previous day
all days

View: session overviewtalk overview

09:00-10:00 Session 16: Normal Tissues and Tumors
Investigating the spatiotemporal neuroinflammatory response of half-brain irradiation in a murine model using 18F-FEPPA PET

ABSTRACT. Introduction Brain metastases have been effectively treated with stereotactic radiosurgery (SRS) delivered to the visible growths followed b whole brain radiotherapy (WBRT) for microscopic disease. SRS alone is the preferred treatment despite high recurrence rates, as conventional WBRT is associated with increased cognitive decline. With improved systemic treatments breast cancer patients are living longer which challenges the decision to withhold WBRT. Cognitive decline has been linked to chronic inflammation; radiation induces inflammation via glial cell (microglia, astrocytes) activation. When glial cells are activated, translocator proteins (TSPO) on the mitochondria upregulate and promote inflammation. Glial activation has been assessed in neurological disease using a 18F-FEPPA ( ligand with high affinity for TSPO with positron emission tomography (PET). The aim of this study is to investigate the neuroinflammatory response of half brain irradiation with glial activation imaging using 18F-FEPPA PET. Materials & Methods To evaluate radiation induced glial activation, half brain irradiation was performed on non-tumor bearing immunocompetent mice (BALB/c) using a micro-CT/RT system with sham (n = 9), 4 Gy (n = 9) and 12 Gy (n = 9) in one fraction. Dynamic 18F-FEPPA PET was acquired for 90 minutes at 48 hours, 2 weeks and 4 weeks after irradiation to quantify level and duration of glial activation in a cohort of the mice. PET kinetic analysis was completed for time activity curves (TACs) and volume of distribution (VT). Immunohistochemistry identifying glial activation was completed with stains for translocator proteins (TSPO) which is the specific ligand for 18F-FEPPA. Results 18F-FEPPA-PET dynamic scans were acquired for n = 8 mice: sham at 48 hours (n=1) and 2 weeks (n =1), 4 Gy at 48 hours (n=1), 2 weeks (n=1) and 4 weeks (n=1) and 12 Gy at 48 hours (n=1), 2 weeks (n=1) and 4 weeks (n=1) with n =2 additional mice at each dose and timepoint for histological analysis. Unirradiated and irradiated hemispheres of the brain for all mice imaged with 18F-FEPPA-PET showed similar radiotracer uptake; suggesting that partial brain irradiation triggers global inflammation in the brain at 48 hours, 2 weeks and 4 weeks. Histological analysis of TSPO also showed global inflammation with half brain irradiation, with similar signal between the hemispheres. However, differences were present in TSPO stain signal corresponding to the different areas of the brain. Conclusion This study will map the spatiotemporal dose-response of the brain when partially irradiated with different levels of dose. It has provided data to understand how the brain, specifically glial cells respond to irradiation and the resulting duration and amount of inflammation. Glial activation immunofluorescent stain with Iba1, further PET and histology data collection and analysis is on-going. Following this, half brain irradiation will be investigated in a breast cancer brain metastasis model to provide a comprehensive understanding of radiation and subsequent glial activation in the context of cancer to discover an alternative radiation treatment that optimizes breast cancer brain metastasis control with reduced cognitive side effects.

Age-induced and ionizing radiation-triggered changes of glial makers in the brain of rats

ABSTRACT. Introduction: In the radiotherapy of patients with metastatic brain and head and neck tumors, fractionated irradiation is preferentially used in cranial indications. Modern irradiation techniques increase the irradiation process's conformity, leading to longer median patient survival. However, the use of therapeutic doses of radiation leads to severe changes. Aging is a complex process that differentially impacts the brain's cognitive, sensory, neuronal, and molecular processes. Advanced age is an important modifier of the radiation effect on the brain's neurogenic population. In this study, we investigated the effect of gradually increasing age and the so-called bystander effect of spinal cord irradiation on glial markers in surrounding structures, e.g., the brain. Materials & Methods: The study enrolled adult Wistar male rats (n=10) monitored in the aging process (at 3, 6, and 9 months), which we further examined after irradiation with age-matched control animals (n=10). Brain metabolites were measured using in vivo proton magnetic resonance spectroscopy (1H MRS) in the hippocampus, corpus striatum, and olfactory bulb (OB). The 12-month-old rats received fractionated spinal cord irradiation with a total dose of 24 Gy administered in 3 fractions (8 Gy per fraction) once a week on the same day for 3 consecutive weeks. Subsequently, we measured metabolites twice, 48 hours (12-month-old rats) and 2 months (14-month-old rats) after irradiation. Histopathological changes were investigated in two brain neurogenic regions: the hippocampal dentate gyrus (DG) and the rostral migratory stream (RMS). Results: Changes in rats at different ages (3, 6, and 9 months old) measured by 1H MRS showed an age-dependent increase in the total choline to creatine (tCho/tCr) and myoinositol to tCr (mIns/tCr) in the OB and hippocampus. 1H MRS measurement performed 48 hours after irradiation showed an increase of mIns/tCr in the corpus striatum and BO in 12-month-old rats, which remained elevated 2 months after irradiation. Using image analysis of immunofluorescent stained brain sections, we identified a nonsignificant increase of astrocytes and microglia in the RMS and hippocampal DG in irradiated 14-month-old rats. Conclusion: The study showed age-related metabolic changes in the evaluated brain regions. The radiation-induced bystander effect influenced the concentration of metabolites in selected brain regions more than the distribution of glial cells. The influence of advanced age as a possible modifier of radiation-induced changes cannot be ruled out.

Development of Clinically Relevant Radiotherapy Methods for Measuring Survival Benefits and Late Effects of Radiosensitising Drugs in Mouse Models of Medulloblastoma

ABSTRACT. Introduction Medulloblastoma is the most common malignant brain tumour in children, and radiotherapy is an important part of treatment. However, many preclinical studies examining radiosensitising drugs are conducted in such a way that they cannot reliably be translated into clinical practice. Additionally, medulloblastoma patients treated with radiotherapy commonly develop long-term late side effects including cognitive deficits. Many preclinical models use a single, high dose of radiotherapy, which does not mimic the fractionated schedule children receive clinically. Therefore, we aimed to develop clinically accurate, replicable radiotherapy protocols for use in a preclinical research pipeline investigating new therapies for medulloblastoma. Furthermore, we examined whether radiotherapy induced measurable long-term neurostructural defects in mice, to determine the potential impacts on the developing brain.

Methods Four irradiation techniques were designed and evaluated using orthotopic xenograft mouse models of medulloblastoma: focal (tumor site), square beam (targeting the brain surrounding the implantation site), whole brain, and craniospinal (CSI - targeting the entire brain and spinal cord). NOD/Rag1KO adult mice were implanted with medulloblastoma cells, and treated with 18 Gy fractionated across 9x 2 Gy doses of x-ray irradiation using an X-RAD SmART 225Cx. Tumour-free survival was measured, and the location of tumour regrowth analysed. To examine late effects, juvenile non-tumour bearing mice were treated at postnatal day 16 with a single dose of 8 Gy or 18 Gy fractionated across 9x 2 Gy doses, both determined to be biophysically equivalent according to the linear quadratic model. Control mice received single or 9x sham treatments including anaesthesia and CT scan(s). Animals were aged to young adulthood (63 days), at which timepoint ex vivo anatomical MRI scans and histology were performed.

Results As expected, focal radiotherapy was insufficient for medulloblastoma control. Square beam irradiation extended survival, however tumours regrew rapidly, especially in areas of the brain outside the treatment field. Whole brain irradiation significantly extended survival, however mice developed metastases along the spinal cord. CSI increased survival in all models of medulloblastoma tested, and relapses occurred locally. The total dose of radiotherapy given varied between mice, as 18 Gy was curative in many patient derived xenograft models. Irradiation of the developing murine brain had significant long-term effects. A single 8Gy radiation dose significantly decreased total brain volume (-7.9%), particularly in the olfactory bulbs (-17.3%), hippocampus (-7.3%), corpus callosum (-9.0%) and motor cortex (-8.9%). In contrast, mice treated with fractionated radiotherapy (18 Gy) showed no decrease in total brain volume, although the olfactory bulbs were reduced (-11.1%).

Conclusion We comprehensively compared multiple radiotherapy treatment plans to define an optimal method for treating medulloblastoma mouse models. To best mimic clinical practice and patterns of disease relapse, our data indicate that multi-fraction CSI is the recommended best practice for translational preclinical research for medulloblastoma. Moreover, fractionated radiotherapy in non-tumour bearing juvenile mice replicated neurostructural late effects experienced by children. Assessment of long-term effects of new therapies in this manner will be a beneficial addition to a comprehensive preclinical testing pipeline to evaluate the safety of novel chemo/radiotherapy treatments for children with medulloblastoma.

Development of preclinical cone-beam CT imaging biomarkers to characterise tumour heterogeneity, radiation response and the early detection of metastasis

ABSTRACT. Introduction: Radiomics is a rapidly evolving area that uses medical images to develop prognostic and predictive imaging biomarkers. Preclinical models play a critical role in oncology research that aim to recapitulate the underlying biological of features and overcome limitations relating to heterogeneity and reproducibility in patients. In this study, we aimed to determine tumour-specific radiomics signatures using genetically defined preclinical mouse models. Materials/Methods: Radiomics features were extracted from cone-beam computed tomography (CBCT) scans obtained from multiple tumour models (subcutaneous and metastatic). Tumour heterogeneity was assessed using different cell lines derived from Kras mutant/p53 null cell lines derived in-house from KrasLSL-G12D/+; p53Fx/Fx (KP) mice. Longitudinal changes in radiomics features following radiation exposure were assessed in subcutaneous xenografts using MC38 colorectal cells irradiated with a single fraction of 16 Gy or four fractions of 4 Gy. A predictive model for the early detection of lung metastasis was developed following tail vein injection of KP cells. Radiomics analysis was performed using a recently developed in house pipeline using ITK-SNAP and Pyradiomics. Original and wavelet features (n = 842) were analysed, and principal component analysis (PCA) was used to identify feature clusters. Prediction models were developed through deep learning using Random Forest Classifier (RFC) models. Scans were split into training (66%) and validation (33%) cohorts. Model robustness was tested through 3-fold-cross-validation and recurring features identified as robust for model prediction. Results: The heterogeneity of different Kras mutant tumour models was achieved using a RFC model trained using 3 radiomics features (AUC 0.95). Early and late radiation response was compared in MC38 tumours with non-irradiated controls. Tumours irradiated with 16 Gy could be differentiated using CBCT imaging at 24 hours post treatment using a RFC based on 3 radiomics features (AUC 0.92), representative of initial response, and 7 features when tumours reached endpoint (AUC 0.88), representative of late response. We also established a radiomics signature for the early detection of metastatic tumours in the lungs of mice using 4 features (AUC 0.78). Conclusions: This work demonstrates the feasibility of radiomics analysis to characterise tumour heterogeneity, detect changes in the radiation response of tumours and the detection of tumour metastasis. These preclinical models will support the development and translation of imaging biomarkers as a cost-effective surrogate to tumour biopsies for precision medicine applications in radiation oncology.

Dose-dependent changes in cardiac function, deformation and remodelling in a preclinical model of heart base irradiation

ABSTRACT. Introduction: Radiation-induced cardiac toxicity (RICT) describes a range of adverse conditions that can manifest years or even decades after treatment. Classically, the heart has been assumed to be a uniformly radiosensitive organ and most preclinical studies have used whole heart or whole thorax irradiations for experimental purposes. Using small animal image-guided radiotherapy, we have recently demonstrated variations in cardiac function resulting from regional targeting and identified the base of the heart as a radiosensitive subvolume. In this study, we apply our model of cardiac base irradiation to assess the dose-dependent changes in cardiac function, deformation and remodelling Also, we aimed to demonstrate the predictive potential of global and segmental longitudinal strain as early biomarkers for late occurring RICT. Materials & Methods: Female C57BL/6J mice were irradiated under image guidance using a small animal radiation research platform (SARRP) with a single-fraction of 16 Gy or 20 Gy or with 3 consecutive fractions of 8.66 Gy targeting the heart base, using an anterior-posterior field arrangement. The respective biologically equivalent doses (BEDs) were 101.3 Gy, 153.3 Gy and 101.3 Gy using an α/β value for the heart 3 Gy. Longitudinal transthoracic echocardiography (TTE) was performed at baseline and at 10-week intervals up to 50 weeks after irradiation. This was coupled with two-dimensional speckle tracking (2D-STE) to analyse early changes in global longitudinal strain (GLS). Results: Irradiation of the heart base leads to BED-dependent changes in systolic and diastolic function 50 weeks post-irradiation. Systolic function measured by EF was significantly decreased in all groups compared to baseline and age-matched control animals with the largest decrease observed following 20 Gy irradiation (baseline = 71.6 ± 3.2%, 20 Gy = 64.79 ± 4.3%). Similarly, FS was significantly decreased in all groups compared to baseline with the largest decreases observed following treatment with 16 Gy and 20 Gy (baseline = 30.9 ± 4.2%, 16 Gy = 25.38 ± 3.18%, 3 x 8.6 Gy = 27.58 ± 2.36%, 20 Gy = 25.45 ± 2.15%). Diastolic function measured by the E/A ratio was decreased in all groups with the most significant changes observed following treatment with 20 Gy (baseline = 1.28 ± 0.08, 20 Gy = 1.10 ± 0.03). BED-independent increases were observed in the left ventricle (LV) mass and volume, and myocardial fibrosis. GLS was significantly decreased in a BED-dependent manner for all irradiated animals, as early as 10 weeks after irradiation from -18.6% to -14.8% for 20 Gy and -14.5% for 16 Gy. Conclusions: Our model of cardiac base irradiation accurately captures clinical observations of the heart base a radiosensitive subvolume with loss of cardiac function dependent on BED. Importantly, we show that GLS can accurately detect BED-dependent radiation-induced changes in cardiac strain at 10 weeks after treatment that are indicative of late functional loss at 50 weeks. These data clearly demonstrate the potential use of GLS as an imaging biomarker of early radiation-induced cardiac dysfunction before detectable loss of systolic and diastolic function.

10:00-10:30Coffee Break & Exhibition
10:30-11:43 Session 17: Dosimetry&Imaging&Technology
(Invited) Ionoacoustics-based dose monitoring in pre-clinical proton beam therapy

ABSTRACT. Introduction The spatially and temporally confined energy deposition of pulsed proton beams results in thermoacoustic wave emissions with characteristics matching those of the underlying dose distribution and specific medium properties (i.e., density and material-specific energy-to-pressure efficiency). Therefore, detecting the acoustic signals generated during proton therapy with pulsed beams allows inferring information on the incident ionizing source either to locate the dose maximum in vivo (e.g., Bragg peak localization) or to reconstruct its distribution. Ionizing radiation-induced acoustics (ionoacoustics) has gained significant interest recently due to its relative simplicity, cost-effectiveness, and potential for near real-time dose monitoring. Furthermore, an ultrasound/ionoacoustics bi-modal imager, ideally relying on the same detection device, could provide dosimetry information promptly registered to the anatomy for treatment adaptation in image-guided therapy. However, advancement in ionoacoustics-based dosimetry is challenged by the inherently low frequency and weak pressures resulting from the energy deposited by pulsed beams available at clinical facilities (typically a few mPa at a frequency around 50 kHz). This is several orders of magnitude lower than conventional ultrasound imaging, requiring the development of dedicated sensing technologies. This work investigates suitable technological solutions enabling ionoacoustics-based in vivo range verification on a prototype irradiation platform (SIRMIO platform) during small animal irradiation at clinical synchrocyclotrons.

Material & Methods State-of-the-art acoustic detectors have been extensively investigated during several experimental campaigns with pre-clinical monoenergetic pulsed proton beams at the MLL tandem accelerator (Munich, Germany) and clinical beams from the synchrocyclotron at the Centre Antoine Lacassagne (Nice, France). The impact of the sensing system characteristics and medium heterogeneities on the Bragg peak localization was quantified in-silico and confirmed experimentally for phantoms of increasing complexity using different reconstruction strategies (i.e., 3D multilateration and time-reversal reconstruction). The accuracy and precision of the dose maximum localization were assessed in silico for a representative treatment plan as expected to be delivered in a mouse irradiated on the SIRMIO platform with pulsed proton beams.

Results and conclusion Capacitive Micromachined Ultrasonic Transducer technology (CMUT, here developed at the University Roma Tre, Italy), enabling detection of the kHz ionoacoustics signals and MHz ultrasound imaging, was identified as a promising candidate for ultrasound/ionoacoustics bi-modal system development. Our results show that the accuracy of both multilateration and time-reversal-based dose reconstruction strongly depends on the sensor array design (i.e., sensor position relative to the Bragg peak), suggesting that the detector arrangement should optimally be adapted on a case-by-case basis. The simulation study in mice demonstrated that each pristine Bragg peak of a treatment plan can be located with a millimeter accuracy and a sub-millimeter precision.

Acknowledgment The financial support from European Research Council (SIRMIO, 725539), the Centre for Advanced Laser Applications, BayFrance and LMUexcellent.

New X-Ray Irradiators for Research: Experience with the USA Cesium Irradiator Replacement Project

ABSTRACT. Wake Forest University School of Medicine (WFUSM) recently obtained two state-of-the-art x-ray irradiators for radiation research. These x-ray instruments are replacing two legacy cesium irradiators. The acquisition of the new x-ray devices was facilitated by the Cesium Irradiator Replacement Project (CIRP), a National Nuclear Security Administration (NNSA), Office of Radiological Security (ORS) initiative that seeks to reduce the risk of high activity radioactive sources, to include Cs-137 and Co-60, by encouraging transition from those sources to alternates such as x-ray technology. The CIRP incentivizes the replacement of a designated cesium unit by an appropriate x-ray device through 1) subsidizing the purchase of the x-ray device, and subsequently 2) providing no-cost removal of Cs-137 irradiators which are permanently disposed of at a federal repository. This presentation reviews WFUSM’s favorable participation in CIRP for a planned transition from cesium to x-ray research irradiators for radiation research, including details of the CIRP initiative, logistics and milestones of participation, relevant stakeholders and communications, and considerations and current status for moving from cesium to x-ray irradiations for cells and small animals.

Towards standardized commissioning data for the Small Animal Radiological Research Platform (SARRP)

ABSTRACT. Introduction: Xstrahl’s Small Animal Radiological Research Platform (SARRP) is one of the most widely used systems for preclinical cancer treatment research worldwide. An essential step in the setup and ongoing QA of a SARRP is the commissioning of the treatment planning system (TPS). During commissioning, computed depth-doses are fitted to measurements to determine collimator-specific scaling, or output factors (OFs), which are subsequently applied to all calculated doses. Ideally, film measurements from each individual machine are used for commissioning. However, it is often necessary to use existing commissioning measurements from machines of the same build type. Until recently, we have used an “exemplar” data set measured on a single machine and scaled this data to the output of the customer’s machine. While this has been shown to provide reasonably accurate OFs, it does not take advantage of our extensive database of commissioning measurements, nor does it give us any indication of how an individual set compares to other measurements. In this work, we present an overview of our commissioning database, showing the degree of variation in measurements for a given collimator/build and comparing measurements to Monte Carlo (MC) simulated depth-dose. Several options for generating commissioning data are given, including using an “exemplar” set, an average set, and MC data only. We also examine OFs of the variable collimator (MVC) to determine the significance of the variation with field size. Materials & Methods: Analysis of the commissioning database is facilitated by a set of Python scripts. These read in the commissioning depth-dose data (in JSON format), scale the data to the output of the machine being commissioned, shift the data to a set of common SSDs, and interpolate it to a set of common depths. For a given collimator/build the script plots the depth-dose curves for each SSD, allowing the user to visualize depth-dose data for individual machines in the database, averages over multiple machines, and MC-calculated depth-dose. As a quantitative indicator of data spread, the scripts also calculate the root-mean-square difference (RMSD) between depth-dose for each machine and the average depth-dose and MC results. Results: For a given collimator/build, depth-dose curves may vary by more than 10%, with some machines evidently systematic outliers. Plots of depth-dose curves clearly indicate when use of data from a single machine or averaged over multiple machines to generate commissioning data is warranted, and when the overall quality of measured data is sufficiently poor that MC data only is preferable. In the case of the MVC, the variation in OF with field size is 2-3 orders of magnitude less than the OF itself. Conclusion: Given an overview of the commissioning database, a straightforward workflow for generating accurate commissioning data using either existing measurements or MC data becomes possible. Variations in MVC OF with field size are small and appear to be machine-specific, making it preferable to simply calculate a fixed OF when commissioning with data measured on another machine.

Automated quantification of tumor volume and characterization of cachexia in a novel orthotopic lung cancer mouse model and patients

ABSTRACT. Introduction: Recent advances in medical imaging and deep learning enable automated tissue segmentation, enhancing our understanding of complex diseases. Cancer cachexia, particularly impactful in non-small cell lung cancer (NSCLC), contributes to weight loss and muscle/fat wasting, emphasizing the need for further investigations to uncover mechanisms and develop interventions. This study focuses on applying deep learning (DL) techniques to quantify cancer cachexia in both small animal orthotopic models and NSCLC patients, aiming to elucidate tissue wasting dynamics during concurrent chemo-radiotherapy.

Materials and Methods: To assess longitudinally tumor and cachexia development in a novel orthotopic lung cancer mouse model, treated with chemoradiation, we present a comprehensive approach involving automated tumor volume quantification and cachexia characterization. Initially, a two-step 3D U-Net architecture was employed to segment muscle volume, trained on a 32-case micro cone-beam CT (μCBCT) database of orthotopic lung model in mouse, correlating with experimentally measured wet muscle mass. Subsequently, a 3D UNet was subsequently used to segment orthotopic lung tumors, enabling the investigation of correlations between tumor growth curves and muscle mass decline. Our third effort utilized a 3D UNet model for precise segmentation of organs-at-risk (OAR) for radiation damage, including the heart, lungs, spinal cord, and thorax bone. For stage III NSCLC patients (n=140), a dataset comprising planning CT (pCT) and 1st fraction CBCT images was utilized to develop an DL-based solution for cachexia monitoring. The solution involves a two-step cascade: (a) DL-based conversion of CBCT to synthetic CT (sCT), and (b) DL-based segmentation of the pectoralis muscle in sCT images. Pectoralis muscle segmentation focused on the first axial slice above the aortic arch in CBCT scans (fractions 1, 10, 20, and 30). The best-performing CBCT-to-CT model converted selected CBCT scans to sCT, resulting in 326 sCT slices. Manual segmentation using Slice-O-Matic software preceded training a 2D U-Net model on sCT datasets, with segmentation performance assessed via Dice Similarity Coefficient (DSC).

Results: For the orthotopic mouse models, the muscle segmentation model achieved an average DSC of 93% and showed strong correlation (R2 = 0.92) with experimentally measured ex-vivo harvested wet muscle mass. The tumor model provided accurate presentations of the lung tumors with average DSC of 80% enabling simultaneous monitoring of tumor growth and muscle mass decline longitudinally. The proposed OAR model provided contours for the heart and lung with respectively 92 and 93% DSC. The performance was slightly lower for spinal cord and thorax bone segmentation with DSC of 79 and 74%, respectively, which is likely a consequence of the lack of stable high-quality ground truth contours for these structures. For the patients, the image conversion model restored the Hounsfield Unit (HU) distribution and visually reduced artifacts in CBCT scans. For segmentation on the sCT dataset, the 2D UNet achieved a mean DSC of 94% on the validation set, ranging from 86 to 98%.

Discussion: We developed various AI-based models for following the cancer cachexia not only in preclinical NSCLC models but also in patients. This also allows to significantly reduce the required number of animals compared to the ex-vivo muscle harvesting method, following the 3R principle.

Funding: This collaborative project is co-financed with a PPP allowance made available by Health~Holland, Top Sector Life Sciences & Health to stimulate public-private partnerships.

An Open-Source Python Pipeline for Multi-Modal Feature Extraction for Data Science in Pre-Clinical Research

ABSTRACT. Goal The primary objective of this study was to develop an open-source Python-based pipeline for pre-clinical radiation research applying animal models, as there are no commercial solutions. The goal was to enable feature extraction from various medical imaging modalities, including MR (Magnetic Resonance), CT (Computed Tomography), as well as dosimetric data from a treatment planning system (TPS). Furthermore, feature extraction which often relies on quantitative image analysis techniques was foreseen. By offering an open-source solution, our intention was to empower researchers in the pre-clinical domain, encouraging collaboration and innovation in the field of radiomics and machine learning-based medical imaging analysis for pre-clinical investigations.

Materials and Methods For developing such a pipeline for pre-clinical research, clinical images covering multimodality images and dosimetric data were used. For testing and validation data from an ongoing mouse brain irradiation study were utilized. The pipeline includes the following steps:

Data pre-processing: standardization of image orientation, voxel size, and intensity, which is in turn essential for subsequent registration. Modality fusion and registration: following registration techniques were embedded: affine, rigid, and deformable methods, to align and fuse CT and MR data. Parameters for registration were customized to cater to the specific demands of pre-clinical data.

ROI Segmentation: Delineation and segmentation of Regions of Interest (ROIs), utilizing label-map atlases for precision. It also features the capability to independently archive segmented volumetric data as well as identified ROIs, to facilitate subsequent analyses.

Radiomics Feature Extraction: Leveraging PyRadiomics a Python library, comprehensive feature extraction from the co-registered and segmented data was implemented. This process entails basic radiomics features, including first-order statistics, shape descriptors, and texture features.

Data Analysis and Reporting: Radiomics features can be analysed using dimension reduction techniques, either based on classical statistical methods or customized machine learning algorithms. The outcomes enable data-driven insights and quantitative assessments that are instrumental for pre-clinical investigations.

Results The in-house developed Python-based pipeline enables data and image handling from commercial micro-CT [spiral cone-beam µCT (Molecubes, Belgium, 0.225mAs, 50kV), micro-MR (15.2T Bruker BioSpin, Germany), as well as dosimetric and microdosimetric TPS generated data (Micro-RayStation, Sweden) from photon and charged particle irradiation. An initial radiomic analysis was successfully conducted on segmented regions of the mouse brain, followed by a comparative evaluation between cohorts of control and proton-irradiated mice. The PyRadiomics software library was used to extract features, resulting in the identification of 45 distinct features for each of the 336 brain segments from the 3D micro-MR images.

Conclusion The ability to effectively manage multiple imaging modalities while focusing on selected ROIs, as well as dosimetric data is essential for pre-clinical radiation research. The data-pipeline is currently being explored and utilized in the context of a study assessing responses to particle irradiation in a mouse brain model of normal tissue. Works in progress aim to include Principal Component Analysis (PCA) to enhance feature extraction and analysis, which will be complemented by multivariate statistical techniques as well.