Knee osteoarthritis (OA) is a prevalent, disabling disease for which early, accurate diagnosis is essential to guide treatment and reduce long-term burden. Machine learning (ML) approaches using quantitative imaging biomarkers show promise for automated OA classification, but their reliability under imperfect image segmentation remains unclear. This study evaluated the robustness of cartilage-based radiodensity and morphological features derived from MRI-registered CT scans against simulated segmentation errors. Manual segmentations of femoral, patellar, lateral tibial, and medial tibial cartilages were modified using morphological operations (erosion, dilation, and closing) to imitate automated and manual segmentation inaccuracies. A total of 130 knee scans (79 control, 51 degenerative) were analyzed. Several ML models, including tree-based models and support vector classifiers (SVC) with different kernels, were trained and tested using nested cross-validation. Statistical analyses confirmed that cartilage density variation and medial tibial cartilage volume and surface area remained significant discriminators despite pixel perturbations. Among ML models, SVC with radial basis kernel (RBF SVC) achieved the highest performance on the original dataset (F1 0.86, ROC AUC 0.91), with Linear and RBF SVC and Logistic Regression performing comparably under error-modified datasets (F1 between 0.80 and 0.86, ROC AUC between 0.86 and 0.92). While tree-based models were more sensitive to dilation errors, most models maintained weighted F1 scores ≥ 0.75. These findings demonstrate that ML classifiers can robustly distinguish degenerative from control knees even when cartilage masks are imprecisely segmented. Thus, highly precise manual segmentations may not be strictly required for reliable OA classification, suggesting potential for scalable and cost-effective deployment of ML-based diagnostic tools in clinical imaging workflows.

Parkinson's disease (PD) is a progressive neurodegenerative disorder characterised by motor impairments extending to balance and postural regulation. Although EEG abnormalities in oscillatory activity and functional connectivity are well documented in PD, large-scale brain dynamics during tasks directly engaging postural control remain poorly understood. To address this gap, we examined EEG microstate organisation during the BioVRSea virtual-reality postural-control task in early-stage PD patients (n = 30) and matched healthy controls (HC; n = 26). EEG microstates, brief quasi-stable scalp topographies representing global neural states, provide a robust framework for characterising the rapid temporal structure of whole-brain activity. Although task-based microstate approaches exist, applications to PD remain limited and largely confined to resting-state research. Topographical analyses revealed pronounced between-group differences in microstates D and E, whose group-averaged maps in PD diverged markedly from canonical configurations. Because these maps were not topographically equivalent between groups, comparisons of temporal parameters were restricted to the comparable microstates A-C. Across all task phases, PD patients showed increased duration and coverage of microstates A and B. Transition-probability analysis, likewise restricted to A-C, indicated different trajectories across phases in PD and HC. A single significant Group × Phase interaction emerged for A→C: the largest between-group difference occurred during the POST-movement phase, when PD showed a higher probability than HC. For the remaining transitions, no phase-dependent differences emerged within PD. Because patients were assessed ON medication and clinical or behavioural correlates were unavailable, these findings represent candidate task-state EEG microstate alterations requiring validation, rather than established disease-specific markers.

INTRODUCTION:To investigate the epidemiology of high-energy thoracic and lumbar spine fractures at Landspítali and evaluate their treatment and associated injuries using international injury classification systems. MATERIALS AND METHODS:Medical records were reviewed with regard to age, sex, nationality, length of hospital stay, mechanism of injury, associated injuries, treatment, neurological injury, seasonality, and injury location. Imaging studies were used for diagnosis and for classification of fractures according to the AO injury classification system. RESULTS:A total of 424 individuals were diagnosed with 596 fractures during the study period. On average, there were 21.2 injuries per year; the mean age was 40.3 years, and men accounted for just over 62% of cases. Motor vehicle collisions were the cause in approximately 21% of cases, rollovers in about 20%, and among rollover accidents, just over 31% involved foreign individuals. The most common fracture levels were L1 (approximately 19% of all fractures) and Th12 (just over 11%). Neurological injury was present in just over 20% of cases involving fractures at Th12, L1, and Th7. Overall, 36% of patients underwent surgical treatment, and an additional 17% received external support. Associated injuries to other parts of the body were present in just over 63% of cases. CONCLUSION:On average, approximately two thoracic and lumbar spine fractures caused by high-energy trauma were treated per month at Landspítali during the period 2003-2022. A typical case was a fracture at Th12 or L1 in a 40-year-old man. Patients without neurological deficits had a significantly shorter length of hospital stay. The number of injuries decreased over time, although not statistically significantly; this trend could, for example, be explained by improved vehicle safety equipment.

Parkinson’s Disease (PD) is a neurological disorder characterized by impaired postural control (PC) and balance issues. To date, few studies have explored the relationship between brain activity and responses during specific tasks designed to challenge balance in individuals with PD. Our exploratory research employs an innovative paradigm to assess PC by integrating virtual reality (VR) and electroencephalography (EEG). In the study, 20 individuals diagnosed with PD who self-reported postural instability participated in the BioVRSea paradigm. This paradigm tested their PC using visuomotor stimuli and collected EEG signals to assess brain responses throughout the experiment. The results of the Parkinson’s group were compared with those of 22 age-matched healthy controls (CTR). From the functional connectivity between brain regions, we extracted brain network states (BNSs) using the k-means++ clustering algorithm. These BNSs capture the dynamic organization of brain activity and were compared with canonical resting-state networks (RSNs) to investigate neural alterations in individuals with PD. Six distinct BNSs were identified, with the dorsal attention network (DAN) dominant in five states. A significant reduction in the coverage of BNS2 was observed in PD patients during both the PRE (adjusted p-value = 0.019) and MOV (adjusted p-value = 0.036) phases compared to CTR. This reduced BNS2 coverage suggests impaired visuomotor integration in PD patients during PC tasks. DAN dominance highlights its crucial role in maintaining attentional control during the task. The findings of this study highlight the potential of using brain dynamics as a biomarker of neural dysfunction in PD, especially during specific PC tasks. Altered BNSs, particularly in networks associated with attention and sensorimotor integration, reveal key neural deficits related to PD.

IntroductionKnee osteoarthritis (KOA) is a chronic and progressive joint disease that affects middle-aged and older adults. Early detection is crucial to prevent progression toward joint replacement and improve long-term outcomes, yet current diagnoses are strongly influenced by subjective symptoms, especially pain perception, which varies widely across individuals and does not reliably reflect structural degeneration. This study introduces a semi-supervised learning (SSL) framework for characterizing KOA stages through combined MRI and CT-derived cartilage features.MethodsA cohort of 133 knee scans was analyzed, including 36 expert-labeled cases categorized as healthy, early degeneration, or advanced degeneration. These labels served as seeds for graph-based SSL using Label Propagation and Label Spreading, producing pseudo-labels for the remaining samples.ResultsLabel stability across ten Monte Carlo runs demonstrated high agreement (0.91 ± 0.14) and substantial reliability (Fleiss’ kappa = 0.781). Supervised classifiers trained on the SSL-labeled dataset achieved robust performance, with Support Vector Machines and Logistic Regression yielding the highest weighted F1-scores (0.84 and 0.81, respectively). Statistical analysis confirmed significant differences among the three classes for all extracted features.DiscussionThe volume-to-surface ratio and density heterogeneity demonstrated the strongest discriminatory power, reflecting progressive cartilage thinning, surface irregularity, and increasing structural heterogeneity consistent with KOA pathophysiology. These results show that combining expert knowledge with SSL enables reliable KOA stratification even with limited labeled data, offering meaningful insights into cartilage degeneration and laying the foundation for quantitative and more objective imaging-based biomarkers and future continuous scoring systems.

Heart Rate Variability (HRV), a marker of autonomic nervous system balance, is influenced by factors such as age, gender, trauma, and overall health. This study used BioVRSea, a virtual reality-integrated motion platform, to assess HRV responses during postural control challenges. We investigated whether the multisensory setup, involving visual, acoustic, and motion stimuli, could differentiate participants based on age (over-40, under-40), gender, and concussion status based on HRV analysis. Pearson correlation was used to identify HRV features associated with age, and statistical analyses revealed significant differences related to both age and gender. Participants with a history of concussion exhibited distinct HRV patterns compared to healthy controls, particularly during the movement and recovery phases of the experiment. Future research will focus on expanding the cohort, especially among individuals with medical conditions, to enhance statistical power. Key goals include improving system portability and further exploring HRV as a potential biomarker for concussion.

The ability to maintain our body's balance and stability in space is crucial for performing daily activities. Effective postural control (PC) strategies rely on integrating visual, vestibular, and proprioceptive sensory inputs. While neuroimaging has revealed key areas involved in PC-including brainstem, cerebellum, and cortical networks-the rapid neural mechanisms underlying dynamic postural tasks remain less understood. Therefore, we used EEG microstate analysis within the BioVRSea experiment to explore the temporal brain dynamics that support PC. This complex paradigm simulates maintaining an upright posture on a moving platform, integrated with virtual reality (VR), to replicate the sensation of balancing on a boat. Data were acquired from 266 healthy subjects using a 64-channel EEG system. Using a modified k-means method, five EEG microstate maps were identified to best model the paradigm. Differences in each microstate maps feature (occurrence, duration, and coverage) between experimental phases were analyzed using a linear mixed model, revealing significant differences between microstates within the experiment phases. The temporal parameters of microstate C showed significantly higher levels in all experimental phases compared to other microstate maps, whereas microstate B displayed an opposite pattern, consistently showing lower levels. This study marks the first attempt to use microstate analysis during a dynamic task, demonstrating the decisive role of microstate C and, conversely, microstate B in differentiating the PC phases. These results demonstrate the utility of microstate technique in studying temporal brain dynamics during PC, with potential applications in the early detection of neurodegenerative diseases.

Aging is a significant factor influencing postural control (PC), leading to a progressive decline in postural stability, with older adults exhibiting greater instability compared to younger individuals. This study investigates cortical activity in 289 healthy individuals, 72% under 40 years old and 28% over 40, during a complex postural task using high-density electroencephalography (EEG) recordings. Relative Power Spectral Density (PSD) was calculated for each electrode, and age-related differences were assessed. The statistical analysis, highlighting electrodes significant at $p < 0.05$ and after False Discovery Rate (FDR) correction, revealed increased Beta and Low Gamma power in the Over40 group, while Theta and Alpha power were higher in the Under40 group across nearly all stimulation phases. Behavioral indices further supported these findings, indicating reduced physical and neurocognitive status in older adults and greater responsiveness to multisensory stimulation in younger participants. These variations across frequency bands and behavioral dimensions reflect distinct PC responses and underline age-related neural and musculoskeletal changes affecting balance. This approach supports the identification of potential biomarkers and can be extended to neurological populations. These findings underscore the importance of age when designing interventions to assess and improve balance, and they may contribute to the early detection of age-related postural deficits.

The growing use of machine learning in healthcare raises concerns about computational cost and environmental sustainability, particularly when working with high-dimensional datasets. In this work, we investigate the optimization of a scalable ensemble feature selection strategy for knee osteoarthritis classification using radiomic data from cartilages segmented on MRI. We explored threshold tuning to balance the trade-off between dimensionality reduction and predictive accuracy in a two-stage ensemble feature selection pipeline. Results show that models trained on optimized subsets of features achieve comparable or superior F1-scores to models trained on the full feature set, while requiring significantly fewer features. The performance increase compared to the full set is almost 20%, using only 4% of the features from the original dataset. Moreover, only 29 radiomic features have been demonstrated to be sufficient for distinguishing between healthy and degenerative knees. These findings demonstrate how sustainable feature selection can enhance technological efficiency and promote generalizable solutions in healthcare AI.

Objectives This study aimed to enhance the comprehension of volumetric bone mineral density (vBMD) changes following Total Hip Arthroplasty (THA) by establishing a protocol to (i) precisely locate alterations in the proximal femur in three dimensions and (ii) evaluate these changes over an extended period. Methods Twelve individuals who underwent unilateral THA, using either cemented or uncemented prostheses, were recruited. CT-scans of the proximal femur were acquired at three distinct time points: 24 h, 1 and 6 years post-surgery. Utilizing the acquired data, 3D models of the proximal femur were generated, and a novel algorithm was developed to categorize them into Gruen zones. Comparative analysis of density values among the three sets of scans allowed the calculation of bone density gains/losses for the entire proximal femur and specific regions. Results A lower trabecular bone quantity was observed in the cemented group compared to the uncemented cohort, with discernible differences in vBMD evolution observed in the overall femur and certain Gruen zones. Noteworthy inter-patient variability was evident, ranging from physiological bone remodeling to unexpected increases/decreases in vBMD (e.g.,+340 % after one year). Conclusions This analysis proves to be a valuable tool to understand the long-term vBMD evolution in THA patients.

The Center of Pressure (CoP) is a widely used metric to assess postural control (PC). In this study, CoP data from 304 healthy individuals were collected during a balance task performed using the BioVRSea system. Spectral, non-linear, and temporal features were extracted to identify differences in PC strategies based on gender and age group (Over40 vs. Under40). For comparisons across phases, the Friedman test was performed, followed by the Wilcoxon rank sum test as a post hoc analysis. To assess differences within each experimental phase, the Mann-Whitney U test or the t-test were used, depending on data normality. No significant gender differences were found within the phases ($p$>0.05). Age-related differences in postural control were significant during the PRE and POST phases (p<0.001), with greater sway and reduced stability in the Over40 group. Some instability measures in the Over40 group persisted after the perturbation, showing sustained balance challenges. These findings demonstrate the utility of the BioVRSea system and both spectral and conventional CoP analyzes to detect age-related PC changes and support their use in balance assessment for aging populations.

Background/Objectives: Sarcomas are a rare and heterogeneous group of malignant tumors, which makes early detection and grading particularly challenging. Diagnosis traditionally relies on expert visual interpretation of histopathological biopsies and radiological imaging, processes that can be time-consuming, subjective and susceptible to inter-observer variability. Methods: In this study, we aim to explore the potential of artificial intelligence (AI), specifically radiomics and machine learning (ML), to support sarcoma diagnosis and grading based on MRI scans. We extracted quantitative features from both raw and wavelet-transformed images, including first-order statistics and texture descriptors such as the gray-level co-occurrence matrix (GLCM), gray-level size-zone matrix (GLSZM), gray-level run-length matrix (GLRLM), and neighboring gray tone difference matrix (NGTDM). These features were used to train ML models for two tasks: binary classification of healthy vs. pathological tissue and prognostic grading of sarcomas based on the French FNCLCC system. Results: The binary classification achieved an accuracy of 76.02% using a combination of features from both raw and transformed images. FNCLCC grade classification reached an accuracy of 57.6% under the same conditions. Specifically, wavelet transforms of raw images boosted classification accuracy, hinting at the large potential that image transforms can add to these tasks. Conclusions: Our findings highlight the value of combining multiple radiomic features and demonstrate that wavelet transforms significantly enhance classification performance. By outlining the potential of AI-based approaches in sarcoma diagnostics, this work seeks to promote the development of decision support systems that could assist clinicians.

Knee osteoarthritis (OA) is a leading cause of disability in older adults, with early diagnosis critical for timely intervention. However, class imbalance and subjective symptom reporting in medical datasets limit the effectiveness of automated diagnostic tools. This study investigates a novel oversampling method called Secondary label SMOTE (S-SMOTE) to improve knee OA classification using imaging-derived features from 133 knee scans. In addition to primary binary labels (control vs. degenerative), each subject was assigned an expert-defined secondary label representing anatomical cartilage degeneration severity across knee compartments. These dual labels lead to a finer subclass definition, yielding six distinct categories. S-SMOTE uses this structural degeneration information to generate synthetic minority samples, improving representation without introducing bias. We evaluated classification performance using Random Forest, Support Vector Machine, and Logistic Regression models across stratified repeated splits. S-SMOTE significantly improved accuracy in imbalance-sensitive models, particularly Logistic Regression, while maintaining or enhancing model stability. Compared to standard SMOTE, Borderline SMOTE, and ADASYN, S-SMOTE better captured biological variability and produced more clinically meaningful samples. Our findings highlight the importance of incorporating expert anatomical assessments in data augmentation to enhance classifier generalization and robustness in early knee OA detection.

Postural control (PC) is a crucial function influenced by both habituation and adaptation processes. The BioVRSea paradigm, a novel virtual reality-based approach, provides a comprehensive $\mathbf{P C}$ assessment by exposing participants to visual and motor perturbations while measuring electromyographic and center of pressure responses. This study aims to compare the PC responses of sailors recently exposed to prolonged maritime conditions with those of a control group of healthy individuals, focusing on differences in adaptation and habituation strategies. Sailors demonstrated increased complexity and variability in postural sway and distinct neuromuscular activation patterns, suggesting functional adaptations to unstable environments. By highlighting these contrasts, the study seeks to understand better the impact of long-term sea exposure on balance regulation mechanisms. This paper demonstrated a significant difference in the habituation and adaptation phases between sailors and controls, with sailors exhibiting greater sway dynamics and complexity, as well as a distinct Soleus activation pattern indicative of efficient, burst-based motor strategies.

Objective: This study introduces an explainable, radiomics-based machine learning framework for the automated classification of sarcoma tumors using MRI. The approach aims to empower clinicians, reducing dependence on subjective image interpretation. Methods: A total of 186 MRI scans from 86 patients diagnosed with bone and soft tissue sarcoma were manually segmented to isolate tumor regions and corresponding healthy tissue. From these segmentations, 851 handcrafted radiomic features were extracted, including wavelet-transformed descriptors. A Random Forest classifier was trained to distinguish between tumor and healthy tissue, with hyperparameter tuning performed through nested cross-validation. To ensure transparency and interpretability, model behavior was explored through Feature Importance analysis and Local Interpretable Model-agnostic Explanations (LIME). Results: The model achieved an F1-score of 0.742, with an accuracy of 0.724 on the test set. LIME analysis revealed that texture and wavelet-based features were the most influential in driving the model’s predictions. Conclusions: By enabling accurate and interpretable classification of sarcomas in MRI, the proposed method provides a non-invasive approach to tumor classification, supporting an earlier, more personalized and precision-driven diagnosis. This study highlights the potential of explainable AI to assist in more secure clinical decision-making.

Osteoarthritis (OA) is a prevalent form of arthritis, characterized by the degradation of joints hyaline cartilage, particularly the knee, which often necessitates surgical intervention due to its limited ability to self-healing. This study investigates the potential of using radiomics and machine learning (ML) for the automatic classification of degenerative and healthy knees in the medial and lateral anatomical compartments. A multimodal dataset comprising MRI and CT images of OA-diagnosed and healthy subjects was used. Image segmentation and registration were conducted using Mimics software, and radiomic features were extracted from CT and MRI scans. Feature selection via Recursive Feature Elimination (RFE) and ML algorithms were employed for classification. Results indicate superior classification performance in the medial compartment, both for MRI and CT images, suggesting their importance in OA diagnosis. This study contributes to advancing non-invasive OA diagnosis, with implications for personalized treatment strategies.

Knee osteoarthritis (KOA) is a prevalent degenerative joint condition. Despite the prevalence of this pathology, there is still a lack of adequate investigation into the importance of musculature toward KOA. This study investigates the relationship between cartilage degeneration and alterations in the density and volume of the leg's adipose and skeletal muscle tissue (SMT) using an innovative virtual histology technique on computed tomography (CT) images. Subjects diagnosed with KOA and healthy controls were recruited. The analysis revealed that intramuscular adipose tissue (IMAT) has a great impact on the progression of the disease, KOA patients had higher IMAT volumes and altered muscle volumes. Gender differences were more pronounced in healthy individuals, while degenerative conditions appeared to level these variations. Moreover, muscle and IMAT densities showed a strong correlation with body mass index (BMI). Our findings highlight the significant impact of IMAT on muscle health and function in KOA, emphasizing it as a crucial factor in disease management and progression.

This study addresses the assessment of knee osteoarthritis (OA), a degenerative joint disease primarily affecting cartilage. While current diagnostic methods like X-rays, CT scans, and MRI scans are subjective, this research aims to develop more objective assessment methods. Focusing on cartilage status in MRI and CT images, the study uses machine learning (ML) techniques to identify predictive parameters for OA severity assessment. Significant differences in cartilage metrics between OA patients and healthy individuals were found. Using the statistical differences a smaller feature set has been extracted and ML analysis (SVM and Logistic Regression) confirmed the effectiveness of these features. This research highlights the importance of precise image analysis and the potential of ML in enhancing OA severity assessment. The emphasis on tibial cartilages aligns with biomechanical considerations, offering insights for personalized treatment approaches.

Background and Objective: Tendon segmentation is crucial for studying tendon-related pathologies like tendinopathy, tendinosis, etc. This step further enables detailed analysis of specific tendon regions using automated or semi-automated methods. This study specifically aims at the segmentation of Achilles tendon, the largest tendon in the human body. Methods: This study proposes a comprehensive end-to-end tendon segmentation module composed of a preliminary superpixel-based coarse segmentation preceding the final segmentation task. The final segmentation results are obtained through two distinct approaches. In the first approach, the coarsely generated superpixels are subjected to classification using Random Forest (RF) and Support Vector Machine (SVM) classifiers to classify whether each superpixel belongs to a tendon class or not (resulting in tendon segmentation). In the second approach, the arrangements of superpixels are converted to graphs instead of being treated as conventional image grids. This classification process uses a graph-based convolutional network (GCN) to determine whether each superpixel corresponds to a tendon class or not. Results: All experiments are conducted on a custom-made ankle MRI dataset. The dataset comprises 76 subjects and is divided into two sets: one for training (Dataset 1, trained and evaluated using leave-one-group- out cross-validation) and the other as unseen test data (Dataset 2). Using our first approach, the final test AUC (Area Under the ROC Curve) scores using RF and SVM classifiers on the test data (Dataset 2) are 0.992 and 0.987, respectively, with sensitivities of 0.904 and 0.966. On the other hand, using our second approach (GCN-based node classification), the AUC score for the test set is 0.933 with a sensitivity of 0.899. Conclusions: Our proposed pipeline demonstrates the efficacy of employing superpixel generation as a coarse segmentation technique for the final tendon segmentation. Whether utilizing RF, SVM-based superpixel classification, or GCN-based classification for tendon segmentation, our system consistently achieves commendable AUC scores, especially the non-graph-based approach. Given the limited dataset, our graph-based method did not perform as well as non-graph-based superpixel classifications; however, the results obtained provide valuable insights into how well the models can distinguish between tendons and non-tendons. This opens up opportunities for further exploration and improvement.

Osteoarthritis (OA) is a common joint disease affecting people worldwide, notably impacting quality of life due to joint pain and functional limitations. This study explores the potential of radiomics — quantitative image analysis combined with machine learning — to enhance knee OA diagnosis. Using a multimodal dataset of MRI and CT scans from 138 knees, radiomic features were extracted from cartilage segments. Machine learning algorithms were employed to classify degenerated and healthy knees based on radiomic features. Feature selection, guided by correlation and importance analyses, revealed texture and shape-related features as key predictors. Robustness analysis, assessing feature stability across segmentation variations, further refined feature selection. Results demonstrate high accuracy in knee OA classification using radiomics, showcasing its potential for early disease detection and personalized treatment approaches. This work contributes to advancing OA assessment and is part of the European SINPAIN project aimed at developing new OA therapies.