ABSTRACT. Aims To derive intraoperative dynamic hip–knee–ankle (dHKA) alignment phenotypes using unsupervised machine learning (ML) and to determine whether phenotype preservation after femoral cuts is associated with KOOS Jr. improvement within a tibia first total knee arthroplasty (TKA) workflow.

Methods Pre cut and post cut dHKA were acquired under neutral manipulation with a force controlled intra articular distractor integrated with computer assisted orthopaedic surgery at 12 flexion angles. A multi surgeon dataset (11 surgeons; 1,890 tibia first TKAs) informed K means clustering trained on pre cut vectors; internal validity metrics selected an eight-angle feature set (10°, 20°, 30°, 45°, 60°, 75°, 90°, 105°) and four clusters. Centroids/boundaries were fixed and applied to post cut data for boundary consistent labelling. Outcomes were assessed in a single surgeon cohort with KOOS Jr. available (n = 141). The primary endpoint was one-year KOOS Jr. improvement. Comparisons of cluster matched versus cluster shifted cases were performed overall and within pre cut cluster strata (subgroups n ≥ 10); Welch t test, α = 0.05.

Results Four phenotypes were identified: valgus/neutral, neutral, low to moderate varus, and moderate to high varus. Phenotype preservation occurred in 72.3% (102/141). Mean KOOS Jr. improvement was greater with preservation than with shift (32.0 ± 19.5 vs 27.2 ± 14.7; p = 0.117). Within low to moderate varus pre cut knees, preservation yielded larger gains than shift to a more varus phenotype (34.8 ± 17.4, n = 36 vs 22.1 ± 15.0, n = 10; p = 0.036). Neutral pre cut knees showed no difference (29.8 ± 20.5, n = 34 vs 28.0 ± 20.1, n = 10; p = 0.808); other series lacked sufficient shifted cases.

Conclusion Force controlled intraoperative dHKA combined with unsupervised ML produced reproducible dynamic alignment phenotypes and enabled boundary consistent assessment across surgical steps. Phenotype preservation showed directionally greater one-year KOOS Jr. improvement—significant for low to moderate varus—supporting phenotype guided, patient specific alignment targets. Larger, multi surgeon cohorts are required for validation.

ABSTRACT. Total knee arthroplasty is one of the most frequently performed procedures for the treatment of end-stage knee osteoarthritis, but studies point out that only approximately 80% of patients are satisfied with the post-operative results. It is hypothesized, that restoring native tibio-femoral kinematics may enhance patient outcomes. However, substantial inter-individual variability in physiological knee motion suggests that personalized implant selection is necessary to replicate the individual natural joint function. Modern navigation and robotic‑assisted systems enable intra-operative kinematic assessment, providing real‑time information to optimize implant selection and component positioning. However, the inter-observer reliability of intra-operative kinematic measurements during passive knee flexion remains unclear. Demonstrating reliable inter-observer agreement is essential to determine whether kinematic phenotyping can effectively guide personalized implant decision-making. The present study investigated inter‑observer reliability of knee kinematics during passive knee flexion in ten human donor knees. Three experienced surgeons performed a standardized passive flexion movement from full extension to 90° flexion prior to any surgical incision, while the OrthoPilot® navigation system continuously recorded the tibio-femoral motion using an infrared camera and reflective markers. The condylar motion was visualized by projecting the femoral flexion axis onto the tibial plateau at different flexion angles. Although minor deviations in absolute condylar positions were observed between surgeons, the overall kinematic patterns were consistent for each knee. Importantly, distinct differences between knees, such as a lateral pivot pattern or a pronounced femoral rollback, were preserved across observers. These findings demonstrate that intra‑operative passive kinematic measurements exhibit sufficient inter‑observer reliability to identify individual kinematic phenotypes.

ABSTRACT. Background: Robot-assisted total knee arthroplasty (TKA) has been shown to reduce radiographic outliers. With improved implant positioning accuracy, functionally aligned (FA) TKA has been introduced to restore native, pre-arthritic knee kinematics while minimizing periarticular soft tissue release. ROSA is an image-based robotic system that enables intraoperative planning based on soft tissue laxity and allows quantitative assessment of postoperative laxity. However, evidence regarding FA TKA performed using the ROSA system remains limited. This study aimed to compare short-term clinical outcomes and soft tissue laxity between FA and mechanically aligned (MA) TKA using ROSA. Methods: Thirty-one knees with varus osteoarthritis underwent MA TKA between February and December 2021, and 40 knees underwent FA TKA between October 2023 and November 2024. All procedures were performed using a single implant design with posterior cruciate ligament resection. Postoperative medial and lateral laxities were measured at 0° extension and 90° flexion. Radiographic alignment was assessed using the hip–knee–ankle (HKA) angle. Clinical outcomes, including range of motion, the 2011 Knee Society Score, and the Forgotten Joint Score-12 (FJS-12), were evaluated up to 1 year postoperatively. Results: Postoperative HKA differed significantly between the groups. Medial laxity at 90° flexion was significantly greater in the MA group. FJS-12 was significantly higher in the FA group at 3 and 6 months postoperatively, whereas no significant difference was observed at 1 year. Conclusions: FA TKA using ROSA demonstrated superior early postoperative FJS-12 compared with MA TKA; however, this advantage was not maintained at 1 year postoperatively.

ABSTRACT. Background: Image-based robotic-arm assisted total knee arthroplasty (R-TKA) enables the precise execution of personalized surgical strategies. However, inaccuracies in achieving the intended posterior tibial slope (PTS) have been reported, even with advanced robotic systems. These discrepancies may be attributed to differences in anatomical landmarks used in CT-based planning compared to those used in standard radiographs. This study proposes a modified tibial knee center (MTKC) designed to improve agreement between CT-based planning and radiographic PTS measurements. Methods: This retrospective study included 204 knees that underwent primary R-TKA. The MTKC was developed by posteriorly shifting the proximal tibial reference point based on a subset of 45 knees and subsequently applied to the entire cohort. Preoperative PTS was assessed using standard lateral radiographs and CT-based planning, referencing both the traditional tibial knee center (TTKC) and the MTKC. Radiographic and CT-based PTS values were compared using paired t-tests and Bland–Altman analysis, and equivalence between postoperative radiographic and MTKC-based target PTS was assessed using TOST. Results: TTKC-based preoperative PTS significantly overestimated radiographic PTS (p < 0.001), whereas MTKC-based preoperative PTS showed no significant difference (p = 0.605). Agreement between MTKC-based target and radiographic PTS was confirmed by Bland–Altman analysis and equivalence testing (TOST), whereas TTKC-based measurements showed systematic overestimation. Conclusion: The MTKC effectively reduces discrepancies between CT-based planning and radiographic assessment of PTS, providing a more anatomically consistent reference for sagittal alignment planning and potentially improving precision in R-TKA.

ABSTRACT. Introduction: Accurate acetabular cup placement is critical for preventing postoperative dislocation after total hip arthroplasty (THA). Ortoma’s Ortoma Treatment Solution™ (OTS) is a CT-based navigation system incorporating artificial intelligence for preoperative planning and bone segmentation. This study aimed to compare acetabular cup placement accuracy between OTS and a conventional CT-based navigation system (Stryker Navigation; S-Navi) in supine direct anterior approach (DAA) THA. Methods: A retrospective review was conducted of primary THA performed via the supine DAA. The OTS group included 39 patients (40 hips), and the S-Navi group included 118 patients (126 hips). Patients with Crowe type IV deformity or incomplete postoperative evaluation were excluded. Radiographic inclination (RI) and radiographic anteversion (RA) were measured using three-dimensional templating software based on CT data. Absolute errors were calculated by comparing intraoperative planned values with postoperative measurements. Patient demographics and absolute errors in RI and RA were statistically compared between the two groups. Results: There were no significant differences between groups in age, sex, body mass index, or underlying diagnosis. The mean absolute error in RI was 1.9 ± 1.5° in the OTS group and 2.4 ± 1.8° in the S-Navi group. The mean absolute error in RA was 1.7 ± 1.0° and 2.0 ± 1.5°, respectively. No statistically significant differences were observed between the two groups. Conclusions: OTS demonstrated acetabular cup placement accuracy comparable to that of conventional CT-based navigation in supine DAA THA. OTS would be a useful tool for achieving precise cup placement.

ABSTRACT. Bicompartmental knee arthroplasty (BKA) is a bone- and ligament-preserving alternative to total knee arthroplasty (TKA) that may better restore physiological knee function. With advances in computer-assisted surgery, robotic-arm assistance enables precise implant positioning; however, objective functional differences between arthroplasty strategies remain incompletely defined. This study compared gait biomechanics and patient-reported outcome measures (PROMs) before and after robotic-arm assisted BKA or TKA. Between June 2022 and January 2025, forty-eight knees (24 BKA and 24 TKA) were retrospectively matched 1:1 according to age, preoperative coronal alignment, and body mass index. All procedures were performed using robotic-arm assistance, and all TKAs were cruciate-retaining. Three-dimensional gait analysis was conducted preoperatively and at one year postoperatively to evaluate lower-limb kinematics, joint moments, and joint power across the gait cycle. PROMs were collected at baseline, one year, and two years after surgery. Preoperatively, no significant differences were observed between groups in gait biomechanics or PROMs. At one year, PROMs improved significantly from baseline in both groups, with no between-group differences. Gait analysis demonstrated that the BKA group exhibited a higher peak hip extension moment compared with TKA (2.54 ± 0.74 vs 2.09 ± 0.57 Nm/kg, p = 0.025), representing an approximately 20% greater magnitude, accompanied by greater hip extension power generation during early stance (p = 0.018). Other kinematic, kinetic, and spatiotemporal parameters were not significantly different, although cadence tended to be higher after BKA (112.2 vs 105.4 steps/min, p = 0.068). Knee extension moment and power waveforms in the BKA group more closely resembled normal control patterns. These findings suggest that robotic-arm assisted BKA may facilitate a more physiological gait strategy postoperatively despite comparable PROMs.

ABSTRACT. Introduction: In total hip arthroplasty (THA), reconstructing hip parameters such as femoral offset (FO), limb length (LL), and femoral neck anteversion (FNA) significantly affects postoperative alignment and gait. While CAOS has enabled precise 3D preoperative planning and intraoperative control of these parameters, established criteria for planning that account for dynamic lower limb alignment remain insufficient. This study aimed to clarify how 3D changes in hip parameters influence lower limb kinematics and kinetics during gait, providing insights for optimal target setting in CAOS-supported THA. Methods: 32 patients (average age 65.9 years) who underwent THA for hip osteoarthritis were included. Changes in FO, global offset (GO), LL, and FNA were measured using 3D preoperative planning software (ZedHip®). Gait analysis was performed using VICON MX to calculate the range of motion for the hip, knee, and ankle joints, as well as peak hip adduction moment (HAM) and knee adduction moment (KAM). Pearson’s correlation coefficients were used to analyze the relationship between hip parameter changes and gait variables. Results: Correlation analysis revealed that GO change was positively correlated with knee varus/valgus motion (r = 0.45) and KAM (r = 0.45). Increased FO was associated with increased KAM (r = 0.41), while increased LL correlated with decreased hip adduction (r = -0.38). Notably, changes in FNA showed significant correlations with knee varus/valgus (r = -0.50) and ankle eversion (r = 0.43). Discussion and Conclusion: Our findings demonstrate that increased hip offset leads to increased knee varus and ankle inversion, whereas increased FNA is associated with increased knee valgus and ankle eversion.

ABSTRACT. In reverse shoulder arthroplasty (RSA), appropriate implant size selection is essential for achieving a favorable postoperative range of motion (ROM). However, shoulder morphology varies by sex, body size, and population. Japanese shoulders are generally smaller than those of Western populations. This study aimed to identify optimal glenosphere sizes by incorporating a Japanese statistical shape model (SSM) stratified by sex and height into an impingement-free ROM simulation model. Twenty random shapes per group were generated using a Japanese SSM stratified by sex and height (Female-140/150/160; Male-150/160/170) and incorporated into a ROM simulation tool. Impingement-free maximum flexion, abduction, external rotation, and internal rotation were measured. Each ROM was normalized relative to the group maximum under each condition, and the harmonic mean of normalized ROM percentages (H-mean) was calculated. Implant parameters included glenosphere diameter (32–44 mm), lateral offset (0/3/6 mm), and neck-shaft angle (NSA) (135°/145°/155°). The glenosphere diameter at which the H-mean reached 95% was defined as the minimum required glenosphere size. The median minimum required glenosphere size across conditions was 37 mm for Female-140, 38 mm for Female-150, Female-160, Male-150, and Male-160, and 39 mm for Male-170. Across all groups, the minimum required glenosphere size was larger when NSA was larger, and lateralization was lower. This study can provide guidance on the minimum required glenosphere size for Japanese patients of each sex and varying stature to optimize ROM. Additionally, because the effect of glenosphere size on ROM varies depending on glenoid lateralization and humeral NSA, these parameter combinations should be considered.

ABSTRACT. In recent years, there has been an increase in the popularity of research focusing on patient-specific preoperative planning. In accordance with Lewinnek's safe zone model, the calculation of individual safe zones is applicable for numerous parameters. The present study investigated the sensitivity of safe zones for anterior overhang and cup coverage using literature-based threshold variations on a data set of 199 Japanese total hip arthroplasty patients. For anterior overhang, a safe zone was identified for anteversion angles greater than 30° for a threshold of 4 mm and greater than 15° for a threshold of 12 mm. The analysis of cup coverage revealed a greater variation in individual safe zones than for anterior overhang. The superimposed safe zone for the entire cohort revealed an optimal range for anteversion angles between 10° and 30°, with an inclination angle of at least 50°, for a threshold value of 70%. As well as a safe zone, in which almost all implant angle combinations are safe, for a threshold value of 50%. The combination of the two parameters in a combined safe zone exhibited a shape that was analogous to the safe zone for anterior overhang, although the safe areas were more constrained in size. The configuration of the safe zone for both parameters, as well as the combination of both, differs considerably from the shape of the Lewinnek safe zone. This underscores the significance and indispensability of patient-specific preoperative planning in contrast to planning with universally applicable thresholds.

ABSTRACT. Although total knee arthroplasty (TKA) is a well-established procedure to address osteoarthritis, 10-20% of patients remain unsatisfied with the outcome after surgery. Considering patient-individual tibiofemoral kinematics could potentially improve patient satisfaction in the future. Using a state-of-the art navigation system, the motion of the femoral condyles relative to the tibia during a passive flexion motion can be assessed. However, to use this as basis to select the implant type which best supports patient-individual kinematics, robustness of this kinematic acquisition method has to be shown. The objective of this study was hence to assess intra-observer repeatability of passive intra-operative kinematics before knee incision in ten human donor knees. Using the OrthoPilot® navigation system, passive intra-operative kinematics were assessed, whereby an experienced surgeon performed passive flexion from 0° to 90° with two repetitions (set 1 and set 2). The femoral condyle positions were projected onto the tibial plateau at different degrees of flexion to depict the relative motion of femur and tibia as Pinskerova view. In the ten donor knees, different kinematic phenotypes could be observed. Intra-observer repeatability differed slightly between specimens but generally a high agreement was shown between set 1 and set 2. Hence, despite small differences in the single repetitions of passive intra-operative kinematics, intra-observer repeatability was generally found to be good, i.e. the same kinematic phenotype resulted thereof. This underlines the potential of using intra-operative kinematic acquisitions to select the knee insert type which best supports patient-individual kinematics. However, further analyses should be conducted to assess inter-observer reliability.

ABSTRACT. Conventional musculoskeletal shoulder models face limitations due to polynomial approximation of scapular motion and lack of physical scapulothoracic contact modeling. This study presents an upper-limb musculoskeletal model that physically represents thorax-scapula sliding contact and reproduces independent scapular motion through coordinated muscle-ligament action, without polynomial　approximation. The model incorporates 63 muscles and 10 ligaments per arm, with scapular geometry decomposed using Volumetric Hierarchical Approximate Convex Decomposition (V-HACD) for contact detection. Model Predictive Control (MPC) optimizes activations of 27 muscles for reaching simulations. Validation compared model outputs with published in-vivo kinematics measured using bone-pin-based direct methods and electromagnetic tracking. Results reproduced scapulohumeral rhythm, with glenohumeral elevation correlations ≥ 0.96 and scapular upward-rotation correlations of 0.75–0.86 versus experimental data. SA was underestimated below 100° but matched well at ≥ 100°. Compared with the ligament-integrated shoulder-complex model of Sahara et al., our model produced larger SA and closer agreement with experiments, attributed to explicit scapulothoracic contact. Maximum elevation angles (155.5° forward, 154.2° lateral) emerged without artificial constraints and agree with　normative adult active elevation ranges. This work reproduces independent scapular motion without polynomial coupling and enables anatomically plausible evaluation of scapular kinematics for dynamic movement analysis and clinical assessment.

ABSTRACT. Background: Glenosphere diameter and inferior overhang influence bony impingement–free range of motion (ROM) after reverse shoulder arthroplasty (RSA), but criteria for selecting a 32‑mm glenosphere across sexes are unclear. Methods: Fifty-one shoulders (25 females, 26 males) underwent 3D‑CT ROM simulations (diameter 32/36/39 mm; inferior overhang 1.5/3.0/4.5 mm). EF, AA, and rotation arcs to first bony impingement were analyzed with sex‑stratified two‑factor repeated‑measures ANOVA. ROC analysis tested anatomic predictors of 32‑mm preference at overhang 4.5 mm (ties allowed) for rotation and EF endpoints. Results: Females had smaller MAPD than males (24.1±2.1 vs 31.4±2.2 mm) and showed consistent ROM advantages with a 32‑mm glenosphere across planes. In males, optimal diameter was plane‑dependent: EF peaked at 36 mm, AA favored 32 mm, and rotation at 4.5‑mm overhang was similar for 36–39 mm. Rotation‑based 32‑mm preference occurred in 33/51 shoulders (25/25 females, 8/26 males); MAPD predicted preference (AUC 0.886) with cutoff ≤27.7 mm. EF‑based 32‑mm preference occurred in 21/51 shoulders (17/25 females, 4/26 males); MAPD showed moderate discrimination (AUC 0.797; cutoff ≤25.2 mm), and humeral head diameter provided a high‑specificity rule‑in threshold (AUC 0.817; cutoff ≤39.2 mm). Conclusion: Females consistently achieved broader impingement‑free ROM with 32 mm. In males, 36 mm better balanced sagittal/rotational motion, while 32 mm remained favorable for AA and for males with small anatomy identified by MAPD/HHD cutoffs.

ABSTRACT. Deep Learning (DL) has many promising applications in the medical field, from automating diagnostics to automatically segmenting anatomical structures from CT scans with great accuracy. Yet, despite its potential, the path from research to clinical adoption remains fraught with challenges. Regulation and safety are often treated as an afterthought during the development and validation of DL methods, hindering the translation of innovations in clinical practice. To bridge this gap, we propose a hybrid confidence estimation framework for AI-generated 3D segmentations of the shoulder. Our approach enhances the interpretability of predictions, and enables automated quality assessment and targeted review. We assess its effectiveness in quality control, failure detection, and uncertainty localization using a dataset of 674 expert-corrected shoulder segmentation predictions.

ABSTRACT. Anterior iliopsoas impingement (AIPI) is a complication following total hip arthroplasty and is commonly attributed to anterior cup overhang caused by irritation between the iliopsoas tendon and anterior rim of the acetabular component. Although over-sizing, under-anteversion, and over-lateralization of the acetabular component have been associated with AIPI, a patient-specific three-dimensional evaluation of the iliopsoas muscle in relation to anterior cup overhang remains lacking.

In this simulation study, we combined AI-based patient-specific muscle segmentation with a radar chart method to circumferentially evaluate the relationship between the iliopsoas muscle and the acetabular rim. Twenty-seven hips from sixteen patients undergoing surgery for osteonecrosis of the femoral head were analyzed. Acetabular cup placement was simulated using default model, 2-mm lateralized model, and 5°-anteverted model. Cup overhang was assessed at 15 intervals around the acetabular rim using the radar chart method.

The iliopsoas muscle contacted the acetabulum rim in all cases, with contact locations ranging from 2:00 to 4:30. Although the contact area varied among individuals, no hips showed complete absence of iliopsoas-rim contact. Cup overhang occurred more frequently in the lateralized and under-anteverted models than in the default model and contact with the iliopsoas was consistently localized at the 2:30 location.

These findings demonstrate that, to reduce the risk of AIPI, particular attention should be paid to avoiding cup overhang around the 2:30 to 4:00 position during preoperative planning and intraoperative assessment.

ABSTRACT. CT/MRI enable 3D quantification of muscle volume and quality but are impractical for routine sarcopenia screening in primary and community care due to cost, access, and examination burden. We propose a three-view long-length lower-extremity radiograph framework (anteroposterior [AP] plus bilateral lateral views [LAT-L/LAT-R]) to derive 3D-informed muscle indices from routine X-rays. Stitched standing long-length radiographs collected at Inha University Hospital (≈400 exams; AP and ≥1 lateral view) are orientation-normalized, metrically calibrated (DICOM pixel spacing or calibration markers), and contrast-enhanced. U-Net++ segments bone, muscle, and subcutaneous fat; Mask R-CNN detects femur/tibia endpoints to define bone axes and standardized sampling. At quartile positions along each axis, orthogonal tissue thickness is measured on AP and lateral views. Thicknesses are fused under an elliptical approximation to estimate cross-sectional area and integrated along bone length (trapezoidal rule) to compute tissue-specific volumes. When both lateral views are present, lateral estimates are aggregated by the median to reduce rotation bias. Baseline AP-only evidence from prior work showed strong segmentation and DXA validity (IoU 0.93; Dice 0.95; r=0.72 with DXA skeletal muscle index; r=0.66 with leg lean mass; p<0.0001). The long-leg keypoint module achieved 97% keypoint detection accuracy with high geometric fidelity (ROI IoU 0.98; endpoint MAE 0.91 mm; HKA MAE 0.7°; ICC 0.921). These results support the feasibility of standardized, low-burden radiograph-based screening that can integrate alignment/length metrics with 3D-informed body-composition estimation for scalable sarcopenia risk stratification.

ABSTRACT. Precise positioning of the acetabular cup and femoral stem is paramount for restoring hip joint function following total hip arthroplasty (THA). CT-based navigation is widely regarded as the gold standard for ensuring optimal implant alignment. Only a single CT-based navigation platform capable of guiding both cup and stem placement has been available. While this technology enhances surgical precision and reduces complications such as dislocation, its adoption remains limited among hip surgeons due to the labor-intensive nature of preoperative planning. Artificial intelligence (AI), a powerful tool for image analysis, has been increasingly utilized in the medical field and applied to perioperative preparation for arthroplasty. Recently, a novel CT-based navigation system was developed by a Swedish company, utilizing AI to automate bone segmentation and streamline preoperative workflows. This study compared the preoperative planning duration and cup placement accuracy of this AI-driven system against the established industry standard. Our findings indicate that the novel system significantly reduces planning time to under 15 minutes while maintaining surgical accuracy equivalent to existing high-precision navigation technologies.

ABSTRACT. Intraoperative fluoroscopy is widely used in total hip arthroplasty (THA), but it provides only 2D projections and therefore lacks volumetric anatomical information. We propose a generative AI-based pipeline to reconstruct a 3D CT volume of hip bones from a single anteroposterior (AP) intraoperative fluoroscopy image in patients with hip osteoarthritis. The pipeline comprises (1) fluoroscopy-to-DRR-like translation using a segmentation-guided, modified U-GAT-IT, followed by (2) 3D CT volume reconstruction using X2CT-GAN. Retrospective data from two institutions were used. Institution A provided preoperative CT scans from 756 patients with hip osteoarthritis to construct paired DRR–CT data for X2CT-GAN training. Hip bones were extracted using TotalSegmentator; volumes were cropped with an ASIS-based range (+2 cm cranial to −15 cm caudal), split into left/right hips, mirrored to unify sides, and resampled to 256×256×256. Institution B provided 244 cases with preoperative CT and AP fluoroscopy acquired at the start of THA; 194 cases were used for training and 50 for testing. Fluoroscopy images were augmented six-fold by random rotations within ±5°, resulting in 2,820 images. U-GAT-IT was trained on unpaired fluoroscopy images and 702 DRRs, using bone masks estimated by a ResNet34 U-Net. The proposed pipeline successfully generated a 3D CT volume of the hip bones from a single AP intraoperative fluoroscopy image. Quantitative evaluation on the 50-case test set yielded a PSNR of 41.3 dB and an SSIM of 0.876. The U-GAT-IT translation preserved the bony anatomy better than CycleGAN, with fewer shape and position distortions. In future work, we plan to further validate accuracy and explore potential applications such as intraoperative navigation and preoperative planning.

ABSTRACT. Accurate assessment of the sacral canal is useful for preoperative planning in spinal surgery and understanding shape variations with respect to aging, sex, or pathologies. Manual delineation of the sacrum and sacral canal anatomy in computed tomography (CT) images is time-consuming and prone to inter-observer variability due to the complicated structures’ geometry. This study presents a fully automated framework for sacrum and sacral canal segmentation to enable rapid analysis of patient-wise as well as diverse patient populations variations. A deep learning-based pipeline using a 2D hierarchical U-Net-based musculoskeletal segmentation model was trained on a single institution database comprising preoperative CT images of 50 hip osteoarthritis patients. Labels of 25 structures, including the sacrum and sacral canal, were manually annotated by a collaborative team and used for training and validation. The pipeline allows for volumetric segmentation and analysis of densitometric (mean HU) and morphometric (volume/cross-sectional area) features of the target structures. The pipeline was validated in a 5-fold cross-validation manner. The segmentation accuracy was assessed using Dice coefficient (DC), average symmetric surface distance (ASD,mm), average intensity error (AIE,HU) and average volume error (AVE,%). The experiments have shown an accuracy for sacral canal/sacrum of 0.896±0.025/0.969±0.009 (DC), 0.332±0.098/0.211±0.074 mm (ASD), 4.510±3.725/1.854±1.842% (AVE), 3.711±2.845/2.982±2.379 HU (AIE). The tool has shown high accuracy in segmenting the sacrum and sacral canal, which could be feasible for a large-scale database analysis of densitometric and morphometric measurements.

ABSTRACT. Total Ankle Arthroplasties (TAA) procedures have steadily increased over recent years, driven by advances in implant design and a growing demand for motion-preserving alternatives to ankle fusion. Nonetheless, TAA remain a technically challenging procedure. In this context, digital twins offer promising opportunities to enhance computer-assisted surgeries by enabling patient-specific modelling, planning, and intraoperative navigation.

However, creating digital twins requires labelling each picture of a CT scan at the pixel level which is prohibitively time-consuming. In this work, we introduce a digital twin framework dedicated to computer-assisted TAA based on a 2 steps deep-learning pipeline that automatically segments the bones at the talocrural joint (fibula, talus and tibia) from a Computed Tomography (CT) scan. We used a multinational, multi-centre, multi-manufacturer dataset gathering 243 CT scans, including the presence of hardware and bones fusion to train our pipeline. The designed solution relies on a two-stages architecture:

1. A first stage with 2D UNet multiclass segmentation base models. 2. A fusion stage: an other shallow UNet takes the probability maps of the first stage’s models as well as the initial DICOM image to refine the prediction.

Our pipeline achieves median Dice scores between 0.957 on the fibula to 0.978 for the distal tibia segmentations which is consistently comparable or better than the published state-of-the-art, despite being evaluated on a more challenging dataset that includes cases with hardware and bones fusions. Therefore, our model is a promising tool towards more efficient TAA planning and navigation.

ABSTRACT. Skeletal muscle mass (SMM) loss is commonly assessed using the skeletal muscle index (SMI) through bioelectrical impedance analysis (BIA). However, because SMI measurement is not routinely performed in many clinical settings, we aimed to develop and validate a CT-based approach to detect whole-body SMM loss using preoperative lower-limb CT in hip osteoarthritis (HOA) surgery. A total of 177 HOA patients undergoing hip surgery across three institutions (A–C) were retrospectively analyzed. All patients underwent preoperative CT (iliac crest to knee) and BIA. Lower-limb muscles were automatically segmented using the validated Bayesian U-net model. Using the segmentation labels, SMM (g) was calculated for each muscle. The summed SMM proximal to the knee was defined as upper-leg muscle mass (SM_leg) and normalized by height squared to calculate SMI_leg. Whole-body SMM loss was defined as low SMI according to AWGS 2025 BIA-based cutoffs, and the diagnostic performance was evaluated at each institution using multivariable receiver operating characteristic analysis incorporating SMI_leg, age and sex. SM_leg was 6820 ± 1552 g, 6851 ± 1322 g, and 6923 ± 2004 g in institutions A–C, respectively. Whole-body SMM loss was identified in 26.2%, 27.7%, and 37.7% of patients. The areas under curve for diagnosing whole-body SMM loss using SMI_leg were 0.91, 0.88, and 0.89, respectively. Our findings suggest that routine preoperative lower-limb CT (iliac crest to knee) may enable opportunistic screening for whole-body SMM loss in HOA patients without additional tests.

ABSTRACT. Thigh muscle fat infiltration (MFI) is a critical marker of adverse muscle quality linked to physical frailty and metabolic disorders. This study performed a proteome-wide association study using Dixon MRI and Olink proteomics data from 10,895 UK Biobank participants to characterize the circulating protein signatures of myosteatosis. Using a stringent Bonferroni-corrected threshold (P < 1.7 x 10-5), we identified 2,924 proteins significantly associated with MFI. Notably, LEP (Leptin) and FABP4 showed the strongest positive correlations, reflecting ectopic fat accumulation, while CA14 exhibited a prominent negative correlation related to muscle metabolic alterations. Individual scatter plots and correlation matrices confirmed a robust biomarker signature that reflects the biological pathways of muscle deterioration. These findings identify potential therapeutic targets and offer scalable, blood-based surrogates for MRI-defined muscle quality, facilitating high-throughput screening for musculoskeletal decline and personalized interventions in aging populations.

ABSTRACT. Background: Incorrect forearm positioning during imaging significantly biases distal radius morphometric measurements, with potential consequences for diagnosis, surgical planning, and outcome assessment in fracture management. Accurate modeling of forearm pronation-supination (PS) motion is therefore essential to standardize 3D forearm repositioning and enable reliable, reproducible measurements across clinical workflows. Conventional kinematic models describe PS as a rotation around a single fixed longitudinal axis linking the proximal and distal radioulnar joints. However, physiological motion involves complex coupled radius-ulna kinematic patterns, including translation at the distal radioulnar joint and variable rotation axes, challenging the fixed-axis assumption.

Methods: We propose a joint radius–ulna Statistical Pose Model (SPM) using Principal Component Analysis to learn population-level distributions of relative 3D bone configurations from 88 static CT scans spanning a broad PS range (7–177°). The model encodes pose using Lie algebra SE(3) and captures coupled rotation-translation patterns through 12 modes of variation. Validation used 17 optically tracked cadaveric PS sweeps from 3 specimens (mean PS range 115°), comparing SPM performance against a subject-specific fixed-axis baseline.

Results: The SPM reproduced PS trajectories with submillimetric accuracy (mean mesh RMSE 0.12 mm, translational error 0.10 mm, rotational error 0.05°), substantially outperforming the fixed-axis baseline (RMSE 2.7 mm, translational error 2.3 mm, rotational error 5.6°).

Conclusions: These results demonstrate that data-driven pose space models can effectively capture the complex coupled kinematics of forearm PS motion and can potentially be used for the standardization of 3D forearm repositioning to reduce pose-related measurement bias in surgical planning, fracture assessment, and longitudinal follow-up.

ABSTRACT. Hallux valgus (HV) is a triplanar deformity of the forefoot, which makes freehanded realignment of the first metatarsal bone during surgery challenging using only 2D radiographs. This paper presents a novel semi-automatic 3D virtual surgical planning (VSP) and a novel design of patient-specific guide (PSG) for a single patient with HV who received a Lapidus surgery. The semi-automatic VSP reduced planning time to one fifth (16 minutes), thereby improving feasibility for clinical implementation. The PSG design allowed for improved foot alignment after surgery: the intermetatarsal angle changed from 20.7° to 16.4°, the rotation of the first metatarsal changed from 7.2° to 0.2° supination, and Meary’s angle changed from 20.9 to 15.5°. The proposed method has potential for further optimization and clinical implementation of VSP- and PSG-guided Lapidus surgery.

ABSTRACT. Custom 3D-printed braces offer improved comfort and comparable effectiveness to traditional splints. However, their clinical adoption is hindered by the requirement for patient-specific digital models, typically obtained through 3D scanning and the registration of a statistical shape model (SSM). Such a standard workflow is, however, time-consuming, operator-dependent, and requires specialized hardware. Furthermore, traditional registration techniques frequently struggle with anatomically complex regions like the fingers and thumb, where linear assumptions fail to capture nonlinear deformations and rotational variability.

This paper presents a novel data-driven framework that predicts 3D forearm geometry from a small set of simple 1D anthropometric measurements, effectively eliminating the need for 3D scanning. To build a robust SSM, dense correspondences were established throughout a population using an iterative registration process that combined landmark, ICP, and elastic registration. To decouple anatomical shape from pose-related noise, a dual-normalization strategy was implemented: Linear Blend Skinning for skeletal alignment and a statistical correction for soft tissue deformation. Multivariate linear regression was then used to link 1D measurement features with principal component weights, allowing the direct prediction of new forearm shapes.

The results demonstrate that the framework can predict forearm shapes with an average measurement error below 1%. Although the thumb region remains challenging due to its high degrees of freedom, the proposed method significantly reduces the dependency on expensive hardware and manual intervention. This measurement-based approach offers a scalable and accessible alternative for the virtual design of well-fitting functional braces in clinical settings.

ABSTRACT. In knee arthroscopy surgery, accurate registration between preoperative and intraoperative anatomy is a critical step for patient-specific navigation. Achieving an accurate registration requires a reliable 3D reconstruction of the joint during surgery. Preoperative 3D models can be obtained from patient imaging through segmentation and reconstruction, but generating an intraoperative 3D representation remains particularly challenging. Arthroscopic imaging suffers from a limited field of view, low surface texture, and strong specular reflections, which make conventional feature-based 3D reconstruction methods unreliable. In this work, we investigate the application of MIS-NeRF (Minimally-Invasive Surgery Neural Radiance Fields) for reconstructing intraoperative knee 3D models from monocular arthroscopic images. The approach is evaluated on five simulated arthroscopic acquisitions representing five patient-specific knee 3D models. Both qualitative and quantitative results are presented to assess the reconstruction quality. The reconstructed knee 3D models were evaluated through their rendered images, achieving PSNR(Peak Signal-to-Noise Ratio) of 23.29 ± 3.82, SSIM(Structural Similarity Index) of 0.86 ± 0.07 and LPIPS(Learned Perceptual Image Patch Similarity) of 0.27 ± 0.13. These preliminary results suggest the feasibility of NeRF-based reconstruction in the challenging context of arthroscopy and may represent a promising step toward accurate in-silico preoperative-to-intraoperative 3D registration for computer-assisted orthopedic surgery. Further, validation on real arthroscopic data will be necessary to assess clinical applicability.

ABSTRACT. Developmental Dysplasia of the Hip (DDH) is commonly screened using 2D ultrasound (US); however, diagnostic metrics such as alpha (α) angle and femoral head coverage (FHC) show high variability due to variations in operator experience, probe placement, and patient position. Although screening with 3D ultrasound (3D US) is more reliable, its high cost and limited availability restrict widespread use. To bridge this gap, we have developed an Optically Tracked Ultrasound (OTUS) system that reconstructs 3D volumes from a tracked sweep of 2D US images. An initial OTUS prototype has demonstrated feasibility in a clinical pilot study, but faced barriers to clinical integration due, in part, to reliance on a non-planar marker system with tight orientation requirements. To address this limitation, we evaluated the accuracy and usability of a new planar marker design. We conducted a clinical pilot study, acquiring OTUS scans from 15 infants (mean age 7.34 weeks, 4.42-12 weeks). OTUS scans were reconstructed into 3D volumes, and relevant anatomy was segmented using 3D U-Net models trained on expert-labeled 3D US data. We extracted three-dimensional equivalents of α angle and FHC and compared them across repeated OTUS scans to assess repeatability and against 3D US to assess agreement with the clinical gold standard. The updated system resulted in a higher proportion of high-quality scans than the previous marker design, but the metrics also demonstrated relatively poor repeatability and agreement with 3D US. Scan quality was not strongly correlated with metric plausibility or repeatability, indicating that improvements in segmentation and metric extraction are required before OTUS can support reliable DDH screening.

ABSTRACT. This study aimed to investigate the relationship between oral function and objectively measured physical activity in patients undergoing total hip arthroplasty (THA). This study included 100 female patients who underwent primary THA. At admission, seven oral functional conditions were assessed, and oral hypofunction was defined as impairment in three or more domains. Physical activity was objectively measured using a wearable activity monitor. Average daily exercise volume (Ex) was calculated preoperatively and at three months postoperatively as the product of metabolic equivalent intensity and time spent in activities ≥3.0 METs. Multivariable linear regression analysis was performed with postoperative Ex at three months as the dependent variable, adjusting for age, body mass index, preoperative Ex, and oral function variables significant in univariable analyses. Oral hypofunction was identified in 18 patients (18%). Compared with patients without oral hypofunction, those with hypofunction were older and showed significantly lower postoperative Ex at three months (both p < 0.01). In univariable analyses, poor oral hygiene, decreased tongue pressure, and decreased tongue–lip motor function were significantly associated with postoperative Ex. In the multivariable model, age was not independently associated with postoperative Ex, whereas preoperative Ex (β = 0.616, p < 0.01) and tongue pressure (β = 0.239, p = 0.022) remained significant independent correlates. These findings indicate that oral hypofunction is present in approximately 20% of patients undergoing THA, and that tongue pressure is independently associated with postoperative physical activity. Oral function, particularly tongue muscle strength, may influence postoperative recovery and warrants consideration in perioperative care.

ABSTRACT. This study investigated whether component positioning influences patient-reported outcome measures at 2 years after handheld robotic-assisted bi-cruciate stabilized TKA (BCS-TKA). A total of 99 patients (108 knees) who underwent handheld robotic-assisted BCS-TKA and were followed for at least 2 years were included. All procedures were performed using the Journey II prosthesis. Postoperative coronal, sagittal, and axial alignment of the femoral and tibial components was evaluated using three-dimensional computed tomography. Postoperative rotational mismatch between the femoral and tibial components was calculated. Clinical outcomes were assessed preoperatively and 2 years postoperatively using the 2011 new Knee Society Score (2011 KSS). Multivariate analyses were performed to identify factors independently associated with each 2011 KSS domain. Accurate implant positioning was achieved in the coronal and sagittal planes. The mean postoperative femorotibial rotational mismatch was -3.7 ± 4.9°. All domains of the 2011 KSS, except for patient expectations, showed significant improvement 2 years postoperatively compared with preoperative values. Multivariate analysis identified postoperative rotational mismatch between the femoral and tibial components as an independent factor associated with the symptoms and patient satisfaction domains of the 2011 KSS. Postoperative rotational mismatch between the femoral and tibial components affected patient-reported outcomes at 2 years after handheld robotic-assisted BCS-TKA. Precise control of femorotibial rotational alignment remains crucial for optimizing clinical outcomes, even when coronal and sagittal alignment is accurately achieved.

ABSTRACT. Purpose: Using computer-assisted equipment (CAE) in total hip arthroplasty (THA) has facilitated improved accuracy of acetabular cup placement, but the impact of aligning the femoral stem on post-operative patient satisfaction remains unclear. In this study, we identified factors related to femoral stem alignment that are associated with post-operative patient satisfaction after THA. Methods: This retrospective study included patients who underwent THA using CAE for acetabular cup placement between May 2021 and August 2023. Femoral stem parameters—including stem anteversion, change in stem anteversion, femoral offset, and coronal and sagittal stem angles—were evaluated using computed tomography scans and radiographs. Patients were categorized into normal and non-normal alignment groups based on defined ranges: stem anteversion 10°–40°, change in anteversion ±5°, offset difference ±3mm, and coronal/sagittal angles ±3°. Patient satisfaction was assessed using a visual analogue scale (VAS) and clinical outcomes were evaluated using the Japanese Orthopaedic Association Hip Disease Evaluation Questionnaire (JHEQ) at 1 year post-operatively. Results: Of the 200 patients included (205 hips; mean age, 67.3 ± 9.8 years), those in the normal alignment group demonstrated significantly higher JHEQ total and movement scores than non-normal group. Through multivariate logistic regression analysis, we identified change in stem anteversion as an independent factor associated with high post-operative patient satisfaction. Conclusion: Accurate femoral stem alignment, particularly that minimizing changes in stem anteversion from pre- to post-operative measurements, was significantly associated with improved post-operative patient satisfaction after THA. These findings confirm the importance of precise positioning of the femoral stem in CAE-assisted THA.

ABSTRACT. Correct implant positioning in total hip arthroplasty (THA) for severe developmental dysplasia of the hip (DDH) is technically demanding due to anatomical deformities. This study evaluated the Ortoma Treatment Solution (OTS), a novel AI-based CT navigation system, focusing on its efficiency and accuracy using 3D-printed Crowe type II–IV models. The AI-driven "OTS Hip Plan" software completed the segmentation of the pelvis and femur in less than one minute. The accuracy was exceptionally high, with a Dice Similarity Coefficient of 0.99 ± 0.01 across 30 hips (10 each for Crowe II, III, and IV). For registration, we compared methods including points on the true/false acetabulum and the iliac crest . While matching using only the true acetabulum often failed in severe cases (success rate 20% in Crowe IV), adding two points on the iliac crest achieved a 100% success rate for all types. Specifically, in Crowe IV cases, iliac crest points significantly reduced the anterior superior iliac spine (ASIS) deviation from 3.29 ± 2.56 mm to 0.28 ± 0.49 mm (p < 0.05). In conclusion, the AI-based system offers rapid, high-precision segmentation. Furthermore, acquiring additional registration points on the iliac crest—accessible through pin cite incisions—is essential for ensuring matching success and spatial accuracy in complex DDH reconstructions.

ABSTRACT. Aims To compare clinical and technical outcomes when transitioning from a femur first measured resection (MR) workflow to a tibia first gap balancing (GB) workflow in TKA, and to assess how early versus late GB experience influences functional results and intraoperative parameters.

Methods A retrospective series of 225 consecutive primary TKAs performed by one senior surgeon using the same CAOS platform was reviewed. Patients were allocated to MR (n=75), early GB (n=75), and late GB (n=75) groups. MR relied on landmark based femoral resections, whereas GB used a tibia first technique with a force controlled distractor applying 90 N per compartment to acquire dynamic medial and lateral gap profiles. KOOS Jr. scores were collected preoperatively and at one year. Intraoperative variables included femoral and tibial alignment, tibial slope, insert thickness, and planned versus final gaps. Statistical analyses included Welch t tests, ANOVA with Tukey post hoc tests, Mann–Whitney U, and Fligner–Killeen variance tests (α = 0.05).

Results Preoperative KOOS Jr. scores were similar. At one year, all groups improved, with the late GB cohort showing the greatest gain (34.1 ± 20.9), significantly exceeding MR and early GB. Early GB outcomes mirrored MR. Technically, GB adoption increased variability in femoral resections, with progressive increases in femoral flexion and tibial slope, and tighter, more symmetric gaps observed in late GB cases.

Conclusion Refinement of the GB workflow produced superior functional outcomes and more consistent symmetric gaps. Early GB matched MR results, while late GB experience yielded clinically meaningful improvement, reflecting increased confidence in soft tissue guided planning.

ABSTRACT. Patient-specific Ti-6Al-4V lattice implants represent a significant advancement for reconstructing weight-bearing segmental defects, yet robust clinical data from large cohorts remain limited. This retrospective single-center study evaluates the clinical, oncological, and functional outcomes of 29 patients treated between 2016 and 2024 for long-bone tumor resections (n=28) or post-traumatic nonunion (n=1) of the femur or tibia. The reconstruction workflow utilized intraoperative 3D-printed patient-specific instruments (PSIs) for guided osteotomies, followed by the insertion of customized Ti-6Al-4V lattice implants designed to promote osseointegration. The cohort had a mean age of 26.2 years, with a median follow-up of 33 months. Results demonstrated a high limb-salvage rate of 93.1% (27/29 patients). Early surgical complications occurred in 10.3% of cases (one hematoma, two deep infections), while late complications also affected 10.3% (one infection, one mechanical failure, and one subtalar fusion). Oncological control was stable, with 23 patients showing no evidence of disease at the latest follow-up and a local recurrence rate of 13.8%. Functional recovery was high, with a median MSTS score of 24. This study indicates that patient-specific Ti-6Al-4V lattice implants offer a reliable and effective solution for critical-sized, weight-bearing defects. The approach achieves accurate anatomic repair and high functional outcomes without compromising surgical or oncological standards, supporting its use as a practical alternative to traditional biological grafting in complex reconstructions.

ABSTRACT. Preoperative three-dimensional (3D) planning has become standard practice in complex orthopaedic procedures, including bone tumor resections. However, quantitative evaluation of surgical accuracy remains challenging due to the lack of standardized measurement methods, resulting in substantial heterogeneity across studies and limiting meaningful comparison and feedback for surgical improvement.

This study aimed to identify a suitable standardized approach for evaluating procedural accuracy by comparing achieved cutting planes with preoperative plans, with a focus on complex orthopaedic and oncological procedures.

A single-centre retrospective cohort study included 10 patients (7 oncological, 3 non-oncological) treated between 2021 and 2024, comprising 28 cutting planes in patients and 17 in allografts. All cases involved 3D preoperative planning and patient-specific instruments. Achieved planes were determined using mesh and point cloud–based approaches, and procedural accuracy was assessed using multiple linear (Point-to-Point, COG-to-COG, COG-to-plane), angular (normal vector, pitch and roll), flatness (ISO-1101 corrected and non-corrected), and surgical margin metrics.

Linear outcome measures showed no significant differences between the mesh and point cloud methods (p > 0.05). However, the COG-to-COG approach differed significantly from both Point-to-Point and COG-to-plane, with differences up to 5.01 mm (p < 0.001), indicating increased sensitivity to plane translation and convex hull inaccuracies. Using the point cloud Point-to-Point approach, a mean procedural accuracy of −0.77 ± 2.03 mm was observed, suggesting slightly deeper resections than planned. Angular outcomes did not differ significantly between methods. Flatness and surgical margin analyses revealed small but significant method-dependent differences, particularly in allograft cases.

Measurement methodology substantially influences post-operative accuracy assessment and can introduce errors comparable to surgical margins. The point cloud method represents a robust alternative to mesh-based analysis and provides a promising basis for standardized evaluation of procedural accuracy in complex orthopaedic surgery.A multicenter study is beeing performed and can be presented at CAOS