Finite element analysis of knee joint contact mechanics is computationally expensive, which has motivated the development of graph neural network surrogate models. However, effectively representing long-range dependencies in joint mechanical responses remains challenging. Topology diffusion and global routing capture nonlocal interactions through different structural assumptions; determining whether either mechanism is sufficient, or whether they are complementary, is important for designing mechanically grounded surrogates. This study systematically compared topology diffusion, global routing, and their hybridization for surrogate modeling of knee joint contact mechanics. Using kinematic and force data from nine soccer players performing change-of-direction maneuvers, finite element simulations generated graph-structured samples for grouped three-fold cross-subject evaluation. Five architectures were compared: standard MeshGraphNet, hierarchical MeshGraphNet, a routing-only transformer, a topology-biased routing transformer, and a hybrid model. The hybrid model performed best overall, reducing RMSE by 20.0
Visually induced motion sickness (VIMS) is a common adverse effect of virtual reality (VR). Magnetic vestibular stimulation (MVS) may modulate vestibular function, but its effect on VIMS remains unclear. This study examined whether different MVS intensities were associated with changes in VIMS symptoms and resting-state EEG network characteristics in healthy participants exposed to VR. Twenty healthy young adults completed five repeated-measures sessions: normal/non-VR, VR-only, and VR combined with MVS at 90
Accurate identification of major depressive disorder (MDD) and characterization of abnormal higher-order brain-state transitions remain important in electroencephalography (EEG)-based depression research. Although EEG microstate analysis provides a promising high-temporal-resolution tool for probing abnormal brain dynamics, most existing studies focus on static microstate parameters and may overlook higher-order transition structure embedded in microstate sequences. To address this issue, we proposed a rule-driven microstate sequence analysis framework for EEG-based depression recognition. Specifically, the RuleGrowth algorithm was used to mine discriminative microstate transition rules, which were incorporated into an attention-based multiple instance learning model for classification. The proposed method outperformed conventional approaches based on static microstate features on both datasets, achieving accuracy, F1-score, and area under the curve (AUC) values of 96.67 q < 0.05 ), with 7 and 13 significant rules identified, respectively, and the C → D → B and A → D → B rules consistently showing higher confidence in the MDD group than in healthy controls across both datasets. These findings indicate that MDD is associated with disrupted higher-order brain-state switching dynamics and support the utility of rule-based microstate representations for EEG-based depression analysis.
Chronic kidney disease staging from longitudinal clinical records is challenging because routinely recorded GP stage labels are scarce, whereas rule-based eGFR labels are more abundant but may not capture the broader clinical context reflected in primary-care coding. This study proposes a hierarchical contrastive learning framework for CKD staging under heterogeneous supervision, designed to maximise the use of available data by leveraging both scarce GP annotations and abundant rule-based labels, which provide complementary but partially inconsistent information. Rather than treating these supervision sources as fully consistent or conflicting, the proposed approach models graded levels of agreement and incorporates this structure into a contrastive objective. This encourages clinically concordant cases to form compact representations while progressively separating increasingly discordant cases, yielding a structured and clinically meaningful embedding space. The framework is extensively evaluated across multiple neural architectures for longitudinal CKD modelling, including recurrent, convolutional, temporal convolutional, Transformer-based, and hybrid models. Results show consistent improvements over classification-only and binary contrastive baselines, with the best performance achieved using a TCN+Transformer backbone. These findings show that structured contrastive supervision can effectively exploit complementary information from clinician-annotated and rule-based labels and provide a practical framework for longitudinal disease modelling, with potential applicability beyond CKD.
Electrohysterogram (EHG) has been reported as promising way of predicting preterm birth, often relying on conventional machine learning approaches. However, the EHG modeling by deep learning has not been sufficiently investigated yet, probably due to both scarcity and complexity of the EHG data. To address these limitations, we propose CWT-AuxNet, an end-to-end deep learning method for preterm birth prediction with improved capability of learning discriminative patterns directly from raw EHG signals. In our method, continuous wavelet transform (CWT) is used to provide multiscale time-frequency representations of EHG data; an auxiliary feature termed peak amplitude (PA) is employed to further enhance discriminative capacity; and a cost-sensitive function with focal loss is designed during model training to handle severe data imbalance between term and preterm classes. Our CWT-AuxNet model is designed in a multibranch convolutional architecture enabling effective feature extraction and producing fine-grained window-level predictions. These window-level predictions are further aggregated using a user-level decision strategy to make individualized decisions regarding abnormalities and preterm risk during inference. Experimental results demonstrate that CWT-AuxNet consistently outperforms both traditional and deep learning baselines, achieving AUCs of 0.741 at the window level and 0.932 at the user level. Our study indicates the effectiveness of time-frequency deep representation learning and supports the potential of CWT-AuxNet as a promising framework for noninvasive preterm birth prediction.
Whole-slide image (WSI) classification is commonly formulated as multiple instance learning (MIL) with only slide-level labels, yet diagnostically relevant evidence is sparse and spatially organized. A persistent challenge is the spatial context dilemma: spatially agnostic MIL may over-attend isolated artifacts, whereas naive global context injection may oversmooth focal lesions or mix unrelated regions. We propose Neighbor-Constrained MIL (NCMIL), a dual-path MIL framework that combines a Nyströmformer global encoder with Neighbor-Constrained Attention (NCA). Rather than merely combining local and global branches, NCMIL introduces three design choices aimed at spatially structured WSI evidence: a fixed physical neighborhood with a hard attention mask to preserve tissue topology, similarity-weighted neighbor aggregation based on frozen tile embeddings to suppress spatially adjacent but morphologically inconsistent tiles, and adaptive local–global fusion to balance microenvironment fidelity with slide-level context. This inductive bias improves spatial coherence without an explicit graph-construction stage. Extensive 5-fold cross-validation on four public benchmarks shows that NCMIL achieves the best overall performance among the evaluated baselines, with absolute gains of up to +1.3 AUC points, +2.2 F1 points, and +2.2 ACC points over the strongest competitor.
Endoscopic retrograde cholangiopancreatography (ERCP) requires reliable bile duct cannulation, yet steerability of wire-driven steerable catheters (SCs) can deteriorate when the shaft is constrained, partly because actuation may induce unintended shaft-side deformation and transmission loss. This study evaluated whether non-straight wire-routing architectures could improve practical steerability of a bidirectional SC for ERCP. Three prototypes were developed: a conventional straight-wiring design (SC#0), a non-straight, non-crossed design (SC#1), and a non-straight, crossed design (SC#2). Steerability was assessed by incremental wire-pulling tests under straight-shaft and loose-shaft conditions and by a mock ERCP study using an actual duodenoscope. Outcomes included the handle displacement and pulling force required to reach 85° distal bending (HD85, HPF85) and target-angle reachability. In the straight-shaft condition, SC#1 and SC#2 reduced HD85 and HPF85 by 54–63
Bone distraction is a widely applied orthopedic procedure based on Ilizarov’s tension–stress principle, which requires high precision and accuracy to guarantee successful bone regeneration. In many low resource locations, the Aybar-type external fixator is the only viable treatment option due to its low cost and accessibility. However, its manual procedure introduces human error, reduces precision, and compromises treatment outcomes. To address this problem, this study develops and implements an automated micrometric distraction system for an Aybar-type fixator, aiming to provide a low cost yet highly accurate alternative, particularly focused on pediatric patients, for whom precision is critical to ensure proper callus formation and bone growth. Four control strategies were designed and evaluated: an ON–OFF controller, a Proportional–Integral (PI) controller, a Linear Quadratic Integrator (LQI), and a Fuzzy Logic Controller (FLC). The mechanical platform integrates a Direct Current (DC) motor, encoder, and worm gear with 1.25 mm pitch, while simulations and experimental tests were performed under two distraction regimens: 1 mm/day in four increments and 1 mm/day in sixty increments. Results indicate that the FLC achieves the best performance in coarse step elongation, while the LQI provides superior results in high-resolution protocols. These findings demonstrate that it is possible to enhance the precision and reproducibility of the Aybar external fixator through low cost automation. The proposed solution contributes to improving treatment quality for pediatric patients in contexts where advanced imported systems are unavailable, filling a critical gap between affordability and clinical efficacy.
Arrhythmias, a significant category of cardiovascular diseases, including atrial fibrillation (AF) and ventricular fibrillation (VF), can result in severe complications and fatal outcomes, necessitating timely detection and intervention. This study proposes an intracardiac heart sound classification method based on feature fusion and voting ensemble learning for accurate discrimination between AF and VF. Using high-fidelity sonocardiogram dataset, the signals are preprocessed using a window function and a pre-emphasis filter. Then, the four types of features, including Mel-frequency cepstral coefficients, envelope autocorrelation, Hilbert-Huang transform, and wavelet scattering transform are extracted. Next, feature fusion and dimensionality reduction are performed using the maximum relevance minimum redundancy algorithm to streamline inputs for machine learning. Finally, a voting ensemble approach incorporating K-nearest neighbors, support vector machines, and artificial neural networks as base classifiers is employed to reliably distinguish between AF and VF. Experimental results demonstrate that the proposed method achieves an accuracy of 96.7
Craniofacial bone regeneration is a demanding application of tissue engineering (TE) in which scaffold architecture—pore size, shape and spatial distribution—is a determinant of regenerative success, so that reliable geometric quality control is a prerequisite for translation. Conventional assessment of printed scaffold geometry is largely manual, time-consuming and subject to operator variability. Here we present a low-cost, semi-automated imaging platform that combines custom 3D-printed hardware—a 12.3 MP Raspberry Pi High Quality Camera with a 6 mm CS-mount lens, an adjustable monopod and a ring illuminator—with a dedicated image-processing pipeline. The system acquires a single zenithal image of a scaffold and segments it to quantify the number, area, perimeter and compactness of the pores of the uppermost printed layer; it therefore characterises two-dimensional surface macrotopography and does not resolve internal three-dimensional architecture or pore interconnectivity. On a printed reference grid the platform was highly repeatable (coefficient of variation, CV = 1.33
Electromyography (EMG)-based interventions have been proposed to improve hand/wrist motor recovery after stroke; however, their clinical value remains uncertain, as evidence for their effectiveness is inconsistent. This systematic review with meta-analysis evaluated the impact of different EMG-based rehabilitation approaches compared with non-EMG-based interventions in individuals with stroke. Seven databases were searched from inception to November 17, 2025, in accordance with PRISMA guidelines. Randomized controlled trials comparing EMG-based and non-EMG-based interventions for post-stroke hand or wrist rehabilitation were eligible. Interventions were classified into four categories: EMG-triggered electrical stimulation (EMG-ES), EMG-ES combined with other interventions, EMG-based robotic platforms, and EMG biofeedback combined with other interventions. Risk of bias was assessed using the Cochrane Risk of Bias 2 tool, and certainty of evidence was evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. Meta-analyses were conducted for clinical outcomes classified according to the International Classification of Functioning, Disability and Health (ICF) domains at post-intervention, follow-up <3 months, and follow-up ≥3 months. Thirty-eight studies involving 1,132 participants were included, of which 31 contributed to the meta-analyses. Eight studies were judged to have a low risk of bias, 13 to have some concerns, and 17 to have a high risk of bias. EMG-ES alone was not superior to conventional therapy. In contrast, EMG-ES combined with other interventions produced greater improvements across all ICF domains. EMG-based robotic platforms reduced wrist and finger spasticity, with effects sustained at long-term follow-up. EMG biofeedback primarily improved outcomes within the ICF body functions and structures domain. The effectiveness of EMG-based interventions depends on the rehabilitation strategy. EMG-ES, combined with other therapies, may be prioritized, whereas EMG-based robotic platforms may be useful in reducing spasticity. Further high-quality studies are needed to establish the clinical effectiveness of EMG-biofeedback approaches. The graphical abstract summarizes the findings of a systematic review and meta-analysis including 38 studies and 1,132 participants. Four types of sEMG-based rehabilitation interfaces were examined across the International Classification of Functioning, Disability and Health (ICF) domains of Body Functions and Structures, Activity, Participation, and Patient-reported outcomes. The symbols indicate the direction of the observed effects and the certainty of the evidence according to GRADE. The results suggest that the effects of sEMG-based rehabilitation vary according to the type of interface. The combination of sEMG-triggered electrical stimulation with other interventions produced the most consistent improvements across the ICF domains. In contrast, sEMG-triggered stimulation alone and sEMG-biofeedback combined with other interventions showed fewer and less consistent benefits. sEMG-based robotic platforms also showed potential, particularly for reducing spasticity. Together, these findings favor the use of combined sEMG-triggered interventions while supporting further investigation of sEMG-based robotic approaches in rehabilitation.
From skin to musculoskeletal prediction is a current challenge in biomechanics. Previous studies attempted head-to-skull prediction using statistical shape relationships, but they did not leverage local shape geometries and reduced muscles to simple 1-D action lines. We introduced Inside from Outside, a novel approach that reconstructs the bones of the skull and the facial muscles from external head surfaces. The method coupled Region-of-Interest (ROI) learning with Statistical Shape Relationship (SSR) modeling to capture detailed anatomical correspondences. We reconstructed 329 head and skull bone meshes from head-and-neck Computed Tomography (CT) scans. Multivariate regression models were employed to learn the statistical relationships among ROI regions. A ten-fold cross-validation procedure was used to estimate the optimal number of shape parameters, the multivariate regression method, and the prediction strategies. As a result, the prediction errors were 1.9293 ± 0.1838 mm, 1.4464 ± 0.3044 mm, and 1.4933 ± 0.2470 mm for the predicted skull bones, skull shapes, and muscle attachment points, respectively. Our novel SSR- and ROI-based approach achieved more reliable accuracy in the head-to-skull prediction problem. The proposed approach will be used to perform patient-specific, head-based simulations of facial expressions to personalize rehabilitation strategies for patients with facial palsy. Fast and accurate prediction of 3D internal geometries from external geometries. Novel ROI–SSR coupling method to predict 3D skull bones and facial muscles from 3D head shape. Head-to-Skull-and-Muscles translation with a mean error range of 1.5–1.9 mm. Enables patient-specific facial rehabilitation prediction and prevention.
Single cell surgery, which involves removing or manipulating subcellular organelles from single cells, is increasingly being utilized in precision medicine to investigate illnesses and their causes. This article describes an optical tweezers (OTs)-assisted mitochondria biopsy method for performing minimally invasive automated organelle biopsy of single cells. A microfluidic chip device is utilized to hold a single cell, and OTs is used to trap and move the mitochondria to the edge of the cell membrane automatically, followed by automatic mitochondria biopsy with a bevelled microneedle. An image processing technique is also being developed to identify the location of the mitochondria and cell. To achieve precise and robust manipulation of organelles within the viscous cytoplasmic environment, a hybrid PID–adaptive sliding mode controller is implemented, ensuring accurate positioning of mitochondria for biopsy. The efficacy of the proposed robotic surgical system is proven experimentally using automated mitochondria biopsy from Hela cancer cells. Following mitochondrial extraction, JC-1 staining and cell viability assays were performed to evaluate immediate mitochondrial integrity as well as biopsied cell viability. The experimental findings indicate that the proposed OT aided mitochondria biopsy system outperforms the current mitochondria biopsy techniques in terms of cell viability and biopsy efficiency.
Digital pathology has enabled large-scale analysis of histological images. However, accurate detection of cellular nuclei remains challenging due to variability in morphology, staining, and especially image resolution. Existing object detection approaches often degrade when applied to low-resolution images, and current solutions typically address either multi-scale detection or image enhancement independently. In this work, we propose a hybrid framework that integrates super-resolution with a dual-branch detection strategy combining full-image and patch-based inference. This design leverages both global contextual information and localized high-detail analysis to improve detection robustness. The outputs of both branches are fused through a confidence-weighted mechanism followed by non-maximum suppression and clustering-based refinement. The proposed method was evaluated on the NuCLS dataset, demonstrating consistent improvements over baseline detection approaches. In particular, the combined workflow achieved up to a 20
Accurate delineation of epidermis and tumor boundaries is central to melanoma staging, yet pixel-level annotation on whole-slide images (WSIs) is labor-intensive and inconsistent across observers. Advancing this field requires standardized, publicly available benchmarks with expert-validated labels. We introduce Mel-DEPTHS, a new benchmark dataset for epidermis and tumor segmentation, designed to accelerate and standardize research for automated melanoma staging. Mel-DEPTHS comprises 50 anonymized melanoma WSIs (40x, 0.25 m/pixel) with pixel-level masks for epidermis and tumor regions. Clinical variables such as invasion depth, ulceration, and pT stage are provided alongside fixed train/test partitions to ensure reproducibility. To mitigate annotation burden, we developed an Expert-Supervised Iterative Self-Training (ESIST) protocol: a pretrained model generates pseudo-labels, which dermatopathologists iteratively refine for retraining. We benchmarked six state-of-the-art segmentation models (UNet, UNet++, UNet3+, UPerNet, TransUNet, ConvUNeXt) using WSI-level precision, recall, IoU, and Dice. TransUNet achieved the best performance, closely followed by ConvUNeXt and UperNet. Three-fold cross-validation also confirmed consistent model rankings and label robustness. Mel-DEPTHS provides the fidelity and diversity necessary for clinically meaningful segmentation. It establishes a standardized benchmark and fosters reproducibility in computational pathology.
Integrating medical data across hospitals has become a critical challenge in medical informatics, largely due to the heterogeneity of electronic medical record (EMR) systems. This study aims to address this issue by developing a diagnosis classification model that automatically maps diagnosis spans in EMR data to the standardized clinical ontology Systematized Nomenclature of Medicine—Clinical Terms (SNOMED-CT). We trained an embedding-based model, ClinicalBERT, on EMR data to obtain latent representations of diagnosis spans. These representations were aligned with SNOMED-CT fully specified names (FSNs) through mean squared error (MSE)-based fine-tuning and subsequently used to construct a downstream classification model. In addition, we analyzed the latent representations to investigate semantic alignment and structural characteristics in the latent space. The proposed model achieved an accuracy of 0.934, a weighted F1-score of 0.923, and a macro-averaged F1-score of 0.823 on 273 SNOMED-CT classes. Despite improved semantic alignment, the fine-tuned model demonstrated performance comparable to the base ClinicalBERT model, while remaining competitive with state-of-the-art approaches such as SapBERT (accuracy: 0.944) and BioSyn (accuracy: 0.931). Representation analysis further revealed improved clustering coherence, as evidenced by reduced similarity distances and more distinct class separation. This approach demonstrates strong potential for scalable and privacy-preserving medical concept normalization in real-world clinical environments. Fine-tuning improved semantic alignment, reducing the average Manhattan and cosine distances by 36.2
Rapid, non-invasive detection of intracranial hemorrhage at the point of care is crucial for improving patient outcomes after head trauma. To overcome the limitations of bulky and expensive gold-standard imaging systems like CT and MRI, this study introduces a portable detection system based on electromagnetic induction. The core of our method is a simplified single-coil sensor that detects hemorrhage-induced changes in tissue conductivity by monitoring variations in coil impedance, facilitated by a high-precision inductance-to-digital converter (LDC1101). We rigorously evaluated the system through electromagnetic simulation, customized coil design, hardware implementation, and software development featuring an RGB localization algorithm. Experiments using a 3D-printed head phantom demonstrated that the system effectively distinguishes simulated ICH lesions (30 mL NaCl solution) from normal brain tissue and successfully localizes the hemorrhage. This work establishes a novel and practical technological pathway for developing portable, low-cost ICH monitoring devices suitable for bedside and point-of-care applications.
As virtual training environments for colonoscopy become more prevalent, addressing their limitations such as the scarcity of simulation scenarios and the lack of open, shareable models becomes essential. This paper introduces an editing tool developed to mitigate these issues by enabling non-technical users, such as surgeons, to create varied 3D models of the large intestine. The software employs a parametrized approach using a Bezier spline with control nodes, presets, blendshapes, and specific meshes to enable the generation of a wide range of large intestine configurations, sizes, and pathology conditions. The paper also includes the results of a pilot evaluation focused on the editor utility, model fidelity, and usability, involving 10 colonoscopy practitioners who completed a modeling task and provided feedback via a questionnaire. Results indicate a positive reception, highlighting the realism of the models, user-friendliness, and overall utility. This solution not only addresses key limitations in colonoscopy training, but also lays the foundation for future advancements in colonoscopy simulation technology.
Soft tissue balance is crucial for the success of total knee arthroplasty (TKA), influencing joint stability, patient satisfaction, and implant longevity. However, intraoperative assessment of ligament tension is subjective and prone to error. In this study, we present a novel, low-cost, and highly sensitive force measurement system based on flexible sensors for real-time evaluation of soft tissue balance during TKA. The system integrates 24 sensor units (12 each in the medial and lateral compartments) embedded within a custom-designed spacer, enabling continuous real-time force measurements throughout the full knee motion range, including clinically relevant positions: full extension (0°), mid-flexion (45°), and deep flexion (90°). The sensors, fabricated from RTV silicone rubber filled with carbon nanotubes (CNTs) and carbon black, show high sensitivity (0.117 N⁻1) at low forces and stable performance across a wide detection range (0–100 N). A complete testing platform, including data acquisition circuitry and a graphical user interface (GUI), was developed to visualize pressure distribution during simulated surgery. The proposed system demonstrates rapid response ( 0.4 s), low hysteresis ( 1.18
Surface electromyography (sEMG) and electrical muscle stimulation (EMS) are increasingly used in rehabilitation and training. However, turning EMG–EMS setups into usable end-to-end interactive experiences remains challenging due to real-time constraints, safety, and user-facing workflow complexity. We investigate an interactive EMG–EMS exercise workflow that supports exercise execution and post-session reflection, and we evaluate it from an interaction-centered perspective focusing on usability, learnability, perceived usefulness, and comfort. The system integrates real-time sEMG visualization with session history and session comparison for reflective analysis. During the study, EMS assistance was delivered in a manual, operator-mediated manner, in which the application provided real-time stimulation cues that guided manual adjustments of a commercial EMS device. A small exploratory pilot study with healthy participants ( n = 19 ) examined the full interaction workflow in two sequential phases (sEMG-only followed by EMS-assisted), using think-aloud, task-level performance logging, and questionnaires (SUS and domain-specific). The evaluation showed positive perceived usability and user experience (SUS mean: 78.95/100), with participants particularly valuing session history and comparison as tools for post-exercise reflection. Real-time sEMG feedback was delivered with stable behavior and an effective processing delay on the order of tens of milliseconds. At the same time, the study revealed recurring challenges inherent to EMG–EMS interaction, including sensor onboarding, electrode placement, and users’ interpretation of sEMG-derived feedback, which point toward important directions for future system design. Findings provide preliminary support for the feasibility of an interactive EMG–EMS exercise experience with positive perceived usability and comfort in a healthy, technologically literate pilot sample. However, the results should be interpreted as exploratory because of the small sample, sequential study design, and operator-mediated EMS delivery. These findings motivate HCI-focused iterations to improve onboarding, interpretability, and future evaluation with more reliable and automated hardware.
