screening, diagnosis, or monitoring, including smartphone applications, wearable or POC devices, and AI‑assisted systems. Screening methods of interest included vibration, thermal, pressure, sudomotor, pupillometry, and validated questionnaires. Invasive methods and non‑digital tools were excluded. Following duplicate removal, 1,920 records were identified. After title and abstract screening, 78 records were retained, and approximately 50 studies met the inclusion criteria after full‑text assessment. Included studies demonstrated substantial heterogeneity in screening modalities, digital architectures, and validation strategies. Multimodal approaches combining subjective questionnaires with objective sensory testing were common, with several studies reporting integration into smartphone‑based platforms. Diagnostic performance metrics were variably reported, and only a subset of studies included external validation, usability assessments, or non‑specialist deployment. AI and machine learning techniques were increasingly used for classification and risk stratification, though transparency, explainability, and regulatory considerations were inconsistently addressed. Digital health solutions for DN screening show promising diagnostic accuracy and potential for scalable, non‑invasive deployment; however, evidence gaps remain in clinical validation, usability, and real‑world implementation. These findings highlight key design requirements and research priorities for developing frugal, user‑centred, and regulatorily compliant mHealth tools for DN screening and monitoring.

ABSTRACT. Purpose: Relative power spectral density (rPSD) and functional connectivity (FC) are widely used quantitative EEG (qEEG) features in computational neuroscience and epilepsy research, yet they are often analyzed independently. How their relationship varies across frequency bands and between physiological and pathological conditions remains unclear. We investigated how the rPSD–FC correlation varies across frequency bands in healthy controls (H) and people with focal epilepsy, with a specific focus on differences between responders (R) and non-responders (non-R). We also evaluated whether correlations between channel-wise rPSD and nodal graph metrics, including centrality and segregation measures, could improve discrimination between H, R, and non-R. Methods: We retrospectively analyzed scalp 19-channels EEGs from 284 people with focal epilepsy with at least 2 years of follow-up and 109 healthy controls. For each frequency band, we computed rPSD and estimated FC using multiple phase- and amplitude-based connectivity measures. From these, we derived global and nodal graph-theoretical metrics and inter-hemispheric asymmetry indices. The rPSD–FC correlation within each frequency band was quantified using Spearman’s rank correlation coefficient. Results: In H, only 23/918 global FC features were significantly correlated with rPSD (ρ>0.4, p<0.05), exclusively in the α frequency band. In contrast, people with epilepsy showed 15 correlated features overall, mainly distributed in the δ band (14 features), with only one in the α band. These moderate correlations predominantly involved FC matrices not corrected for volume conduction, i.e. phase locking value (PLV), amplitude envelope correlation (AEC), and magnitude-squared coherence. At the nodal level, α-band correlations between rPSD and closeness centrality progressively decreased from H to R and non-R for both wPLI (median values H: 0.028; R: –0.035; non-R: –0.11) and ImCOH (H: 0.09; R: 0.01; non-R: –0.08), with significant differences between H and non-R, and between R and non-R (p<0.05). In contrast, correlations between α-rPSD and clustering coefficient increased from H to epilepsy groups for both wPLI (H: 0.11; R: 0.22; non-R: 0.28) and ImCOH (H: –0.02; R: 0.10; non-R: 0.18). Clustering coefficient showed significant differences between H and both epilepsy groups, and across all ImCOH group comparisons (p<0.05). Conclusions: rPSD and FC capture complementary rather than interchangeable qEEG information. The structured α-band spectral–network coupling observed in H is disrupted in focal epilepsy, indicating a pathological reorganization of brain network dynamics. In particular, nodal correlations between spectral and topological features differentiated H, R, and non-R groups, suggesting their potential utility as multimodal EEG biomarkers for drug-response characterization and classification pipelines.

ABSTRACT. Ultrasound tomography (UST) is a promising imaging modality for biomedical and non-destructive testing applications, as it enables the reconstruction of spatially distributed mechanical properties in an operator-independent manner. Despite significant methodological advances, the validation and comparison of tomographic and data-driven imaging approaches remain hindered by the limited availability of publicly accessible experimental datasets acquired under controlled conditions. This work addresses this gap by presenting a two-dimensional experimental ultrasound tomography database acquired using an in-house multiple-input–multiple-output (MIMO) circular array system. The dataset comprises a wide range of acquisition scenarios involving targets of different sizes, shapes, materials, and spatial configurations. By providing coherent multistatic measurements together with a well-documented acquisition protocol, the proposed database is intended to serve as a benchmark platform for the development, testing, and validation of conventional, learning-based, and ultrasound imaging algorithms.

ABSTRACT. Multiple Sclerosis (MS) significantly impairs sensory integration, leading to balance disturbances that often go undetected by conventional clinical tests due to limited sensitivity. This study introduces a Mixed Reality (MR) framework designed to objectively assess postural instability through an immersive overground walking task. Developed in Unity for the HoloLens 2, the protocol utilizes four virtual labyrinths with increasing sensory and cognitive loads, including dynamic obstacles and visual-vestibular conflicts. A key contribution of this work is the Lateral Variability Index (LVI), a metric specifically defined to capture rapid micro-corrections and episodic instability that traditional indices, like sway%, often fail to detect. Preliminary validation on healthy volunteers demonstrated that the protocol effectively elicits scenario-dependent gait behaviors, with LVI proving highly responsive to localized challenges. These findings suggest that LVI could serve as a sensitive digital biomarker for the early identification of subtle MS-related balance impairments. Future developments will focus on clinical validation and leveraging high-fidelity devices like the Apple Vision Pro to further enhance diagnostic reliability.

ABSTRACT. Evaluation and validation of hepatic vascular segmentations remain an open methodological challenge: traditional volumetric overlap metrics fail to cap-ture the geometric fidelity and topological integrity of highly branched, clini-cally critical structures. This limitation undermines comparability across segmentation methods and hinders the assessment of their actual clinical suitability. This paper proposes a multi-domain evaluation framework de-signed as a reusable comparative protocol for hepatic vascular segmentation: rather than evaluating a single algorithm in isolation, it provides a structured methodology for benchmarking multiple segmentation methods against a common Ground Truth, producing method-specific diagnostic profiles across three complementary and non-redundant evaluation dimensions, volumetric (DSC, IoU, MCC), geometric (MSD, RMS, HD95), and topological (clDice). The framework is independent of the segmentation technique used and in-cludes a rank aggregation system for comparing heterogeneous methods. To demonstrate its discriminative capability, the framework was applied to five multi-scale vesselness filters and an nnU-Net v2 model on the 3D-IRCADb-01 dataset. Results show that methods with comparable volumetric perfor-mance can exhibit significantly different behaviour in terms of geometric ac-curacy and topological continuity, revealing differences invisible to DSC alone. The proposed framework constitutes a generalisable methodology for vascular segmentation evaluation and provides a solid foundation for the de-velopment of advanced approaches oriented towards clinical application.

ABSTRACT. Standardised certification and preventive maintenance of Biosafety Cabinets (BSCs) require precise airflow measurements. Manually placing an anemometer leads to errors due to operator influence, angle misalignment, and airflow disturbances. To eliminate these confounding variables, this study developed and evaluated mechanical alignment fixtures designed to enforce fixed spatial positioning and stable angular orientation during testing. Two prototypes were fabricated: a Polyamide 12 (Nylon PA12) model via Selective Laser Sintering (SLS) and a Grade 316L stainless steel model via Selective Laser Melting (SLM). Both designs were experimentally validated within a Class II Type B2 BSC across a standardised measurement grid. The results demonstrate that both fixtures significantly enhanced measurement consistency and accelerated sensor stabilisation compared to manual deployment, with the stainless steel design exhibiting optimal structural rigidity. Furthermore, while the 3D-printed nylon prototype provides an effective platform for rapid prototyping, its inherent micro-porosity subjects it to long-term chemical degradation, swelling, and dimensional warping when exposed to aggressive clinical sterilising agents. Conversely, Grade 316L stainless steel demonstrated absolute chemical immunity, making it the superior material for clinical field deployment. Although current prototypes are limited by model-specific geometries, the proposed way forward involves developing an adjustable telescopic framework with indexed locking tracks to achieve universal adaptability across diverse cabinet profiles.

ABSTRACT. Dynamic cell culture systems represent promising tools for the in vitro study of osteogenesis, as they overcome the limitations of traditional static cultures. Bone cells respond not only to biochemical signals but also to physical and mechanical stimuli, including medium flow, shear stress, and the convective transport of nutrients and metabolites. These factors activate mechanotransduction pathways regulating cell proliferation, differentiation, and mineralized extracellular matrix deposition. In this context, dynamic bioreactors recreate a more physiologically relevant microenvironment, promoting more homogeneous and functionally mature cell cultures. Among the available configurations, perfusion bioreactors are particularly effective for three-dimensional (3D) cultures on porous scaffolds. Controlled medium flow improves oxygen and nutrient distribution while generating mechanical cues associated with osteogenic differentiation. However, most current dynamic culture systems still rely on end-point analyses or discrete sampling of the culture medium to evaluate cell activity. This approach provides limited information and does not capture the dynamic evolution of cellular metabolism. To address these limitations, this project focuses on the development of a perfusion bioreactor integrated with electrochemical biosensors capable of continuously monitoring key metabolites involved in cell proliferation and osteogenic differentiation. The bioreactor consists of two main functional units: a perfusion unit responsible for generating controlled medium flow and a control unit regulating the operating conditions of the system. The platform also includes modular multiwell systems in which cells are cultured on biomimetic scaffolds under controlled flow conditions, enabling reproducible modulation of fluid dynamic parameters and the mechanical microenvironment. This approach combines dynamic 3D culture with in-line monitoring of the culture medium, allowing real-time tracking of cellular metabolic evolution and providing sensitive information on biological responses. The proposed platform may support advanced in vitro models for osteogenesis studies and biomaterial screening, contributing to the development of physiologically relevant tissue engineering systems.

ABSTRACT. Recent advances in Regenerative Medicine (RM) are enabling increasingly high degrees of biomimicry. These kinds of innovations must be supported by adequate structures in terms of biomimetic design and responses. Limitations in Additive Manufacturing (AM) technologies are evident when it comes to the fabrication of complex architectures (i.e. anisotropic structures and specific fibres disposition). We have developed many solutions to overcome these limitations and provide useful and powerful tools for researchers in the field of RM and biofabrication. We tackled the issue of complex pores designs availability, and printing path optimization by developing GIPPO (Graph-based Iterative Printing Path Optimization). It is an open-source and user-friendly platform for designing complex porous scaffolds while improving 3D printing outcomes. GIPPO can generate ad hoc bio-inspired designs and optimize the printing path, thus boosting the final performances of the devices. This resulted in enhanced printing resolution and mechanical performances which were tested through tensile and out-of-plane deformation assessments. We demonstrated that lattices performances can be enhanced by overcoming software-inherent limitations. To unlock the full potential of GIPPO as non-planar lattices generator and optimizer, we combined it with RAVEN (Robot-Assisted Volumetric ExtrusioN), a hardware-agnostic platform for robotic printing in small-scale RM applications. RAVEN is based on ROS2 (Robot Operating System) and enhanced by AI for real-time control and adjustment of extrusion flow and positioning. Our open-source platform consists of a neural network-based computer vision algorithm which identifies printed filaments, and a high precision inertial measurement unit that can accurately measure the 3D orientation of the extruder. This control on the non-planar printing process allowed us to fabricate complex “in air” and unit-cell based designs, by combining custom-made algorithms for generative 3D lattices design and automatic orientation calculation. The integration with GIPPO was successfully tested by optimising complex 3D lattices and comparing them with non-optimized ones.

ABSTRACT. The rapid integration of artificial intelligence (AI) and digital technologies into clinical environments has prompted a fundamental reconceptualization of the physician–patient–technology triad. Rather than operating as autonomous decision-makers, modern digital solutions must be designed as human-centered tools that function within a shared care ecosystem, prioritizing explainability and the humanization of healthcare. This presentation introduces the framework of Augmented Clinical Intelligence, a paradigm in which AI serves to support, amplify, and refine, rather than disrupt, traditional clinical practices, actively strengthening the foundational bond of trust and comfort between doctors and patients. To demonstrate the efficacy of this collaborative approach, we highlight two long-standing, model-driven clinical applications developed to seamlessly integrate into established medical workflows. First, in obstetrics, an interactive multinormal probability model paired with artificial neural networks allows real-time checking and correction of human errors during fetal echobiometry, significantly improving the accuracy of fetal weight estimations. Second, in cardiac surgery, a machine-learning clinical model optimizes preoperative blood resource allocation, radically reducing blood bag waste while safeguarding patient safety. Both examples demonstrate that state-of-the-art AI systems achieve their full clinical potential when they are transparent, interpretable, and subordinate to medical judgment. By acting as a robust "second opinion," these technologies reduce cognitive burdens and minimize clinical errors without altering the relational integrity of medical care. Ultimately, this work posits that the future of medical engineering lies not in the replacement of human expertise, but in its deliberate, harmonious, and ethically grounded technological augmentation.

ABSTRACT. Medical imaging has been fundamentally transformed by the advent of deep learning architectures, particularly convolutional neural networks (CNNs) and, more recently, vision transformers (ViTs). This presentation provides a comprehensive overview of the trajectory from algorithmic detection to full-scale clinical integration, examining both the technical underpinnings and the translational challenges inherent in deploying image-based AI in real-world healthcare settings.

We review state-of-the-art performance benchmarks across key imaging modalities, including radiology, pathology, ophthalmology, and cardiology where deep learning models have demonstrated diagnostic accuracy comparable to or exceeding specialist-level performance. Particular attention is devoted to the methodological rigor required for clinical validation: prospective study design, external cohort testing, and the construction of demographically representative training datasets.

A central focus of this session is the growing body of real-world deployments in which AI imaging tools have achieved full clinical integration. In chest radiology, AI-assisted triage systems are now operational in multiple European hospital networks, automatically flagging critical findings such as pneumothorax and pulmonary embolism to reduce reporting delays. In ophthalmology, deep learning platforms for diabetic retinopathy screening have received regulatory clearance and are deployed at population scale in national programmes across several State. Pathology has seen the emergence of AI-powered whole-slide image analysis for cancer grading, with several tools embedded directly into pathology workstation software. These examples illustrate how imaging AI, when designed around established clinical workflows and validated on representative populations, transitions successfully from research prototype to standard of care. Case studies from European hospital networks are used to illustrate successful integration pipelines, and a phased implementation model is proposed in which deep learning tools are introduced as augmentative instruments within structured quality-assurance frameworks.

ABSTRACT. Dermatology represents one of the most advanced clinical domains for the deployment of AI-based diagnostic algorithms, owing to the inherently visual and pattern-dependent nature of skin disease classification. This presentation systematically reviews the current landscape of machine learning tools applied to dermatological diagnosis, with a focus on their clinical utility, performance characteristics, and translational potential. The recognition of cutaneous malignant melanoma at the early stage rests a critical issue for dermatologists worldwide. Early diagnosis based on dermoscopic examination rest the first tool to decrease MM mortality rate. However, this non-invasive diagnostic method is largely influenced by the dermatologists’ experience and by the subjectivity in visual perception of dermoscopic features inside melanocytic skin lesions. Moreover, standard dermoscopy fails to reach adequate diagnostic accuracy in differentiating the early melanomas from the atypical nevi with overlapping clinical and biological features. More recently, the medical imaging field drew its attention to convolutional neural network, deep learning models that remind the neurons connections of the human visual cortex. In the last few years, DCNNs became of interest as decision support systems for dermoscopic analysis of MM and skin lesions in general, and different CNNs models competed in international challenges on large datasets. Promising results have been reported overall, but adequate comparison was hard due differences in the composition of the training and of the testing set, beside the intrinsic differences in the DCNN architecture. We here aim to summarize the status of the art of DCNN in melanoma diagnosis and present the newly proposed hybrid DCNN models integrated with clinical data. Based on the current encouraging results, future architectures should consider to incorporate some additional clinical data of lesions and patients to improve their diagnostic accuracy in difficult MSL detection. Moreover, we report on our recent studies concerning the use of DL models in the non-invasive diagnosis of squamous cell carcinoma and basal cell carcinoma with line-filed confocal optical coherence tomography, a new imaging device of dermatologic use able to give an in vivo tridimensional overview of the skin at cellular resolution. Finally, we illustrate a series of experiments on the use of GAN-generative adversarial network in the generation of synthetic images of melanomas and atypical nevi for educational purposes of dermatologists and implementation of defective datasets useful for the training of future diagnostic models

ABSTRACT. The talk presents a multimodal generative AI pipeline for structured dermoscopic reporting, combining CNN-based classification of melanocytic lesions with LLM-driven clinical report generation. The framework builds on a prior validation study of synthetic dermoscopic images, assessed through a visual Turing test with expert dermatologists, which confirmed their perceptual quality and value for model training. Implications for reliability, reproducibility, and governance of generative models as decision support systems in clinical practice will be discussed.

ABSTRACT. As artificial intelligence systems become increasingly embedded in clinical practice, a rigorous examination of their limitations is both scientifically necessary and ethically imperative. This presentation offers a critical appraisal of the structural, epistemic, and normative constraints that define the boundaries of AI applicability in medicine, with particular emphasis on algorithmic bias, patient safety, and the distribution of moral responsibility.

We begin with a taxonomic analysis of bias sources across the AI pipeline — from data curation and annotation to model training and deployment — demonstrating how systematic underrepresentation of minority populations, atypical presentations, and low-prevalence conditions generates discriminatory error gradients that may exacerbate existing healthcare inequalities. Documented case studies across imaging, risk stratification, and clinical triage are presented to ground theoretical concerns in empirical evidence.

Safety considerations are addressed through the lens of failure mode analysis: we characterize the conditions under which AI systems produce high-confidence erroneous outputs, and argue for mandatory adversarial testing as a prerequisite for clinical certification. The ethical framework is developed around the concepts of non-maleficence, accountability, and informed consent in AI-mediated care. We critically interrogate the diffusion of clinical responsibility in human–AI hybrid decision systems and propose a structured model of shared accountability. The session concludes with a set of evidence-based recommendations for governance bodies, clinical institutions, and AI developers committed to the safe and equitable integration of AI into medicine.

ABSTRACT. The rapid evolution of Digital Health is transforming healthcare delivery through Software as a Medical Device (SaMD), artificial intelligence (AI), connected medical devices, interoperability, cybersecurity, remote patient monitoring, and data-driven clinical decision support systems. As healthcare organizations increasingly rely on digital technologies, Clinical Engineers are expected to play a critical role in technology assessment, implementation, integration, lifecycle management, risk management, and governance of these systems. However, the extent to which Digital Health competencies are formally represented within the Body of Knowledge (BOK) used for Clinical Engineering certification in Canada and the United States remains uncertain.

This presentation examines the current Clinical Engineering certification frameworks administered by the American College of Clinical Engineering (ACCE) to assess their coverage of Digital Health competencies. Topics including SaMD, AI governance, cybersecurity, interoperability standards, health informatics, clinical data management, digital health regulations, and implementation science will be reviewed against emerging industry needs and evolving regulatory expectations. Findings from recent ACCE BOK surveys and Canadian Digital Health initiatives will be discussed to identify strengths, gaps, and opportunities for modernization.

The session aims to stimulate international discussion on the future competencies required of Clinical Engineers and provide recommendations for evolving certification frameworks to prepare the profession for leadership in the safe, effective, and responsible adoption of Digital Health technologies.

ABSTRACT. Software as a Medical Device (SaMD) is rapidly transforming healthcare delivery, with increasing numbers of software-based technologies being licensed by Health Canada and the U.S. Food and Drug Administration (FDA) for diagnosis, treatment, and clinical decision support. While manufacturers have adapted to evolving regulatory frameworks and quality management requirements, healthcare delivery organizations are only beginning to establish the operational processes needed to manage these technologies throughout their lifecycle. Traditional clinical engineering and healthcare technology management programs were developed around physical medical devices and often lack established approaches for software-centric technologies. Challenges related to ownership, inventory management, cybersecurity oversight, software updates, risk management, maintenance strategies, and integration with computerized maintenance management systems (CMMS) continue to emerge as SaMD adoption accelerates. This presentation examines the evolving role of clinical engineering in supporting SaMD deployment and lifecycle management within healthcare organizations. Drawing on implementation experience and practical examples, the session discusses governance considerations, operational workflows, and lessons learned in adapting existing healthcare technology management practices to software-based medical technologies. The discussion highlights opportunities for collaboration among clinical engineering, information technology, and clinical stakeholders as organizations work to bridge the gap between regulatory guidance and hospital practice.

ABSTRACT. Four revisions in 28 years. That is the update cadence of CESOP , Canada's foundational guidance document for clinical engineering practice. At an average of one revision every 7 to 9 years, the math is unforgiving: by the time consensus is built, drafts are circulated, and a revision is published, the technology landscape has already moved on. The 2023 revision deserves credit though, cybersecurity governance, MDS2 evaluation, SaMD lifecycle management, and network security all received meaningful attention. But artificial intelligence, which began reshaping clinical environments in late 2022, is entirely absent. With no announced revision schedule, the next update could arrive well into the 2030s. This session asks an uncomfortable question: can a well-intentioned, consensus-driven committee process ever keep pace with exponential digital health innovation…and if not, what do we do about it?

ABSTRACT. Critical medical devices in modern hospitals – infusion pumps, defibrillators, patient monitors, ventilators, ECG carts, mobile beds – move continuously between wards, leading to misplacement, accelerated wear, delays in patient care, and a substantial hidden workload for nursing and biomedical engineering staff. This paper presents a pilot subsystem for real-time indoor positioning of critical medical devices, developed and deployed under Greece’s national project Horizontal Digital Transformation Interventions for Supervised Bodies of the Greek Ministry of Health – Section 3: Unified Medical Equipment Management System. The pilot is implemented by INBIT as a specialised subcontractor and integrates with the National Medical Equipment Inventory and Vigilance platform, which covers 129 public hospitals and an approximately 170,000 medical assets. The proposed architecture combines two complementary positioning tiers within a single platform: a high-precision tier based on Bluetooth Low-Energy (BLE) Angle of Arrival (AoA) for clinically critical areas, with typical accuracy of 0.1–0.5 m, and a lower-cost tier based on BLE beacons for general areas. Switching between tiers is transparent to operators. The system displays the live position of tagged devices on hospital floor plans, records movement history for analytics, and triggers real-time alerts based on configurable geofences. Beyond device tracking, the same infrastructure is designed to support broader smart-hospital services, including patient and personnel guidance, accessibility services for visually impaired visitors – building on the previously published HOSP4ALL pilot at the University General Hospital of Patras – and operational analytics. We describe the system architecture, the deployment approach, expected operational benefits, and open challenges for hospital-wide and nation-wide adoption.

ABSTRACT. Access to advanced diagnostic imaging remains a challenge in many low- and middle-income countries due to high acquisition and maintenance costs. In Ghana, a national effort to standardize Magnetic Resonance Imaging (MRI) systems across teaching hospitals was initially successful, but maintenance issues later rendered many machines non-functional. This paper presents a successful case of institutional networking among teaching hospitals to restore MRI functionality at the Komfo Anokye Teaching Hospital (KATH). Through a collaborative agreement facilitated by the Ministry of Health, essential spare parts were sourced from other institutions with decommissioned MRI units. This initiative restored diagnostic capacity at KATH, improved access to imaging services, and highlighted the strategic value of coordinated institutional partnerships and resource pooling.

ABSTRACT. Oculopalpebral defects caused by congenital conditions, oncologic disease, or severe trauma led to major esthetic, functional, and psychosocial impairment. Although biomaterials, imaging, and computer-aided design tools have progressed substantially, clinical fabrication of facial prostheses still relies heavily on conventional artisan workflows that demand repeated patient visits and prolonged production times. This study describes and preliminarily validates a standardized digital workflow for patient-specific oculopalpebral prostheses based on computed tomography, 3D reconstruction, CAD modeling, and additive manufacturing. A cross-sectional clinical case with technological intervention was conducted in a 62-year-old woman with ocular loss and facial defect secondary to mucormycosis. Anatomical data were acquired by computed tomography and 3D facial scanning, segmented in 3D Slicer, and modeled in Blender. A physical facial model and anatomical negative were fabricated to decouple fitting from patient presence. Prototypes were produced by 3D printing and cast with a biocompatible silicone resin. The proposed workflow generated high-fidelity anatomical models with deviations below ±0.25 mm, reduced the need for repeated in-person fitting sessions, and shortened total manufacturing time to approximately 90 days. In vitro testing in NIH-3T3 fibroblasts showed cell viability above 80%, consistent with non-cytotoxic behavior under ISO 10993-5 criteria. These findings support the clinical feasibility of a transferable digital workflow for facial rehabilitation that improves reproducibility and patient-centered care.

ABSTRACT. Segmental mandibular resection following oral squamous cell carcinoma can produce complex bone defects that compromise mastication, swallowing, speech, and facial symmetry. Accurate reconstruction planning is therefore essential to support anatomical restoration and improve surgical decision-making. This study presents the design of a patient-specific mandibular graft for reconstruction planning using medical image processing. Computed tomography (CT) images in DICOM format from a patient with a segmental mandibular defect were processed to generate a three-dimensional model of the mandible. Bone tissue segmentation was performed in 3D Slicer, followed by mesh optimization, defect delimitation, and graft modeling in Blender. The personalized graft was designed using the contralateral healthy mandibular segment as an anatomical reference through mirror-based reconstruction and was adapted to the resection margins to achieve virtual restoration of mandibular continuity. The proposed design enabled detailed visualization of the defect and generation of a geometrically consistent patient-specific graft model for reconstruction planning. These findings support the feasibility of medical image processing as a practical approach for personalized mandibular graft design. Further work should incorporate quantitative fit assessment, finite element analysis, additive manufacturing, and mechanical or biological validation.

ABSTRACT. Health care systems are increasingly expected to make sure that decisions on the adoption, purchase, and disinvestment of technologies are supported by robust, multidimensional evidence. Biomedical engineering departments can provide detailed operational data on technologies, which are directly pertinent to such decisions. However, this type of information is seldom used in the formulation of public health policies. Health Technology Assessment (HTA) is recognized as a structured, multidisciplinary process for assessing different implications of health technologies, although the connection with biomedical engineering is limited. This paper presents a study on HTA as a translational architecture that can bridge biomedical engineering and public health policy. A structured approach was used to undertake a systematic domain analysis, and a comparative structural analysis that focused on differences in the unit of analysis, time horizon, evaluation logic, and accountability. Then, a translational integration approach comprising three layers, with HTA as a formal intermediary, was proposed. The findings reaffirm the premise that the disconnect between engineering and health policy is organizational, where the metrics of the former are inherently compatible with the existing evaluation domains of HTA, but the pathways to integration are lacking. The approach illustrates the potential role of the clinical engineer as a ‘translational agent,’ providing operational evidence to support procurement, capital investment, and disinvestment decisions through the HTA process. The improvement in integration between biomedical engineering and HTA has significant implications for transparency, sustainability, and equity in the governance of health technology, and indicates the need for curricular reform in biomedical engineering.

ABSTRACT. Engineering hydrogels with independently tunable network architectures is critical for understanding and directing cell behavior in 3D microenvironments. In this study, we develop a dual-crosslinking strategy using thiolated pectin and pectin methacrylate to fabricate hydrogels with controllable physicochemical and biological properties. The system leverages thiol–ene Michael addition for rapid, spontaneous network formation under mild conditions, followed by photocrosslinking of methacrylate groups to further stabilize and modulate the network. By varying the ratio of thiol to methacrylate functionalities and the extent of photoexposure, we systematically tailor gelation kinetics, crosslinking density, mechanical stiffness, and degradation behavior. This modular platform enables decoupling of initial cell encapsulation conditions from subsequent network reinforcement. Encapsulated cells are evaluated for viability, proliferation, morphology, and mechanosensitive responses to elucidate how sequential crosslinking influences cell–matrix interactions. Our results demonstrate that the interplay between thiol–ene and photoinitiated crosslinking governs matrix elasticity and ligand presentation, which in turn regulate cell spreading and function. This study provides a versatile framework for designing pectin-based bioadhesive hydrogels with spatiotemporal control over material properties, with potential applications in tissue engineering, wound healing, and injectable therapeutics.

ABSTRACT. This study evaluated the therapeutic efficacy of hyaluronic acid (HA) and folic acid (FA)-doped magnetofluorescent carbon dots (MFCDs) against breast cancer in a murine model. Red-emissive MFCDs doped with FA and HA were synthesized via a microwave-assisted method, designated as FMCD and HMCD, respectively. In vitro cell viability assays using 4T1 murine breast cancer cells indicated that FMCD and HMCD exhibited negligible cytotoxicity. Under radiofrequency (RF)-induced hyperthermia, FMCD and HMCD suspensions achieved a temperature increase of 16.7 °C, resulting in a 4T1 cell viability reduction of 87% and 75%, respectively. To study the cellular heat-shock response, Western blot analysis of exosomal markers showed upregulated expression of CD63, GAPDH, and HSP70 following hyperthermia treatment. In vivo studies demonstrated that mice receiving four cycles of magnetic hyperthermia following tail vein injection showed the most significant tumor growth inhibition, whereas RF treatment alone failed to suppress progression. These multifunctional nanoprobes show great potential for targeted dual-modal imaging and minimally invasive hyperthermia therapy.

ABSTRACT. Trabecular bone regeneration requires a scaffold architecture that combines pore interconnectivity, structural continuity, and direction-dependent mechanical behavior. In this work, a porous scaffold for trabecular bone regeneration was developed through a workflow that integrated mathematical design, mechanical homogenization, and resin-based additive manufacturing. A custom unit cell was generated in nTop 5.0 from an implicit trigonometric formulation and periodically replicated to construct the complete scaffold. The geometry was parameterized to evaluate the effect of thickness on the apparent mechanical elastic response. Mechanical homogenization was performed on a representative unit cell using a linear elastic isotropic material model in order to estimate the effective stiffness distribution arising from the architecture. The resulting stiffness matrices showed direction-dependent behavior and progressive changes as scaffold thickness increased, suggesting that the proposed topology can be tuned toward a more uniform or more porous structural response depending on design requirements. After numerical evaluation, the full scaffold was fabricated as a single prototype by high-resolution resin-based additive manufacturing using a Q-touch HALOT-X1 16K printer and Tough 1500 Resin. The printed construct was successfully obtained and UV-cured, demonstrating the feasibility of translating the computational design into a physical porous structure. Although no geometric metrology, experimental mechanical testing, or biological validation was performed at this stage, the proposed workflow establishes a preliminary platform for future biofabrication and experimental studies aimed at trabecular bone tissue engineering.

ABSTRACT. Sepsis is a critical illness characterised by elevated mortality rates and significant financial strain on healthcare systems. Due to the problems in detection and treatment, it is considered as one of the most serious global public health challenges. Early identification of sepsis and initiation of timely and effective treatment are crucial for patient survival. This research presents a novel model named Intelligent Sepsis Predictor (iSP) for the early prediction of sepsis development in patients admitted to intensive care units. We present a machine learning ensemble using the XGBoost algorithm and k-nearest neighbours with a Dynamic Time Warping distance metric. The suggested model calculates static and time series data from vital signs and laboratory results. We trained our suggested model utilising MIMIC IV and a proprietary database from a hospital in Colombia, adhering to the sepsis-3 definition. The results of the study demonstrate that the iSP model generates more reliable predictions than conventional scores, providing an area under the characteristic receiver curve (AUROC) for MIMIC IV: (0.88, 0.89, 0.78, 0.84) and for the private database (0.88, 0.86, and 0.86) for 1 hour, 2 hours, and 3 hours prior to sepsis onset, respectively. Sepsis syndrome can be anticipated by integrating physiological and laboratory test findings inside a machine learning framework.

ABSTRACT. Biomedical Engineering has become one of the most multidisciplinary and rapidly evolving fields in higher education. In addition to its traditional foundations in engineering, biology and medicine, the field now includes artificial intelligence, digital health, medical devices, health data science, robotics, regulatory frameworks and health technology assessment. These developments create new opportunities, but also place increasing pressure on curricula, as emerging subjects compete for space within already crowded programmes. Since the early 1990s, several initiatives have attempted to harmonise Biomedical Engineering education, including IFMBE-related activities, BIOMEDEA and European projects such as TEMPERE, CRH-BME and BME-ENA, developed under ERASMUS, TEMPUS and related EU programmes. These efforts although aligned with the Bologna Declaration aiming to facilitate student and staff mobility, ECTS and mutual recognition of degrees, necessitated reconsideration every ten years. Today, the situation is more complex. A recent survey identified more than 550 active Biomedical Engineering programmes, representing an approximate 60% increase compared with the 2020 baseline. Bachelor-level Biomedical Engineering programmes have become increasingly prominent, while the field itself is becoming more specialised and diverse. This growth intensifies longstanding challenges related to curriculum comparability, student mobility, ECTS recognition and professional competence alignment. This mini-symposium will take the form of a round table discussion and will address the central question: Is there still room for harmonisation of Biomedical Engineering education, and if so, how should it be achieved? The session will explore whether harmonisation should continue to focus on curricula, or whether a more flexible model is needed, based on common core competences, learning outcomes, specialisation tracks, modular mobility and micro-credentials. Emphasis will be placed on open discussion with the audience, inviting proposals and positions on the advantages and disadvantages of renewed harmonisation efforts. The expected outcome is to initiate a broader dialogue on a possible Biomedical Engineering education framework for the coming decade, capable of supporting quality, mobility and professional identity while preserving curricular flexibility and innovation.

ABSTRACT. The origins of Biomedical Engineering (BME) at the University of Ljubljana, Faculty of Electrical Engineering (UL-FE), date back to the late 1960s, while BME was formally introduced into the undergraduate curriculum in 1977. Three decades later, the European Tempus project CRH-BME coincided with the establishment of the Department of Biomedical Engineering and the Bologna reform of first- and second-cycle study programs at UL-FE. Building on these developments, a Master's-level specialization in Biomedical Engineering within the Electrical Engineering program was launched in the 2012–2013 academic year. The curriculum was rooted in earlier educational activities while also incorporating recommendations developed within the CRH-BME project, including the balance of credit allocation across course categories and the inclusion of courses considered at the time part of the essential BME core. The project thus provided an important framework for harmonizing the curriculum with contemporary European BME educational guidelines while preserving local research strengths and expertise. Since its inception, the curriculum has remained a living framework that continuously integrates developments emerging from BME research groups at UL-FE. Over the past fourteen years, the BME specialization has established itself as a stable component of the Master's program in Electrical Engineering and is currently one of seven available study tracks. Apart from a small number of shared electives, it maintains a dedicated curriculum and course portfolio. While most students enter from the first-cycle Electrical Engineering program at UL-FE, the specialization also attracts graduates from Physics, Mechanical Engineering, Chemical Engineering, and the Biotechnical Faculty. Although the overall course structure has remained largely unchanged, course contents are continuously updated to reflect advances in biomedical technologies, data science, and artificial intelligence. AI is increasingly shaping both biomedical engineering practice and engineering education, creating new opportunities in areas such as signal and image processing, medical decision support, and personalized healthcare. At the same time, it raises important questions regarding curriculum design, teaching methods, assessment, and the competencies expected of future biomedical engineers. As with other engineering programs at UL-FE, our specialization is adapting to these opportunities and challenges while maintaining the interdisciplinary foundations of the field.

ABSTRACT. Artificial Intelligence is rapidly transforming healthcare, medical technology, and biomedical research. While many educational initiatives focus on teaching algorithms, programming techniques, and data science methods, the successful adoption of AI in healthcare depends on a broader set of competences. In practice, biomedical engineers rarely act only as AI developers. More frequently, they serve as leading intermediaries between technology experts, clinicians, hospital managers, regulatory bodies, and industrial stakeholders.

Biomedical Engineering education should move beyond a purely technical view of AI and focus on preparing graduates and PhD to become effective facilitators of impactful AI innovation and adoption. Future biomedical engineers need sufficient technical literacy to understand AI capabilities and limitations, but they also require competences in clinical workflow analysis, health technology assessment, regulatory science, risk management, ethics, data governance, sustainability, change management, and communication with both top management and middle management.

A competitive competence-based educational framework structured should be based around five pillars: (1) AI literacy and technical understanding; (2) assessment of clinical, organizational, economic, and societal impact; (3) implementation and adoption within healthcare organizations; (4) leadership and stakeholder engagement; (5) ethics, regulatory science and business models. Such a framework would enable biomedical engineers to identify opportunities for AI, critically assess solutions, support procurement and implementation decisions, and foster responsible innovation.

The presentation will discuss whether these competences could constitute part of a common (trans) European core curriculum for Biomedical Engineering in the AI era. Harmonisation efforts should focus less on specific technologies, which evolve rapidly, and more on durable competences that allow graduates to continuously adapt to future innovations while maintaining a central role in healthcare transformation.

Biomedical Engineering educators should therefore move beyond content-centred curricula and embrace a new educational paradigm. Our primary mission should no longer be to transfer a finite set of technical notions, but to cultivate professionals capable of continuously learning, unlearning, and relearning throughout their lives.

The ultimate goal is not to produce graduates who master today's AI tools, but graduates who can confidently adapt to tomorrow's technologies—many of which do not yet exist. This requires fostering intellectual curiosity, critical thinking, systems thinking, ethical responsibility, and the ability to autonomously acquire new competences. Universities should aspire to develop what might be called “Olympians of Learning”: professionals trained to continuously strengthen their own cognitive capabilities, capable of navigating uncertainty and leading innovation across healthcare systems. During the relatively brief period students spend at our institutions, educators have a unique opportunity not only to teach knowledge, but also to transmit vision, values, intellectual discipline, and a lifelong commitment to learning.

As Biomedical Engineering education enters the AI era, harmonisation efforts should therefore focus not only on defining common competences, but also on reimagining the very purpose of engineering education itself.

ABSTRACT. Biomedical Engineering is an emerging field at the intersection of science and technology. The Department of Biomedical Engineering at the University of West Attica is the only higher education institution in Greece offering a comprehensive range of studies in this field. Its programs include a five-year integrated undergraduate degree (equivalent to a master’s degree), a master's-level conversion program, and doctoral studies.

Graduates of the Department are employed in hospitals and clinics, as well as in biomedical equipment companies across manufacturing, service, and sales sectors. They are also equipped to develop software solutions for medical informatics, conduct research in medical image and signal processing, and design, implement, and support computer-aided decision support systems.

ABSTRACT. The rapid evolution of healthcare technologies is reshaping the role of clinical engineers, requiring continuous adaptation to increasingly complex and interconnected systems This contribution explores the main challenges and opportunities that future clinical engineers will face, including the integration of digital health solutions, artificial intelligence, and data-driven decision-making processes. Particular attention is given to identifying the key skills needed to effectively manage innovation, such as interdisciplinary collaboration, strategic thinking, regulatory awareness, and proficiency in emerging technologies. The presentation aims to offer a practical perspective on how future clinical engineers can contribute to technological development and strengthen their role as enablers of safe, efficient and sustainable healthcare innovation.

ABSTRACT. Biomedical engineering education varies significantly across countries and institutions. Each system believes it is preparing graduates for real-world demands. While universities continue to provide strong technical foundations, questions remain about whether current educational models fully prepare graduates for the realities of modern biomedical engineering careers. This paper reflects on my experience as an Erasmus Mundus scholar in the EMMBIOME program, studying across Greece, Serbia, and Romania. Through exposure to different educational systems, research environments, healthcare settings, and cultural contexts, I observed significant differences in how biomedical engineering competencies are developed and applied. These experiences highlighted skills that are often difficult to teach through traditional coursework alone, including adaptability, cross-cultural collaboration, problem-solving in unfamiliar environments, leadership, and the ability to connect technical innovation with real clinical needs. Rather than presenting a single best model, this paper will explore how combining different educational environments can better prepare future biomedical engineers. Beyond formal coursework, it will present the importance of student-led initiatives in biomedical engineering, and the crucial lessons that emerge when building them in unfamiliar countries, cultures, and systems. This paper offers concrete recommendations for curriculum development, drawn from experience inside the system — arguing that the goal of education is not only technical preparation, but building in graduates the steadiness, care, and capacity to keep learning that a constantly changing field demands.

ABSTRACT. Recent advances in neural engineering and bionics have demonstrated the growing potential of invasive and non-invasive interfaces to restore bidirectional communication with the nervous system. This talk will explore how closed-loop human–machine interfaces can support motor control and sensory feedback restoration in individuals with limb loss. Starting from recent studies on invasive neural interfaces for tactile feedback in upper-limb prostheses, the presentation will then address non-invasive approaches based on transcutaneous electrical nerve stimulation for eliciting somatotopic sensations in both upper- and lower-limb prosthetic applications. The role of multimodal sensing, bio-inspired encoding strategies and AI-based control will be discussed as a means to improve prosthesis usability, reduce cognitive burden and promote more natural human–machine interaction. Finally, the talk will open toward future perspectives, highlighting current challenges and research directions for the development of next-generation bionic prostheses.

ABSTRACT. Postural control (PC) impairment is a key feature of Parkinson’s disease (PD), arising from altered sensorimotor integration and reduced adaptability to environmental demands. The BioVRSea paradigm enables the assessment of PC under controlled visual and mechanical perturbations through the simultaneous acquisition of center-of-pressure (CoP) and electromyographic (EMG) signals. This study compared PC responses between patients with PD and healthy controls (CTR). CoP analysis revealed consistent between-group differences across phases, with CTR showing greater sway amplitude, spatial dispersion, and directional variability, especially in the POST phase. In contrast, PD participants exhibited reduced sway excursions, suggesting a constrained and less flexible postural strategy. Notably, PRE and POST, which share similar visual conditions without platform movement, showed clear differences, with POST highlighting the persistence of altered postural responses following perturbation. EMG analysis demonstrated muscle-specific patterns. The gastrocnemius lateralis showed widespread differences across amplitude, variability, and spectral features, already evident at baseline and persisting after perturbation. The tibialis anterior exhibited marked baseline alterations with more selective phase-dependent effects, while the soleus showed limited differences, mainly related to variability. These findings indicate that PD-related postural alterations emerge both at rest and after perturbation, reflecting reduced adaptability and altered neuromuscular organization.

ABSTRACT. Human movement is a complex process involving both the central nervous system and the musculoskeletal system. Walking, in particular, represents a sophisticated motor task requiring precise coordination of multiple joints, especially in the lower limbs, along with continuous neuromuscular control. Mathematical modeling plays a key role in describing the dynamics and temporal evolution of gait patterns. In this study, we present the development of a mathematical model and a MATLAB Simulink-based framework for simulating human gait in the sagittal plane during the single-support phase. Specifically, a Proportional Derivative (PD) controller is implemented in conjunction with the compensation of the underlying nonlinear dynamics. Simulations were conducted using data from a healthy control (HC) subject and an age- and gender-matched subject affected by idiopathic Parkinson’s disease (iPD). The results were analyzed and compared with experimental gait data acquired through a motion capture system developed by BTS Bioengineering. Agreement between simulated and experimental trajectories was quantified using the Root Mean Square Error (RMSE). The proposed framework achieved low tracking errors for both HC and iPD conditions, with RMSE values < 0.006 m for hip coordinates, < 0.09° for trunk motion, and < 0.6° for the lower-limb joint angles, confirming the ability of the model to accurately reproduce subject-specific gait dynamics.

ABSTRACT. Electrical stimulation is currently one of the most established neuromodulation strategies and is widely used to activate peripheral nerves in both research and clinical settings. By delivering controlled electrical currents, this approach can reliably induce neuronal depolarization and action potential generation. However, electrical stimulation presents inherent limitations, including limited spatial selectivity, current spread to surrounding tissues, unintended activation of non-target fibers, and challenges associated with the electrode–tissue interface. These limitations have driven the exploration of alternative stimulation modalities capable of providing more localized, contact-free, and potentially cell-type-specific neuromodulation. In this context, infrared optical stimulation (INS) has emerged as a promising approach for peripheral nerve activation. By using light to induce localized photothermal and/or thermosensitive effects, optical stimulation may offer improved spatial precision and tunability compared with conventional electrical stimulation. Within the framework of the Optonerve project, this study aims to develop and validate an optical neuromodulation platform for peripheral nerve stimulation. The project encompasses complementary in vitro, ex vivo, and in vivo investigations designed to progressively evaluate the feasibility, underlying mechanisms, and translational potential of infrared optical stimulation. The preliminary results presented here focus specifically on the in vitro phase, involving dorsal root ganglion (DRG) neurons. Experiments were performed using whole-cell current-clamp recordings to directly monitor membrane potential dynamics during both electrical and optical stimulation. Optical stimulation was delivered through a multimode optical fiber with a 200 µm core at a wavelength of 1480 nm and a stimulation frequency of 5 Hz, while varying the duty cycle to modulate pulse duration and radiant exposure. This experimental protocol enabled a direct comparison between electrically evoked excitability and optically induced neuronal responses. The results demonstrate that infrared optical stimulation can evoke action potentials in a subset of DRG neurons. Electrical stimulation was employed to assess neuronal excitability and to define the depolarization range required for action potential generation. Additional control recordings performed in the absence of both optical and electrical stimulation allowed optically evoked responses to be distinguished from spontaneous firing activity and baseline fluctuations. Preliminary pharmacological experiments were also conducted using HC067047, a TRPV4 channel inhibitor, to investigate the potential contribution of thermosensitive pathways to infrared-mediated neuronal activation. Although some recordings were affected by noise, membrane potential instability, and limited cell viability, the in vitro findings support the feasibility of optical stimulation as a strategy for modulating DRG neuronal activity. Overall, these results provide an initial experimental foundation for the Optonerve platform and contribute to the optimization of optical stimulation parameters for subsequent ex vivo and in vivo validation. Future work will focus on improving reproducibility, refining stimulation thresholds, and elucidating the biophysical and molecular mechanisms underlying infrared-induced activation of peripheral sensory neurons.

ABSTRACT. Electrical stimulation has long served as the cornerstone of therapeutic neuromodulation, underpinning technologies from cochlear implants and deep brain stimulators to spinal cord and peripheral nerve interfaces. Decades of clinical deployment have yielded robust evidence for efficacy across a range of conditions, yet the field continues to grapple with fundamental limitations: poor spatial selectivity, chemical/mechanical mismatch that worsens foreign body reaction at the tissue/electrode interface, and an incomplete mechanistic understanding of how applied fields translate into durable therapeutic outcomes. In the first half of this talk, the current state of the art in electrical neuromodulation will be reviewed, with particular attention to where clinical promise has been realized and where persistent engineering and biological barriers remain. Emerging strategies to expand the functional capabilities of spinal and peripheral nerve interfaces will be presented, including data-driven methods for accelerating stimulation protocol optimization, computational model-guided approaches to surgical implantation, and the use of complex multisynaptic reflex pathways to engage residual neural circuits. The second half of the talk turns to an emerging landscape of modalities that seek to transcend these barriers and exploits alternative source of energies or stimuli-responsive nanostructured materials to interrogate the nervous system in the framework of neuromodulation or peripheral nerve regeneration. Focused ultrasound neuromodulation offers millimeter-scale targeting at depth without implanted hardware, with accumulating evidence for both excitatory and inhibitory effects whose biophysical basis (mechanosensitive ion channels, intramembrane cavitation, thermal micro-gradients) is only now becoming tractable. Chemogenetic and optogenetic approaches, while still largely preclinical, are redefining what cell-type specificity and closed-loop control can mean in practice, and early-phase human trials are beginning to test their translational limits. Light-based approaches, including infrared neuromodulation and photobiomodulation, have shown strong potential as wireless methods to interrogate the nervous system. These modalities were demonstrated to enable reversible stimulation or inhibition of action potential propagation in peripheral nerves and are emerging as promising strategies for chronic neuropathic pain management and peripheral nerve regeneration. Magnetic peripheral nerve stimulation and transcranial temporal interference stimulation round out a toolkit that increasingly allows investigators to interrogate the nervous system with spatial and temporal precision unimaginable a generation ago. Across all modalities, common themes emerge: the need for bidirectional interfaces that record and stimulate simultaneously, the imperative of closed-loop architectures driven by physiologically meaningful biomarkers, and the challenge of long-term biocompatibility in devices designed to function across a human lifetime. This talk argues that the transition from empirical parameter tuning to mechanistically informed, adaptive neuromodulation represents the defining engineering challenge of the coming decade, and that the convergence of beyond-electrical modalities with advanced neural decoding may finally make it achievable.

ABSTRACT. The increasing adoption of artificial intelligence (AI) in healthcare is creating unprecedented computational demands, particularly in applications involving large-scale medical imaging datasets. High-resolution imaging modalities, such as four-dimensional (4D) ultrasound, generate massive volumes of data that require substantial computational resources for model development, training, validation, and deployment. High-Performance Computing (HPC) has therefore become a key enabling technology for translating AI innovations into clinically applicable solutions. This presentation introduces a case study focused on AI-assisted prenatal neurodevelopmental assessment based on the Kurjak Antenatal Neurodevelopmental Test (KANET). A custom deep learning pipeline was developed for automated analysis of fetal neurobehavioral patterns extracted from 4D ultrasound recordings. Computational benchmarking was performed between a conventional local workstation environment and an on-premise HPC infrastructure. The results demonstrated significant improvements in computational efficiency, scalability, stability, and model training performance when utilizing HPC resources, highlighting their importance for processing large medical datasets and supporting advanced AI workflows. Beyond the technical case study, the presentation will showcase opportunities available through the EuroCC initiative and the National Competence Centre for HPC in Bosnia and Herzegovina. Special emphasis will be placed on the services available to researchers, healthcare institutions, startups, SMEs, and public organizations, including access to European supercomputing infrastructures, AI and data analytics support, application assistance for EuroHPC resources, training programs, proof-of-concept development, digital innovation services, and capacity-building activities. The presentation will demonstrate how HPC is evolving from a specialized research resource into a strategic component of modern healthcare innovation ecosystems.

ABSTRACT. Enzymes represent an important class of functional ingredients for nutritional supplements, particularly in digestion, nutrient utilization, and targeted biochemical support. Their successful application requires a scientific testing platform that links analytical characterization with functional performance under relevant formulation and physiological conditions. Such a platform integrates three complementary approaches: kinetic analysis, stability profiling, and structure–function evaluation. Kinetic analysis provides measurable information on enzyme activity, substrate specificity, and catalytic efficiency. Determination of parameters such as Km, Vmax, kcat, and kcat/Km supports comparison and selection of enzymes with appropriate functional profiles. For supplement development, kinetic testing can be adapted to simulate the gastrointestinal environment, including variations in pH, temperature, and substrate composition. Stability profiling is equally essential, since enzyme activity may be affected by manufacturing, storage, moisture, excipients, coatings, and interactions with other active ingredients. Assessment of thermal stability, pH tolerance, shelf-life behavior, and residual activity after stress exposure provides insight into formulation robustness and product consistency. Structure–function analysis adds a deeper scientific basis by connecting molecular properties with enzyme performance. Techniques such as structural modeling, circular dichroism, mass spectrometry, and comparative or mutational studies help explain how enzyme conformation, active- site features, and amino acid composition influence activity and stability. By integrating these methodologies, enzyme-based supplement development can move beyond simple activity claims toward an evidence-based formulation strategy. This approach supports rational enzyme selection, blend optimization, compatibility testing, and quality specification, contributing to more consistent and functionally relevant supplement products.

ABSTRACT. Enzymes play essential roles in biological systems and have widespread applications in medicine, biotechnology, environmental science, and industry. Accurate characterization of enzyme structure and function is crucial for elucidating their mechanisms and for the development of novel therapeutics or engineered enzymes. However, traditional experimental approaches are often time-consuming and expensive, particularly given the rapid expansion of genomic and proteomic data. Consequently, computational methods have become indispensable for accelerating enzyme discovery and functional analysis. This review examines the principal computational approaches for enzyme study, emphasizing three core areas: structural modeling, catalytic site prediction, and functional simulation. Techniques such as homology modeling and advanced artificial intelligence tools, including AlphaFold, have significantly improved the accuracy and accessibility of three- dimensional protein structure prediction from amino acid sequences. These structural models serve as foundational resources for subsequent computational investigations into enzyme mechanisms. The review also discusses methodologies for identifying catalytic and ligand-binding sites, including sequence conservation analysis, pocket detection algorithms, and molecular docking techniques. These approaches facilitate the prediction of enzyme substrate specificity and the identification of key amino acid residues involved in catalysis. Additionally, the review addresses molecular dynamics simulations and related techniques, which provide insights into enzyme conformational flexibility and dynamic behavior that static structural models cannot reveal. Integrating structure prediction, active site analysis, and dynamic simulation provides a comprehensive framework for enzyme investigation and experimental planning. These computational strategies offer significant reductions in laboratory time and cost, while advancing efforts in protein engineering, drug discovery, and biocatalyst identification. The review highlights recent progress in artificial intelligence and molecular simulation, and considers the potential future applications of these technologies in enzyme research and biomedical science.

ABSTRACT. Neurodegenerative diseases, including Alzheimer’s disease and Parkinson’s disease, are increasingly recognized as complex multifactorial disorders arising from dynamic interactions between genetic susceptibility, environmental exposures, epigenetic regulation, and molecular dysregulation. Among environmental risk factors, air pollution and particulate matter have emerged as important contributors to neuroinflammation, oxidative stress, mitochondrial dysfunction, and progressive neuronal damage. Understanding these interactions requires the integration of diverse biological and environmental datasets that exceed the analytical capabilities of traditional approaches. This presentation provides an overview of recent advances in precision neurodegenerative medicine through the integration of environmental health research, epigenetics, multi-omics technologies, and artificial intelligence. Particular attention will be given to the role of pollution-induced epigenetic modifications, including DNA methylation changes, chromatin remodeling, and non-coding RNA dysregulation, as potential mechanistic links between environmental exposures and disease progression. Emerging evidence from transcriptomics, genomics, proteomics, and metabolomics studies will be discussed, highlighting how these complementary data layers contribute to the identification of novel biomarkers and therapeutic targets. Furthermore, the presentation explored how artificial intelligence, machine learning, and explainable AI are transforming neurodegenerative disease research by enabling multi-omics integration, biomarker discovery, patient stratification, and predictive modeling. These computational approaches provide new opportunities for identifying high-risk populations, improving early diagnosis, and developing personalized prevention and treatment strategies. The convergence of environmental epidemiology, systems biology, and AI-driven analytics represents a promising framework for advancing precision medicine and supporting a transition from reactive disease management toward proactive and individualized healthcare.

ABSTRACT. Enzyme cofactors represent an important class of functional components for nutritional supplements, particularly in supporting metabolic activity, energy production, and targeted biochemical regulation. Their successful application requires a scientific testing platform that links analytical characterization with functional performance under relevant formulation and physiological conditions. Such a platform integrates three complementary approaches: chemical structure analysis, activation mechanisms, and molecular modulation. Chemical structure analysis provides essential information on cofactor identity, purity, and reactivity. Characterization of metal ions, vitamins, coenzymes, and other organic cofactors supports comparison and selection of components with appropriate functional profiles. For supplement development, structural testing can be adapted to assess cofactor behavior under conditions that simulate the gastrointestinal environment, including variations in pH, temperature, and ionic composition. Activation mechanisms are equally essential, since cofactor function depends on proper binding, availability, and interaction with target enzymes. Assessment of cofactor–enzyme association, activation efficiency, and stability after stress exposure provides insight into formulation robustness and product consistency. These factors may also be affected by manufacturing, storage, moisture, excipients, coatings, and interactions with other active ingredients. Molecular modulation adds a deeper scientific basis by connecting cofactor properties with catalytic regulation. Techniques such as structural modeling, spectroscopic analysis, mass spectrometry, and comparative studies help explain how cofactor structure, binding-site features, and chemical composition influence enzyme activation and overall catalytic performance. By integrating these methodologies, cofactor-based supplement development can move beyond simple activity claims toward an evidence-based formulation strategy. This approach supports rational cofactor selection, blend optimization, compatibility testing, and quality specification, contributing to more consistent and functionally relevant supplement products.

ABSTRACT. Bacterial exoenzymes are catalysts located outside or on the surface of cells. These enzymes enable bacteria to acquire nutrients, interact with hosts, compete with other microorganisms, and facilitate key ecosystem processes. They degrade complex substrates such as proteins, polysaccharides, lipids, and host- derived molecules, which is essential for bacterial survival and adaptation. Advances in structural biology, genomics, and computational biology have significantly enhanced the understanding of exoenzyme mechanisms, regulation, and ecological significance. This review synthesizes current knowledge on the structure and function of bacterial exoenzymes, emphasizing active-site organization, catalytic mechanisms, cofactor requirements, and the regulation of enzyme production and secretion. It further examines the interplay among genetic regulation, cell signaling, and environmental factors in exoenzyme management. The review underscores the accelerating role of computational approaches, including protein structure prediction, comparative genomics, protein language models, machine learning, and multi-omics, in exoenzyme discovery and characterization. Additionally, it evaluates the influence of bacterial exoenzymes on microbiomes, particularly their contributions to community structure, nutrient exchange, probiotic functions, and microbiome engineering. By integrating molecular, computational, and ecological perspectives, this review demonstrates that bacterial exoenzymes are central to microbial interactions with hosts and hold significant potential for applications in biotechnology, medicine, and environmental science.

ABSTRACT. Artificial intelligence (AI) has rapidly emerged as a transformative technology in the healthcare sector, enabling improvements in patient care, diagnosis, and system efficiency. This paper presents a systematic review of AI applications in healthcare, including clinical decision support systems, medical imaging, predictive analytics, robotic surgery, and healthcare administration. It also highlights key challenges such as data privacy, ethical concerns, and algorithmic bias. Finally, future directions including explainable AI, IoT integration, and precision medicine are discussed.

ABSTRACT. The integration of AI-based techniques, such as Deep Learning (DL), into lung cancer diagnostics represents a pivotal challenge in computational oncology, requiring large-scale, high fidelity datasets to ensure model robustness. However, the development of reliable DL pipelines is frequently impeded by data heterogeneity, such as DICOM, NIfTI, and MHD and stringent data privacy regulations. To address these bottlenecks, we present an extensible framework built upon the Medical Open Network for AI (MONAI). Our architecture contains a complete and reliable workflow, which entails four specialized modules: (i) multi source data gathering, (ii) data transformation, (iii) standardized data preprocessing, and (iv) result visualization. By providing a scalable "plug-in" architecture for MONAI, this framework facilitates the integration of several public and private repositories, ensuring technical interoperability for multimodal applications, including PET imaging.

ABSTRACT. The establishment of robotic and AI calibration centre in Vihiga County is essential for improving the accuracy, reliability, and safety of measurement systems in healthcare, industry, and trade. Calibration ensures that equipment operates within acceptable standards by comparing measurements against recognized references. This study examines the need, feasibility, and potential impact of setting up a calibration facility within the county. A descriptive research approach was used, relying on secondary data from government reports, healthcare institutions, and engineering practices in Kenya. Findings indicate that the absence of a local calibration Centre leads to increased operational costs, delays in service delivery, and reduced efficiency due to dependence on distant facilities. The study further highlights the benefits of localized calibration services, including improved healthcare outcomes, enhanced industrial productivity, and strengthened regulatory compliance. The establishment of a calibration Centre is therefore recommended as a strategic investment to support technological growth and socio-economic development within the county. 1. Introduction Calibration is a fundamental process in science, engineering, and healthcare that ensures measurement instruments provide accurate and reliable results. In developing regions such as Vihiga County, access to calibration services remains limited, despite the increasing use of sophisticated equipment in hospitals, laboratories, and small-scale industries. Healthcare facilities, including Vihiga County Referral Hospital, rely heavily on properly calibrated devices such as thermometers, blood pressure monitors, and diagnostic machines. Inaccurate measurements can lead to misdiagnosis, ineffective treatment, and potential risks to patients. Currently, many institutions in the county depend on calibration services from major cities like Nairobi, which results in delays, high costs, and logistical challenges. This paper explores the importance of establishing a local calibration centre to address these challenges and improve service delivery. The establishment of this Calibration, and Testing Center will not only bring Kenya to the forefront of the MedTech industry but also stimulate economic development through job creation and the incubation of startups in the MedTech sector

ABSTRACT. Stargardt disease (STGD1), caused by autosomal-recessive mutations in the ABCA4 gene, is the most common inherited macular dystrophy. Fundus autofluorescence (AF) imaging reveals a characteristic centrifugal sequence of six disease-specific patterns that reflect disease stage and carry clinical significance for grading progression and evaluating the response to emerging treatments. Despite growing interest in deep learning-based analysis of STGD1 images, methodological progress is hampered by the scarcity of large, rigorously annotated image datasets. No dedicated tool currently exists for annotating the full six-pattern taxonomy in AF images of Stargardt disease. This paper presents AnnoStar, an open-source, browser-accessible web application enabling expert annotation of these six patterns in short-wavelength (SW-AF) and near-infrared (NIR-AF) fundus images. Built on a Vue.js (Nuxt) frontend, the tool provides an intuitive mouse-driven interface with automatic overlap resolution, undo/redo support, per-class visibility control, and structured JSON export. AnnoStar constitutes a foundational resource for the construction of annotated datasets that will enable the development of computer-aided diagnosis systems for Stargardt disease.

ABSTRACT. In LMIC hospitals, inefficiencies in health technology management (HTM) contribute to significant hidden costs including downtime, repeated repairs, premature equipment replacement, and lost service capacity. These costs are rarely quantified, limiting evidence-based decision-making and resource allocation in tertiary healthcare facilities. This study aimed to quantify hidden HTM costs and identify key drivers of inefficiency in a Kenyan tertiary hospital to inform improvement strategies. A mixed-methods design was used involving retrospective analysis of 12 months of maintenance and procurement data. Cost-of-downtime estimation models were applied, incorporating equipment downtime, emergency repairs, spare parts usage, and idle capacity due to incompatibility or poor procurement decisions. Key informant interviews with biomedical engineers and clinical staff supplemented quantitative findings. Hidden costs were estimated using productivity loss and replacement cost modeling approaches. Hidden HTM costs accounted for approximately 20–35% of annual equipment-related expenditure. Equipment downtime contributed the largest share of losses, followed by inefficient procurement decisions and delayed maintenance response. Intensive care and imaging departments experienced the highest impact, with productivity losses estimated at up to 28%. Hidden HTM costs are substantial and measurable in LMIC hospitals. Strengthening procurement systems, preventive maintenance, and lifecycle management can significantly reduce inefficiencies and improve healthcare delivery sustainability.

ABSTRACT. Background: The rapid growth of wearable devices and mobile health technologies has accelerated the adoption of the Internet of Medical Things (IoMT) for real-time physiological data acquisition. However, current data transmission and device authentication mechanisms often provide insufficient protection against unauthorized access, raising concerns regarding data integrity, source trustworthiness, and patient privacy.

Objective: This study aimed to develop a secure IoMT data transmission framework based on Fast Healthcare Interoperability Resources (FHIR) APIs and to evaluate the security enhancement achieved by combining mutual Transport Layer Security (mTLS) with OAuth 2.0 compared with conventional unidirectional TLS.

Methods: A secure IoMT data transmission framework based on HL7 FHIR was implemented using an IoMT client, FHIR transformation module, API Gateway, and FHIR server. ECG signals and annotations from the MIT-BIH Arrhythmia Database were converted into FHIR Observation resources, with Provenance and AuditEvent resources added to provide traceability and audit logging. Resources were packaged into FHIR Bundles and transmitted through the API Gateway under two security scenarios: conventional TLS and mTLS combined with OAuth 2.0. TLS provided server-side authentication, whereas mTLS enabled bidirectional authentication using X.509 certificates, with OAuth 2.0 managing user authorization. The API Gateway validated credentials and recorded transmission events. The framework was evaluated for interoperability, resource transmission success, traceability, auditability, authentication effectiveness, resistance to unauthorized access, and compatibility with existing FHIR-based healthcare systems.

Results: The proposed framework successfully achieved standardized FHIR-based encapsulation and transmission of physiological data. Observation, Provenance, and AuditEvent resources provided consistent medical semantics, complete traceability, and audit records. Compared with conventional TLS, mTLS effectively prevented unauthorized requests at the transport layer and strengthened device authentication without requiring modifications to existing FHIR standards. All transmission and access activities were successfully logged and monitored.

Conclusions: This study demonstrates a practical and interoperable IoMT security architecture that integrates FHIR APIs, mTLS, and OAuth 2.0 to enhance data security, device trustworthiness, and traceability. The proposed framework provides a valuable reference for secure IoMT deployment, FHIR API implementation, and healthcare data exchange systems.

ABSTRACT. The relocation of medium- and low-technology medical equipment to the new San Cataldo Hospital in Taranto requires a comprehensive assessment of the existing biomedical equipment fleet to support decision-making on equipment suitability for transfer and replacement due to obsolescence. Methods: The study proposes the implementation and use of a Transferability Index (ITA) as a decision-support tool for assessing the technological status and level of obsolescence of biomedical equipment.

The ITA index was developed starting from models reported in the literature (IPS), which were adapted by introducing two new parameters: clinical adequacy (X4), assessed through questionnaires administered to hospital departments using a Likert scale (1–5), and IT obsolescence (X5), determined based on the software acquisition year. In particular, the index considers seven variables: equipment age relative to its expected service life (X1), availability of spare parts (X2), type of equipment (X3), clinical adequacy (X4), IT obsolescence (X5), device usage condition (X6), and maintenance costs relative to purchase costs (X7). Results: Statistical descriptive and multivariate analyses were conducted to evaluate the behavior of the ITA and the contribution of variables X1–X7. Mean and standard deviation analyses highlighted different levels of parameter dispersion, confirming the multidimensional nature of the index. The distribution of the ITA identified three classes of equipment: transferable (low ITA), obsolete (high ITA), and intermediate cases requiring further technical evaluation. Principal Component Analysis (PCA) revealed the relationships among the parameters and their contribution to the overall variability of the index. The index was applied to 1,939 biomedical technologies, demonstrating the model’s ability to objectively support decision-making processes for the transfer and renewal of the technological asset base.

ABSTRACT. In neonatal care, every second counts — real-time monitoring turns data into life-saving action In most LMICs, there is lack of affordable, continuous and reliable neonatal monitoring systems. Many newborn units face high nurse-to-patient ratios and very limited functional medical equipment. Reliance on manual device checks have led to delayed detection of critical changes to infant’s health and thus infant mortality is on the rise. Hadli provides continuous equipment tracking of certain aspects such as functionality, power fluctuations, temperature and humidity, with automated alerts and data visualization sent to facility BMEs. Some of the challenges faced during patient care include inefficient or unmonitored medical devices, insufficient budgetary allocations towards neonatal device management and increased nonfunctional equipment in hospitals.

This system was developed by NEST360 and it is an IoT based real-time monitoring solution. Its key Objectives were to improve efficiency, safety and device uptime in neonatal care. Currently we are monitoring 26 devices at Pumwani maternity hospital where we can see how the devices are utilized and the frequency of outage. The system comprises of, IoT device sensors, Voltage monitor, Temperature & humidity sensor, central gateway which collects all data from sensors and transmits to cloud. The Cloud-Based Dashboard (HADLI) does remote monitoring and data visualization and alerts. Tracked parameters includes enhanced device uptime, prompt power outage detection, improved maintenance through early warning of faults, optimal thermal conditions → reduces neonatal hypothermia risks which informs data-driven decision-making in equipment budget allocation

Key benefits includes preventive maintenance enabled by real-time alerts, better resource allocation by identifying underused or overburdened devices, improved operational transparency, supports timely clinical interventions and could help in budgeting plans for PPM and continuous maintenance activities – cumulative run hours help in this. Some challenges faced with the Hadli system includes sensors failure over time and not transmitting data, due to hardware related issues like breakages, membrane switch failure, etc. Device users bypassing sensor connections during patient connection hence loss of valuable data. At times, the getaways go offline due to hardware problems and data subscription delays. There are also limitation of use, as sensors are not meant for some devices and SIM and gateway connectivity challenges In conclusion, HADLI system revolutionizes neonatal care in resource-limited settings. The real-time monitoring ensures that there is better outcomes for neonates and more efficient hospital operations and the model can be scaled to other departments within the hospital.

ABSTRACT. Computer vision has emerged as a promising technology for unobtrusive health monitoring in home environments. This study evaluates an AI-enabled biomedical smart mirror designed for remote health monitoring to support independent living among older adults. An interactive workshop was conducted during the 2nd National Geriatrics Conference, where 11 healthcare professionals interacted with the system and evaluated it through usability and user experience surveys. The system exhibited high user acceptance, highlighting its potential as a tool for home-based elderly health monitoring.

ABSTRACT. Medical devices are developed to solve health problems and improve quality of life. Health care providers require medical devices for effective and efficient preventive, diagnosis, treatment and rehabilitation services.

Due to the importance of Medical Devices, the World Health Organization (WHO) expanded its scope and replaced the term with Healthcare Technologies (HT). In Kenya traditionally, use of herbal or alternative medicine approaches was widespread and even today, we still have communities which still practice “herbal medicine”. Kenya also have some religious beliefs which prohibit people from going to hospital when sick and this practice is on the rise. Others believe that treatment using electricity can kill someone quickly and so patient resistance increases.

In Pumwani hospital, we conduct community outreaches every now and then so that we can educate the population on safe deliveries so as to reduce maternal deaths. Unfortunately, the more the device technology advances, the more expensive the treatment becomes and so pregnant mothers may be unable to afford it.

Again, these newer health technologies are not available in most parts of the country. This could be due to lack of infrastructure and other important amenities such as electricity. These technologies have the capabilities to: 1) improve quality of Health delivery service 2) Improve efficiency 3) Improve effectiveness 4) Reduce costs 5) Improve accessibly of health services.

HT management (HTM) has a life cycle including Invention (conception), Production, Installation, Commissioning and User Training, Operation/ maintenance and repair, Decommissioning, Disposal, and then Replacement or new technology. Clinical or Biomedical Engineers (CE-BME) are responsible for management of all health technologies in the hospital and particularly medical equipment for life cycle management noted above as well as solving health care delivery problems for equipment use and improving quality of care.

In Pumwani maternity hospital, most of these management/maintenance services are done in-house but we also out-source some services. We also have partners who assist with donations and required accessories for our equipment, including: 1) NEST 360, who deals with Neonatal Care, https://nest360.org/ 2) Muthaiga rotary club, https://rcnairobimuthaiganorth.org/ and 3) AMREF https://amref.org/ who supplies us with oxygen, and many others resources.

In most cases, we have so much inoperable equipment due to some lack of one or more of the following: 1) Technical skills to repair this equipment 2) Inadequate technical knowledge on HT issues 3) Technical Documentation / Manuals 4) Need spare or repair parts 5) Needed repair funds 6) Administrative support

In summary, Kenya needs the following to change this narrative repeated in many LMIC countries: 1) Clear national HT regulations and standards 2) Clear professional ethics and code of CE-BME practitioner procedures 3) Health policy and guidelines 4) Adequate technical training 5) Hospital Management trust and partnership

The Way forward: (1) HT Policy and guidelines (at the national level); (2) Team work…..and best management practices within CE-BME; and (3) Encouraging innovation and creativity within CE-BME.

In conclusion, proper use and management of health technologies can bring a revolutionary change in the health sector. For health workers, embracing HT is inevitable. For patients, they are becoming increasingly aware of their health rights and available clinical best practices due to accessible information on the internet and other information sources, and are also aware when these practices are not followed. For health leaders, there is an increasing need for CE-BME to expand and communicate their knowledge of best practices for best quality HTM.

ABSTRACT. Combining optical imaging (OI) for functional/molecular data with ultrasound (US) for anatomical structure offers significant synergistic potential. However, the effective integration of these modalities has been hindered by the opacity of conventional transducers, which prevents coaxial light delivery and leads to bulky, inefficient probes. While transparent ultrasound transducers (TUTs) offered a path to integration, progress has been limited to single-element transducers. These are impractical for clinical settings, lacking real-time capabilities and requiring mechanical scanning. We bridge this critical gap with a handheld, tri-modal imaging system built around a 64-channel, 7-MHz transparent ultrasound transducer (TUT) array. This design enables simultaneous and coaxial real-time ultrasound (US), photoacoustic (PA), and fluorescence (FL) imaging from a single, compact probe that integrates all necessary optics and a camera. System capabilities were validated in preclinical models using indocyanine green (ICG) to map blood and lymphatic vasculature. We also demonstrated dual-modal (US/PA) imaging on healthy volunteers, mapping superficial vascular structures and hemoglobin oxygen saturation (sO2). As a key clinical demonstration, the system was employed for guidance during lymphaticovenous anastomosis (LVA) microsurgery in lymphedema patients. The tri-modal data provided precise, real-time localization of both lymphatic and blood vessels, facilitating the procedure. This work establishes the clinical viability of TUT-array technology as a powerful platform for surgical guidance and real-time diagnostics.

ABSTRACT. Cardiovascular diseases (CVDs) remain a leading cause of morbidity and mortality in Europe, with current treatment strategies often failing to account for interindividual variability in drug response. Pharmacogenomics (PGx) offers a promising approach to optimize cardiovascular therapies by integrating genetic information into clinical decision-making. However, its implementation across Europe remains fragmented due to heterogeneity in healthcare systems, regulatory frameworks, data infrastructure, and levels of clinical adoption. The COST Action CA24165 CardioPharmaGENET was established in October 2025 to address these challenges through a coordinated, multidisciplinary, and pan-European network. The Action is structured into five Working Groups focusing on: (1) mapping the current PGx landscape, (2) developing and harmonizing clinical guidelines, (3) integrating artificial intelligence and machine learning tools, (4) analyzing regulatory and policy frameworks, and (5) ensuring effective communication, dissemination, and stakeholder engagement. CardioPharmaGENET aims to bridge the gap between research and clinical implementation by fostering collaboration across scientific, clinical, and policy domains. Within its first six months, the network has engaged over 250 participants from 41 countries, initiated key deliverables, and established a strong digital presence. By enabling knowledge exchange and capacity building, the Action provides a unique platform to advance precision cardiovascular medicine and promote equitable, data-driven healthcare across Europe.

ABSTRACT. Biomedical text simplification aims to make specialized medical language accessible to non-expert readers while preserving its original meaning. However, progress in this area is still limited by the lack of reliable benchmark resources. Existing datasets differ in construction strategy, alignment quality, and linguistic complexity, making fair comparison across models difficult. In this work, we present a comparative evaluation of the main biomedical text simplification datasets, analyzing them in terms of readability and semantic alignment between expert and simplified texts.We further assess two open large language models, Mistral and BioMistral, on the same resources without task-specific fine-tuning. Our results highlight substantial variability across datasets, showing that not all of them provide equally reliable grounds for benchmarking. We also find that BioMistral generally achieves higher overlap-based scores, while both models preserve semantic content to a similar extent according to BERTScore. These findings emphasize the importance of dataset quality for fair and reproducible evaluation and provide a benchmark-oriented reference for future research in biomedical text simplification.

ABSTRACT. While chimeric antigen receptor (CAR) T-cell and natural killer (NK) cell therapies, alongside clustered regularly interspaced short palindromic repeats (CRISPR) genome-editing technologies, have revolutionized treatment for hematological malignancies, their efficacy in solid tumors is hindered by the complex tumor microenvironment, immune evasion, and safety concerns such as cytokine release syndrome. Current artificial intelligence (AI) tools show promise in improving specific stages of therapy development, but they remain fragmented and lack a unified, end-to-end computational framework for integrated design. To address this, we introduce graph immune optimization (GIO), an integrated AI architecture that utilizes graph attention networks to model complex tumor–immune interactions and spatial features. The framework employs conditional generative adversarial networks to optimize CAR constructs and reinforcement learning for CRISPR gene-editing strategy selection, integrated with a digital twin simulation layer that predicts therapeutic responses through a feedback-guided iterative loop. This closed-loop optimization system allows for the continuous refinement of engineered immune cells by identifying optimal therapeutic candidates based on predicted safety, efficacy, and persistence metrics before physical manufacturing. Ultimately, this study provides a robust computational foundation for next-generation precision immunotherapy, leveraging simulation-guided design to overcome current barriers in treating solid tumors and improving the precision and scalability of cell and gene therapies.

ABSTRACT. Objective pain assessment using physiological signals is a significant challenge, with photoplethysmography (PPG) and electrodermal activity (EDA) emerging as promising non-invasive methods. A key assumption in PPG-based approaches is that physiological responses temporally align with pain-stimulation windows; however, this has not been validated. Delays caused by autonomic cardiovascular responses mediated by baroreflex-driven vasoconstriction introduce latency between sympathetic activation and peripheral vascular changes, which conventional labelling schemes do not address. In this study, we utilised the PainMonit dataset, consisting of repeated thermal pain stimulations in healthy volunteers, to explore feature-pain associations across three temporal windows: stimulation, early recovery (10–20 seconds), and late recovery (>20 seconds). We applied Kendall’s Tau correlation and linear mixed-effects models to evaluate these associations and examined the impact of temporally shifted labelling on machine learning-based pain classification. Key findings revealed that significant PPG features peak during early recovery rather than aligning with stimulation. EDA features demonstrated stronger associations during stimulation. Temporally shifted labels reduced bias in PPG models and improved classification in EDA models. These results indicate that the labelling strategy is a critical factor in physiological pain assessment, highlighting its importance in dataset design and model development.

ABSTRACT. Understanding the molecular mechanisms governing protein–receptor interactions is essential for advancing pharmaceutical and biomedical applications. Lysozyme, a small cationic enzyme widely recognized for its antimicrobial activity, also exhibits complex, non-enzymatic interactions with biological membranes and cellular receptors that remain insufficiently characterized at the molecular level. In this study, we employed high-performance computing (HPC)-enabled molecular dynamics (MD) simulations to investigate the interaction landscape between lysozyme and selected biologically relevant targets, including lactoferrin and immune-related receptors such as TLR2, TLR4, and CXCR2. All-atom MD simulations were performed using GROMACS with GPU acceleration, applying the AMBER03 force field and explicit solvent conditions. The simulation protocol included energy minimization, equilibration under constant Number, Volume, Temperature (NVT) and Number, Pressure, Temperature (NPT) ensembles, and 10 ns production runs. Interaction stability and binding characteristics were assessed through structural and dynamic metrics, including interface distance, hydrogen bonding, and contact frequency analysis. The results indicate that lysozyme interactions are predominantly governed by electrostatic and multivalent low-affinity contacts rather than specific receptor–ligand binding. Stable complexes were characterized by persistent hydrogen bond networks and compact interface geometries, while weaker interactions exhibited higher structural variability. The use of HPC resources enabled efficient exploration of multiple protein systems and provided detailed insights into interaction dynamics. These findings contribute to a deeper understanding of lysozyme behavior at the molecular level and highlight the importance of computational approaches in guiding the design of lysozyme-based therapeutic and formulation strategies.

ABSTRACT. This review synthesizes recent evidence (2020–2026) on the roles of collagen genes COL5A1 and COL1A1, and the inflammatory cytokines IL‑6 and TNF‑α, in sports‑related musculoskeletal health and injury. COL1A1 and COL5A1 are central to tendon, ligament, and muscle extracellular matrix integrity, with common polymorphisms (e.g., COL1A1 rs1800012, COL5A1 rs12722) associated with increased risk of tendinopathy, ACL rupture, and muscle strain. Meanwhile, IL‑6 and TNF‑α act as key mediators of exercise‑induced inflammation and tissue remodeling. IL‑6 has been shown to upregulate COL1A1 expression (around 5‑fold in tendon‑derived cells), promoting collagen synthesis and tendon repair, whereas TNF‑α tends to drive catabolic processes and ECM degradation when chronically elevated. The interplay between collagen‑gene variants and cytokine dynamics creates a genotype-inflammatory-phenotype axis that modulates individual susceptibility to overuse and acute soft‑tissue injuries. These findings underscore the potential of combining polygenic risk scores with dynamic cytokine profiling to enable precision‑medicine approaches in sports, including risk stratification, load management, and injury prevention strategies for athletes.

ABSTRACT. In response to the global challenge of bacterial resistance, carbon quantum dots (CQDs) have emerged as biocompatible nanomaterials with remarkable antimicrobial potential. In this study, copper-doped CQDs (Cu-CQDs) were synthesized via a green hydrothermal method using natural extracts from Vitis vinifera and Vaccinium spp. as sustainable carbon precursors. The intrinsic polyphenolic networks and functional groups from these biomasses contributed to stabilizing the carbon core and enhancing cellular affinity. Copper incorporation was designed to modulate the electronic structure of the nanomaterial by introducing intermediate energy levels that promote efficient electron transfer. The obtained Cu-CQDs exhibited characteristic optical properties, including UV absorption (250–350 nm) and intense blue photoluminescence (excitation at 340 nm, emission at 440 nm), confirming the formation of stable carbon nanostructures with functionalized surface groups. Antibacterial assays demonstrated significant inhibitory activity against Gram-positive (Staphylococcus aureus) and Gram-negative (Pseudomonas aeruginosa) bacteria, revealing a synergistic effect between biomass-derived ligands and copper doping. The bactericidal mechanism is attributed to the selective generation of reactive oxygen species (ROS). These findings highlight Cu-CQDs as a versatile and sustainable nanobioactive platform for next-generation antimicrobial therapies.

ABSTRACT. Light-triggered drug delivery provides precise spatial and temporal control over when and where therapies are activated, allowing for on-demand release and reducing systemic toxicity. However, its clinical use is still limited because light does not penetrate deeply into biological tissues, making it difficult to treat tumors located deep within the body. Recently, optical fibers have emerged as promising tools to address this challenge by delivering light directly to target sites in a minimally invasive and localized way. In this review, we focus on optical fibers as platforms for controlled and feedback-enabled drug delivery, shifting attention from using light as the main therapy to its role in regulating drug activation. We begin by outlining the basic mechanisms of light-responsive drug delivery, including photothermal, photodynamic, and photochemical processes, and emphasize how these approaches can control drug release. Next, we examine fiber-based systems in terms of their materials, designs, and release strategies, highlighting methods such as surface functionalization, nanoparticle-assisted delivery, and hybrid multimodal platforms. We pay special attention to the integration of sensing and actuation within fiber-based systems, which enables real-time monitoring of the tumor microenvironment and supports feedback-informed therapeutic strategies. While fully autonomous closed-loop systems are still in development, current examples already show the potential for adaptive control of drug activation. Finally, we discuss key challenges for clinical translation, such as material stability, drug loading capacity, device integration, and regulatory issues. By positioning optical fibers as multifunctional components of drug delivery systems, this review highlights their potential to support interventional, programmable, and adaptive therapeutic strategies for precision oncology.

ABSTRACT. Atrial fibrillation is the most treated type of arrhythmia in clinical practice. Such arrhythmias represent a significant risk factor for the development of ischemic stroke, peripheral embolism, and can ultimately lead to death. This research paper evaluates the cost-effectiveness of switching patients to therapy with novel oral anticoagulants (NOACs) compared to warfarin therapy. The study is based on the perspective of the healthcare system of Bosnia and Herzegovina, and it compares the value of NOAC therapy for stroke prevention in patients with atrial fibrillation with traditionally used Warfarin therapy. The pharmacoeconomic analysis covered a time horizon of 15 years with a discount rate of 3% and included a complete evaluation with appropriate methods such as CEA (Cost-Effective Analysis), CUA (Cost-Utility Analysis), CBA (Cost-Benefit Analysis), BIA (Budget Impact Analysis) and sensitivity analysis.

ABSTRACT. Biomimetic scaffolds incorporating bioactive cues hold significant promise for tissue regeneration. Natural polymers, such as hyaluronic acid (HA), provide intrinsic biological functionality capable of modulating cellular responses. Electrospinning is a versatile technique for fabricating nanofibrous porous matrices that closely resemble the architecture of the extracellular matrix. In this work, HA was incorporated into nanofibrous membranes with high surface-area-to-volume ratios through electrospinning process. Processing parameters were optimized to maximize HA content while maintaining fiber uniformity and structural integrity. The resulting scaffolds exhibited a homogeneous nanofibrous morphology. To improve stability under physiological conditions, post-fabrication crosslinking was implemented. The developed scaffolds demonstrate strong potential for tissue engineering applications, combining biomimetic structure with bioactive functionality. Additionally, this approach enables the fabrication of three-dimensional tubular constructs and hybrid materials, expanding its applicability in regenerative strategies.

ABSTRACT. Resting heart rate variability (HRV) has been widely investigated as a psychophysiological marker of anxiety- and stress-related states, yet findings remain heterogeneous due to the use of non-equivalent psychological measures, redundant HRV metrics, and limited consideration of time-of-day effects. This study aimed to identify which questionnaire-derived dimensions of psychological burden, spanning anxiety, perceived stress, and distress, are most consistently associated with the latent structure of resting HRV, and to determine whether these associations vary across times of day. A total of 106 adults underwent standardized 300 s resting ECG recordings obtained in the morning, afternoon, or under night-time/sleepiness conditions. After quality control, 283 valid resting windows were retained. A broad set of HRV indices was extracted and reduced via principal component analysis (PCA), and associations with psychological questionnaires were examined using exploratory Spearman correlations and generalised additive mixed models (GAMMs), adjusted for age, sex, body mass index, and habitual physical activity. Four latent HRV components explained 91.8% of total variance. After adjustment, the PCA component dominated by short-term vagally mediated variability was the only HRV dimension showing evidence of association across all six questionnaire totals, whereas the remaining components showed only selective or more context-dependent effects. These associations were strongest at night and generally attenuated in the afternoon. Overall, the findings support a multidimensional, physiologically interpretable, and time-aware approach to resting HRV assessment in anxiety- and stress-related research.

ABSTRACT. Ultrasound-guided femoral vascular access is widely used to improve procedural safety, but effective training requires realistic simulators reproducing anatomical landmarks, ultrasound appearance, haptic feedback, and flow-related confirmation of successful puncture. This work presents a patient-specific, flow-enabled ultrasound phantom for femoral vascular access training. A CT-derived mold was used to reproduce the inguino-femoral anatomy, while 3D-printed removable cores enabled controlled positioning of the femoral artery, arterial bifurcation, femoral vein, and inguinal ligament. PVC-plastisol was selected as tissuemimicking material, and heat-shrinkable PTFE tubes were used to reproduce the arterial wall, providing both ultrasound contrast and perceptible resistance during needle penetration. A closed-loop flow circuit was integrated to provide visual fluid return after successful puncture and to support Doppler-based vessel differentiation. Preliminary face and content validity were assessed by eight experienced endovascular surgeons using a structured 24-item Likert questionnaire. The phantom received positive evaluations for ultrasound landmark visibility, haptic arterial-wall feedback, usability, and training utility. The arterial bifurcation was rated as a reliable anatomical landmark, and the flow system was identified as a meaningful improvement over conventional phantoms. These findings support the proposed simulator as a cost-effective and reusable platform for structured ultrasound-guided femoral access training.

ABSTRACT. Colorectal cancer (CRC) screening is limited by cost, invasiveness, and scalability, motivating investigation of non-ionising alternatives. This study assesses a cost-conscious RF/microwave medical imaging (MMI) proof-of-concept for CRC using electromagnetic simulations and tissue-mimicking material (TMM) fabrication. The work comprised full-wave electromagnetic simulations of tumour and non-tumour cases, dielectric characterisation using a single-port open-ended coaxial probe (OECP) with Komarov inversion implemented in PyOECP, and fabrication of gelatine-based colon and healthy-tissue analogues. The dielectric characterisation pipeline achieved <4% average permittivity error on reference liquids, enabling iterative tuning of TMMs over the 2-4 GHz band. Simulations showed measurable tumour-induced scattering perturbations, with peak SAR of 0.022 W/kg, below the cited 4 W/kg limit.

ABSTRACT. Computed tomography (CT) is a commonly utilized diagnostic imaging method that relies on ionizing X-ray radiation, which can pose potential risks to patients. The challenge of reconstructing CT images under low doses of X-rays persists, as conventional algorithms often lead to considerable degradation in image quality. Although advanced iterative reconstruction techniques partially mitigate this problem, recent approaches based on deep learning have exhibited impressive performance, albeit at the expense of needing extensive, high-quality training datasets that are often scarce in the field of medical imaging. In this study, we introduce a novel iterative reconstruction technique that incorporates neural-network-based modeling into the reconstruction process without the need for any training data. This proposed method seeks to leverage the significant representational capabilities of neural networks while maintaining the flexibility and robustness characteristic of model-based iterative methods. Initial results obtained from phantom data containing human-like tissue inserts indicate that the method can successfully reconstruct high-quality images and evidence of its efficacy in scenarios involving reduced doses of X-rays.

ABSTRACT. Virtual populations are a promising approach for investigating the cardiovascular system under both physiological and pathological conditions. The accuracy of virtual populations in representing the hemodynamics of a real population depends strongly on the subject-specific model used for their generation. Currently, the majority of the subject-specific cardiovascular models available in the literature account only for age, neglecting the effects of body size and sex differences. Moreover, the models often present an open-loop architecture and do not provide information on some important cardiovascular regions such as the cerebro-ocular and coronary circulations. In this work, we propose a new subject-specific multiscale cardiovascular model constructed starting from a validated closed-loop multiscale cardiovascular model previously applied to study arrhythmias, postural changes, and the effects of micro/hyper-gravity on the cardiovascular system. The proposed subject-specific model was validated against clinically relevant hemodynamic variables and used to generate a virtual population of 300 healthy subjects in supine posture. The model showed good agreement with clinically observed age- and BSA-related trends. As a representative example of its practical applicability, it captures the role of age and BSA in shaping pressure waveforms along the aorta.

ABSTRACT. Medical devices are essential to modern health systems and central to achieving universal health coverage, strengthening health security, and expanding access to diagnosis, clinical monitoring, treatment, rehabilitation, prevention, emergency care, and assistive technologies in low- and middle-income countries (LMICs). In resource-constrained settings, well-designed technologies can address critical gaps across maternal and newborn care, infectious and non-communicable diseases, emergency response, surgery, and digital health. However, regulatory pathways for clinical investigation, evaluation, and market authorization in many LMICs remain fragmented, under-resourced, and inconsistently implemented which hinders translation of medical device innovation into safe and effective use. This is particularly evident in Africa, where device clinical trials are limited, unevenly distributed, and often disconnected from emerging local innovation ecosystems despite growing biomedical engineering capacity. This policy-or

ABSTRACT. Biomedical engineering (BME) plays a critical role in strengthening healthcare systems, particularly in low- and middle-income countries (LMICs), where medical equipment functionality and sustainability remain persistent challenges. Ghana has made notable progress in developing BME education and training; however, gaps remain in practical competency development, infrastructure, and clinical integration. This paper presents a systems-based approach, aimed at improving biomedical engineering education and training in Ghana; integrating academic programmes, equipment-centered training, and professional upskilling initiatives. It elaborates key interventions, including the GIZ develoPPP programme and the Korea Foundation for International Healthcare (KOFIH), and examines the role of adaptive digital learning platforms. A multi-tier training model and a retrofitted equipment-based workshop framework are proposed. Findings indicate that combining hands-on training with adaptive learning technologies significantly improves competency development and healthcare technology management outcomes. The paper concludes with recommendations for scaling sustainable training systems in LMICs.

ABSTRACT. In neonatal care, every second counts — real-time monitoring turns data into life-saving action In most LMICs, there is lack of affordable, continuous and reliable neonatal monitoring systems. Many newborn units face high nurse-to-patient ratios and very limited functional medical equipment. Reliance on manual device checks have led to delayed detection of critical changes to infant’s health and thus infant mortality is on the rise. Hadli provides continuous equipment tracking of certain aspects such as functionality, power fluctuations, temperature and humidity, with automated alerts and data visualization sent to facility BMEs. Some of the challenges faced during patient care include inefficient or unmonitored medical devices, insufficient budgetary allocations towards neonatal device management and increased nonfunctional equipment in hospitals.

This system was developed by NEST360 and it is an IoT based real-time monitoring solution. Its key Objectives were to improve efficiency, safety and device uptime in neonatal care. Currently we are monitoring 26 devices at Pumwani maternity hospital where we can see how the devices are utilized and the frequency of outage. The system comprises of, IoT device sensors, Voltage monitor, Temperature & humidity sensor, central gateway which collects all data from sensors and transmits to cloud. The Cloud-Based Dashboard (HADLI) does remote monitoring and data visualization and alerts. Tracked parameters includes enhanced device uptime, prompt power outage detection, improved maintenance through early warning of faults, optimal thermal conditions → reduces neonatal hypothermia risks which informs data-driven decision-making in equipment budget allocation

Key benefits includes preventive maintenance enabled by real-time alerts, better resource allocation by identifying underused or overburdened devices, improved operational transparency, supports timely clinical interventions and could help in budgeting plans for PPM and continuous maintenance activities – cumulative run hours help in this. Some challenges faced with the Hadli system includes sensors failure over time and not transmitting data, due to hardware related issues like breakages, membrane switch failure, etc. Device users bypassing sensor connections during patient connection hence loss of valuable data. At times, the getaways go offline due to hardware problems and data subscription delays. There are also limitation of use, as sensors are not meant for some devices and SIM and gateway connectivity challenges In conclusion, HADLI system revolutionizes neonatal care in resource-limited settings. The real-time monitoring ensures that there is better outcomes for neonates and more efficient hospital operations and the model can be scaled to other departments within the hospital.

ABSTRACT. Africa’s health innovation challenge is no longer a shortage of ideas; it is the failure of promising ideas to move reliably from student projects, laboratories, hackathons and pilot studies into regulated, manufactured, adopted and sustained solutions. This opinion review argues that Africa’s innovation-commercialization gap is also a health-systems gap: when local innovations fail to scale, health systems remain dependent on imported technologies that are often unaffordable, poorly maintained or mismatched to local realities. The paper positions the African Biomedical Engineering Consortium (ABEC) as a practical continental platform for addressing this gap through human capital development, curriculum harmonization, innovation training, South-South collaboration and ecosystem building. ABEC’s future value lies in evolving from a training and networking platform into a full translation ecosystem linking education, prototyping, clinical validation, regulation, manufacturing, procurement, financing and policy uptake.

ABSTRACT. Diabetic foot complications remain a critical challenge in Latin America due to fragmented clinical workflows and the absence of data-driven preventive tools. This paper introduces Diapetics, a robust translational framework that operationalizes the transition from clinical multi-source data to precision medical device fabrication. Grounded in patent No. NC2021/0015248 (Resolution No. 51357, July 30, 2025) —Technology Readiness Level (TRL) 4—, the technology integrates a cloud-based architecture with a dual-stage fuzzy logic inference engine designed to handle the stochastic nature of clinical variables. The first computational module utilizes weighted Mamdani-type fuzzy sets to process systemic risk factors, while the second module refines this through region-specific neuropathy and vasculopathy mapping for personalized insole design. To ensure scalability, the framework is designed for integration within the Electronic Health Records (EHR) of the Colombian public health system, aiming for national coverage with a modular architecture transferable across Latin America. Future work aims to expand this patient-centered model to encompass all diabetes-related complications, fostering personalized care and digital patient empowerment. Currently in the proof-of-concept phase, Diapetics demonstrates high fidelity in translating complex clinical reasoning into effective, scalable prescriptions for resource-constrained environments.

ABSTRACT. The aim of the presentation is to inform about the situation of the professions of biomedical engineer and medical physi-cist in the Czech healthcare system and Slovakia. The system of undergraduate, postgraduate, and lifelong education in the fields of Biomedical Technology, Biomedical Engineering, and Medical Physics (MP) is regulated by the Act No.96/2004 Coll. on non-medical health service occupations and related regulations. The Act and related regulations define position of technical personnel in the health service system. This legal regulation distinguishes the following categories of technical personnel: another professional, health service professional with technical competence (bio-medical technician (BMT) – B.Sc., biomedical engineer (BME) – M.Sc.), and health service professional with special-ized competence (clinical technician – B.Sc., clinical engineer – M.Sc.), and medical physicists. There exists a system of accredited medical institutions that can perform defined types of postgraduate and lifelong education. As part of lifelong education, credit system was introduced that specifies activities for which credit points can be awarded. The Act No. 96/2004 Coll. specifies the following types of educations: undergraduate education, i.e. bachelor and master study (minimum requirements are given by the official regulations No. 39/2005 Coll.); accredited qualification course; specialized education; lifelong education (this education is obligatory for all health service professionals and also for those working in the category another professional in health service). Basic rules and requirements concerning undergraduate (Bc. and M.Sc.) and postgraduate (Ph.D.) study in general are defined by the Higher Education Act No. 111/98 Coll. and its amendment No. 137/2016 Coll. In this context, it is necessary to stress that in addition to standard accreditation of a study programme or field of undergraduate educa-tion performed by the Accreditation Board of the Ministry of Education, Youth and Sports of the Czech Republic (MEYS), the biomedical study programmes or fields must get the accreditation of the Ministry of Health Care of the Czech Republic (MHC) in the sense of the Act No. 96/2004 Coll. and related regulations. The graduates of these ac-credited fields get the certificate of qualification to perform health service occupations according to the Act No. 96/2004. The BME (BMT) qualification can be obtained by graduation in the Biomedical Engineering field of study (BME) or the Biomedical Technology field of study (BMT). Graduates of another bachelor or master study pro-gramme in electrical engineering can obtain the qualification for health service professionals with technical compe-tence if they pass the accredited course in Biomedical Engineering (for M.Sc./Eng.) or Biomedical Technology (for B.Sc.) The accreditation for these courses is delivered by the MHC. The conditions for this accreditation are defined in the official regulation No. 424/2004 Coll. The MP qualification can be obtained by graduation in the Medical Phys-ics study programme (MSc.) only. In Slovakia, there are similar study programmes and legal regulations. However, the professions of BME and MP are defined as “another healthcare professional” (Act. No. 578/2004, governmental regulation No. 296/2010 and No. 321/2005). It is necessary to note that BME is studied as independent study programme, similarly to other European countries. However, qualification of MP (in Slovakia named clinical physicist) is acquired in two steps: 1. Graduation in MSc. programme physics, nuclear physics or biomedical engineering; 2. Specialized study in clinical physics, guaranteed by a medical faculty.

ABSTRACT. Biomedical engineering is becoming increasingly relevant in healthcare, within the profound transformation and growing complexity of hospital environments. This trend is driven by the emergence of disruptive technologies such as massive genomic sequencing, digital health, artificial intelligence, cloud and edge computing, robotics, digital twins, virtual and augmented reality, 3D printing, regenerative medicine and nanomedicine, to name just a few. This paper raises the need for a deep reflection on the main roles of engineering in healthcare systems, provides an initial analysis of the situation of engineering within the Spanish National Health System, outlines the main challenges and difficulties for the incorporation of engineers, proposes an initial description of professional profiles, roles, and competencies, and aims to address the urgent need for a training program for internal resident engineers.

ABSTRACT. Trustworthiness and reliability are difficult to embed in a concrete, verifiable manner in AI-driven systems—and even harder to demonstrate or assess. This is a particular challenge in the medical domain, where AI-based systems are increasingly present on the market. Many of these systems are de facto agentic: they can plan, decide, and act across multi-step sequences with a degree of autonomy that classic interactive systems never possessed. Yet these agents often fail to help clinicians grasp the reasoning behind proposed actions, and rarely allow clinicians and health operators to ask for clarification dynamically during interaction. This raises a fundamental question: if AI-driven devices merely deliver algorithmic outputs, what added value does AI actually provide? The real potential lies in systems that go beyond algorithmic decisions and become tools for exploring alternative decision-making pathways together with clinical operators. This requires systems whose reasoning operators can genuinely understand and interrogate. The EU AI Act is clear that transparency and explainability are essential to trustworthiness and to protecting fundamental rights. Nevertheless, explainability is too often conceptualised as a desirable add-on rather than a design requirement. In this talk, I argue for a shift from post-hoc explainability to explainability by design, and from technical transparency to perceived explainability—explanations that are meaningful, actionable, and calibrated to the clinical context and the operator's needs. This reframes explainability from a purely technical property into a measurable, user-centred one. We define perceived explainability as the degree to which an AI-driven interface makes perceivable, at any moment of an agent's action sequence: (a) the reasoning behind the agent's behaviour, (b) the means to request further clarification, and (c) the means to act on disagreement. In this session, we will discuss our proposed approach to assessing perceived explainability as a measurable proxy for the usability—and ultimately the trustworthiness—of AI-driven medical systems.

ABSTRACT. Overview regarding the fundamental motivations in favor of "Open Data" and current developments, organisations and developments with emphasis on applications in health care.

ABSTRACT. Artificial intelligence promises a profound transformation of clinical practice. However, its translation into routine care remains constrained by a structural bottleneck: the scarcity of high-quality, diverse, and accessible real-world data. Privacy regulations, the statistical rarity of orphan conditions, fragmented and unstructured electronic health records, and the underrepresentation of minority populations together confine usable clinical data within institutional silos. This talk examines how Generative AI can address this data availability crisis within the broader vision of digital health. The architectures driving the field are first reviewed, from the Transformer backbone to large language models for clinical text, generative adversarial networks and diffusion models for high-fidelity medical imaging, variational autoencoders for structured records, and multimodal models that bridge notes, scans, and omics. Three complementary strategies are then analysed, namely the synthetic generation of privacy-preserving cohorts, the targeted augmentation that rebalances rare classes, and the cross-modality imputation that recovers missing longitudinal information. The constraints that govern responsible adoption are further discussed, including trustworthiness and hallucination, limited explainability and auditability, the computational and environmental cost of foundation models, and the risk of model collapse when systems are trained recursively on their own synthetic outputs. Generative AI offers a credible path towards democratising healthcare data, provided that its deployment is anchored in rigorous validation, in federated and human-in-the-loop governance, and in a sustained commitment to patient safety and clinical reliability.

ABSTRACT. The rapid digitalization of healthcare, enabled by integrated information systems such as Hospital Information Systems (HIS) and Electronic Health Records (EHR), has improved care quality while substantially expanding the cybersecurity attack surface. This study presents an integrated framework for analyzing the interdependence among healthcare workflows, security threats, and protection mechanisms in this evolving landscape. Healthcare workflows are examined along two dimensions: intra-hospital operations and inter-hospital workflows. The former involves internal clinical processes, system integration, and access management, while the latter includes health information exchange (HIE) and telemedicine, which introduce additional complexity through distributed systems and external network communication. Major threats to healthcare operations and patient safety are discussed, including ransomware, application vulnerabilities, social engineering, and supply chain attacks. Corresponding technological and managerial protection mechanisms are then reviewed, including authentication, access control, cryptographic techniques, security governance, and business continuity planning. To ground the framework in practice, Taiwan’s healthcare infrastructure is examined as a case study, including the National Health Insurance (NHI) IC card system, Healthcare Certificate Authority (HCA), and Electronic Medical Record Exchange Center (EEC), as practical implementations of secure identity verification and cross-institutional data exchange. Finally, emerging trends and challenges are discussed, including the decentralized potential of blockchain for secondary data use, the dual role of artificial intelligence as both a security defense mechanism and a source of novel attack vectors, as well as the growing need for post-quantum cryptography (PQC) to safeguard long-term medical data against future quantum computing threats.

ABSTRACT. Distinguishing vascular from non-vascular cutaneous ulcers is a critical clinical task that drives treatment selection. However, in non-specialist settings where most chronic ulcers are first assessed, misdiagnosis rates remain alarmingly high. This paper describes the design of an explainable binary classification pipeline that addresses this gap by combining three architecturally distinct backbones through late fusion: EfficientNet-B0 (a convolutional neural network), BiomedCLIP (a biomedical vision--language Vision Transformer), and UNI (a computational pathology Vision Transformer). The proposed fusion configuration aims to evaluate the individual and combined contributions of each backbone systematically. The pipeline incorporates Macenko colour normalisation to handle photographic variability, and dual explainability (GradCAM++ for the CNN and attention rollout for the Vision Transformers) to support clinical interpretability. The architectural rationale and the training protocol are presented and discussed for the intended clinical deployment as a triage tool in primary care and telemedicine.

ABSTRACT. Accurate prediction of 3-month functional outcome after endovascular thrombectomy (EVT) in patients with acute ischemic stroke due to large vessel occlusion (LVO) remains challenging. Quantitative EEG (qEEG) provides a non-invasive assessment of brain activity and may offer complementary information for prognostic modeling. This study aimed to evaluate whether qEEG-derived features improve outcome prediction within an interpretable Naive Bayes framework. We analyzed 71 patients with LVO stroke treated with EVT. Functional outcome at 3 months was dichotomized as good (modified Rankin Scale ≤2) or poor (modified Rankin Scale >2). EEG recordings were processed offline to extract relative spectral power in delta, theta, alpha, and beta bands, together with the delta/alpha ratio (DAR) and the (delta + theta)/(alpha + beta) ratio (DTABR). Candidate predictors were selected through a univariate feature-selection step and entered into Naive Bayes classifiers. Age, admission NIHSS, hypertension, recanalization status, symptomatic intracranial hemorrhage, and EEG-derived features including delta power, alpha power, DAR, and DTABR were associated with outcome. The clinical-radiological model achieved 68% accuracy and an AUC of 0.73, whereas the EEG-enhanced model reached 73% accuracy and an AUC of 0.78. These findings suggest that qEEG features provide complementary neurophysiological information useful for improving outcome stratification after thrombectomy.

ABSTRACT. Physical exercise in individuals with Type 1 Diabetes (T1D) significantly increases the risk of hypoglycemia, and despite the widespread adoption of continuous glucose monitoring (CGM) systems, its prediction remains a challenging task. Deep learning classifiers, such as Bidirectional Long Short-Term Memory (Bi-LSTM) networks, have shown promising results as a clinical decision support system, but their performance is often limited by data scarcity and severe class imbalance. To address these limitations, this study proposes a framework for data augmentation purposes based on Generative Adversarial Networks (GANs). In particular, we propose mWGAN-GP, a modified version of the Wasserstein GAN with Gradient Penalty (WGAN-GP), incorporating a diversity regularization term into the generator loss function to improve training stability and variability in the generated signals. This strategy promotes the augmentation of diverse and physiologically plausible CGM sequences. Experimental results on the DirecNet dataset show that mWGAN-GP-based augmentation improves classification performances, with an AUC of 90.9% with respect to that obtained with conventional data augmentation techniques (AUC=88.6%). These findings highlight the potential of GAN-based augmentation to enhance model generalization and support the development of more reliable predictive systems for T1D management.

ABSTRACT. Neurodevelopmental disorders (NDDs) such as attention-deficit/hyperactivity disorder (ADHD) and specific learning disorders continue to go undiagnosed during early childhood because of the lack of access to specialist evaluation and delays in referral channels. Non-invasive screening instruments that can be deployed in schools, at large volumes, are thus needed. The paper presents a serious-gaming platform called Dino by Paperbox Health, which targets children aged between 4 and 8 years of age. Interaction features, which are captured by the system, include response time, accuracy and behavioural variability and allow the analysis of cognitive performance to be done using data. Dino is school-deployable and has age adaptive interaction in Reception, Key Stage 1 and Key Stage 2. The platform structure, the gameplay, and the mood tracker system are shown. There is pre-deployment that proves possible and high participation in classroom environments. Dino offers a scalable baseline of early screening assistance and the subsequent AI-driven digital health applications.

ABSTRACT. Photoacoustic microscopy (PAM) has become a vital tool in biomedical research, providing high-resolution 3D anatomical and functional imaging. While handheld PAM systems are preferred for clinical and point-of-care applications, achieving a compact design without sacrificing imaging speed or spatial resolution remains a significant challenge. This study demonstrats a practical handheld PAM probe that integrates a resonant fiber scanner with a high-frequency transparent ultrasound transducer. The compact device, measuring 17 mm in diameter and weighing only 11 g, achieves high lateral and axial resolutions of 7 and 47 μm, respectively, with a 2.6 mm diameter field of view. Volumetric images are acquired in 1.5 s, effectively minimizing motion artifacts during handheld operation. We validated the system through in vivo imaging of surgically exposed rat abdominal organs, including the stomach and intestine. Furthermore, we investigated the high-speed monitoring capabilities of the system by characterizing the temporal profile of epinephrine-induced vasoconstriction in vivo. Our quantitative analysis of vascular responses highlights the probe's potential for real-time pharmacological assessment and intraoperative guidance.

ABSTRACT. Wearable inertial sensors are widely used in sleep, mobility, and digital-health studies, but their potential for beat-to-beat cardiac monitoring from non-chest locations remains poorly explored. We investigated whether heart rate (HR) and heart rhythm information can be recovered during sleep from inertial measurement units (IMUs) placed at the lower back and wrists. Nine healthy adults underwent overnight at-home monitoring with lower back and wrist IMUs and a wearable electrocardiograph. Sleep posture was estimated from the lower-back IMU, and heartbeat detection was performed within stable posture bouts using a template-matching cross-correlation approach. Performance was evaluated at the beat-to-beat level using RR intervals and at the 5-minute window level using mean HR and RMSSD index of HR variability. Estimation of RR intervals was best from the lower-back IMU (concordance correlation coefficient, CCC = 0.91 ± 0.30), albeit with a marked reduction in the supine posture, while the dominant and non-dominant wrist showed lower performance (CCC = 0.65 ± 0.23 and 0.63 ± 0.29, respectively). Mean HR was recovered reliably across locations, whereas RMSSD estimates were less robust. These results support the feasibility of opportunistic nocturnal cardiac monitoring from sparse inertial sensor placements, with the lower back emerging as the most promising location.