Computer Methods And Programs In Biomedicine[JOURNAL]
Sun
200 papers
Effects of tibial fracture-induced gait alterations on healing outcomes: Implications for patient-specific rehabilitation strategies.
Ding Q, Li L, Miramini S … +4 more , Shi S, Ebeling P, Wei Y, Zhang L
Comput Methods Programs Biomed · 2026 May · PMID 41690261 · Publisher ↗
BACKGROUND AND OBJECTIVE: Partial weight-bearing (PWB) exercise is critical in fracture rehabilitation. However, the effects of tibial fracture-induced gait alterations on musculoskeletal loading conditions, particularly...
BACKGROUND AND OBJECTIVE: Partial weight-bearing (PWB) exercise is critical in fracture rehabilitation. However, the effects of tibial fracture-induced gait alterations on musculoskeletal loading conditions, particularly regarding body weights (BWs) and walking speeds, remain unclear. These patient-specific gait deviations can substantially modify mechanical stimuli at the fracture site, thereby affecting healing trajectories. This study aims to evaluate how fracture-altered gait mechanics influence loading conditions during rehabilitation to inform patient-specific strategies. METHOD: A gait-based biomechanical model was developed to simulate PWB walking in patients with tibial fractures, based on changes in ground reaction forces (GRFs) compared to the uninjured limb. Fracture-induced effects on joint and muscle loading rates were quantified and integrated into the fracture healing model, simulating mesenchymal stem cell (MSC)-mediated tissue differentiation under various walking speeds and BWs. RESULTS: Tibial fracture-induced gait changes increased peak loading rates at knee and ankle joints by 0.9-1.3 BW/s and 0.8-1.2 BW/s, respectively, as speed rises from 0.6 to 1.0v(v = 5 km/h). The tibialis posterior and rectus femoris exhibited the largest increase in peak loading rates, by 92 % and 220 %, respectively. Ignoring fracture effect could underestimate the mechanical stimulation and risk of fracture non-union. There exists an optimal walking speed for patient-specific BW to promote endochondral ossification, while controlling fibrous tissue formation (e.g., walking at 0.6 v for a 65 kg patient under 30 % PWB). CONCLUSION: This study offers clinically relevant insights to assist physiotherapists in prescribing effective, patient-specific gait rehabilitation strategies to enhance tibial fracture healing during early recovery.
Patient-specific fluid-structure interaction modeling of cerebral aneurysm: influence of wall compliance, tissue prestress, and blood rheology.
Raj A, Gupta R, Singh A
Comput Methods Programs Biomed · 2026 May · PMID 41690260 · Publisher ↗
BACKGROUND AND OBJECTIVE: Cerebral aneurysms are pathological dilations of intracranial arteries that commonly develop at arterial bifurcations. At these locations, hemodynamic forces significantly affect structural prop...
BACKGROUND AND OBJECTIVE: Cerebral aneurysms are pathological dilations of intracranial arteries that commonly develop at arterial bifurcations. At these locations, hemodynamic forces significantly affect structural properties of the vascular walls leading to focal weakening and vessel remodeling. This study aims to evaluate the influence of wall compliance and tissue prestress on aneurysmal hemodynamics and wall mechanics using a fluid-structure interaction (FSI) framework. The effect of shear thinning of blood is also studied. METHODS: The flow of blood and its effect on the vessel walls is modelled in a patient-specific cerebral aneurysm. Physiologically realistic inflow conditions derived from PC-MRI is used as the inlet boundary condition and three-element Windkessel model is used to specify the outlet boundary condition to account for the effect of downstream vasculature. Prestress is applied to the arterial wall to mimic the in-vivo stressed state of the vessel wall. Simulations are performed using the Arbitrary-Lagrangian-Eulerian (ALE) FSI approach under different considerations of wall compliance, blood rheology, and prestress, both individually and in-combination. The computational framework is validated against analytical and numerical solutions available in the literature. RESULTS: Accounting for wall compliance leads to increased inflow into the aneurysm sac and a reduced pressure drop between the inlet and outlet over a cardiac cycle. In the flexible wall model, a single, stable vortex core is observed in the dome instead of the multiple vortices which are observed in case of rigid wall. Further, consideration of flexible walls results in the reduction of peak time-averaged wall shear stress (TAWSS) by ∼20%, reduces the dome area exposed to low TAWSS and regions having high oscillatory shear index (OSI). Including the prestress in model proves critical, as it reduces wall displacement up to 72% and peak tensile stress up to 83% at peak systole. Consideration of shear thinning behaviour of blood further decreases peak TAWSS by up to 25% and reduces area having low TAWSS, but has minimal effect on wall displacement and tensile stress. CONCLUSIONS: Wall compliance, blood rheology, and prestress substantially influence aneurysmal hemodynamics and wall mechanics, with prestress having the most dominant effect in reducing wall deformation and stress.
Corrigendum to "Energy loss minimization-based side branch flow model for FFR calculation based on intracoronary images" [Computer Methods and Programs in Biomedicine 269 (2025) 108872].
Lai X, Xue X, Gao Z … +5 more , Xie B, Zhang H, Yong D, Jia H, Liu X
Comput Methods Programs Biomed · 2026 May · PMID 41678980 · Publisher ↗
Abstract loading — click title to view on PubMed.
DNA-Driven EEG monitoring for rapid seizure prediction in healthcare.
Ansari K, Chaurasia U, Pathak HK … +2 more , Singh KK, Pradhan J
Comput Methods Programs Biomed · 2026 May · PMID 41678979 · Publisher ↗
BACKGROUND AND OBJECTIVE: Worldwide, over 50 million people suffer from epilepsy, a neurological disorder characterised by recurrent seizures due to abnormal electrical activity in the brain. These occur as a result of s...
BACKGROUND AND OBJECTIVE: Worldwide, over 50 million people suffer from epilepsy, a neurological disorder characterised by recurrent seizures due to abnormal electrical activity in the brain. These occur as a result of sudden electric surges and the symptoms vary based on the region of the brain being affected, including brief staring spells and confusion to convulsions and loss of consciousness. Physicians typically classify seizures into four main phases: Interictal, Preictal, Ictal, and Postictal. Accurate analysis of EEG signals around seizure onset is extremely critical for timely clinical intervention. However, the current methodologies majorly utilise complex Convolutional Neural Networks (CNNs) with millions of parameters. They require high computational power, and, hence, it is difficult to deploy them in wearable devices. The core idea of this work is to develop a computationally compact architecture for seizure onset discrimination that offers potential for future integration with wearable devices. METHODS: To achieve this, this work proposes employing a DNA-based encoding framework for Electroencephalogram (EEG) signals. Existing DNA-based compression techniques have demonstrated significant potential in reducing data complexity. Multichannel EEG signals using 23 scalp electrodes are obtained from the CHB-MIT dataset and normalised using min-max scaling. The signals are then windowed to capture temporal dependencies and transformed into integer safe magnitudes before being converted to binary. This approach then involves genetic coding-based preprocessing: genetic transcription and translation (DNA → RNA → Codons → Amino Acids) occur. By converting EEG signal data to amino acid sequences, the proposed encoding scheme aims to capture underlying patterns in the data and provide a compact representation of temporal patterns. The encoded sequences are subsequently processed using a lightweight one-dimensional multi-level parallel CNN architecture. RESULTS AND CONCLUSION: These DNA-encoded EEG sequences are then used as input to the proposed 1D multi-level parallel CNN model, with drastically fewer parameters. After extensive testing, the proposed model achieves an accuracy of 96.22%. Additionally, the applicability of the proposed encoding framework on early seizure prediction tasks under a subject-wise protocol has been evaluated. An accuracy of 93.87% has been achieved. Overall, these findings indicate that the proposed approach provides a compact and effective representation for EEG-based seizure analysis across related onset and early prediction tasks.
Stable EEG source estimation for standardized Kalman filter using rate-of-change tracking.
Lahtinen J
Comput Methods Programs Biomed · 2026 May · PMID 41678978 · Publisher ↗
BACKGROUND AND OBJECTIVE: Localization of brain activity is an important guiding tool for presurgical evaluation and treatment planning, particularly in the context of various brain-related diseases such as epilepsy. Sin...
BACKGROUND AND OBJECTIVE: Localization of brain activity is an important guiding tool for presurgical evaluation and treatment planning, particularly in the context of various brain-related diseases such as epilepsy. Since brain activity is highly dynamic and arises from the firing of neuronal networks, advanced spatiotemporal modeling is needed to capture this time-varying behavior accurately. METHODS: A new parameter tuning and model utilizing the rate-of-change of brain activity distribution were developed to improve the filtering-parametrization-stability of the otherwise accurate estimation of the recently introduced Standardized Kalman filter. Namely, we propose a backward-differentiation-based measurement model for the rate of change. Simulated and real non-invasive electroencephalography data, along with realistic head models from two real subjects, were used in time-evolution tracking and localization experiments focusing on somatosensory evoked potentials. The method was compared to the original Standardized Kalman filter and Standardized Low-resolution Brain Electromagnetic Tomography (sLORETA). RESULTS: Results indicate that the proposed parametrization yields high localization accuracy, as the original Standardized Kalman filtering localizes 7 and 6 out of 8, and sLORETA found 8 and 6 of the literature-defined originators of short latency somatosensory evoked potentials. The proposed standardized filtering method identified 7 of 8 expected originators. The change-rate-based model exhibits greater tracking stability than filtering without it against changes in filtering parameters. In addition, the method is more stable against badly set model parameters. CONCLUSIONS: With new model parametrization, the studied standardized methodologies provide assumably accurate and stable estimations to explore the location and dynamical properties of cortical and subcortical brain activity. The results showing correct sub-thalamic localization demonstrate the significant potential of these methods in the guidance of stereo-electroencephalography sensor placements or the placement of implant electrodes for deep-brain stimulation.
A computational approach for classification of HIV drug resistance based on the self-consistent extreme classifier.
Stolbov LA, Rudik AV, Stolbova EA … +7 more , Pokrovskaya AV, Shemshura AB, Kireev DE, Lagunin AA, Filimonov DA, Poroikov VV, Tarasova OA
Comput Methods Programs Biomed · 2026 May · PMID 41671705 · Publisher ↗
BACKGROUND AND OBJECTIVES: The development of viral resistance can significantly reduce the effectiveness of therapy. Human immunodeficiency virus type 1 is the cause of chronic immune dysfunction, leading to the develop...
BACKGROUND AND OBJECTIVES: The development of viral resistance can significantly reduce the effectiveness of therapy. Human immunodeficiency virus type 1 is the cause of chronic immune dysfunction, leading to the development of co-infections and serious complications. Despite worldwide progress and consolidated efforts to overcome HIV drug resistance, the development of novel approaches for rational drug therapy of HIV infection is still needed for building models with high accuracy of prediction and that can be applied for evaluation of resistance against wide variety of inhibitors. Our study is dedicated to the development of a novel computational ML-driven approach for the ternary classification of HIV protease, reverse transcriptase, and integrase sequences. Binary classification approaches naturally are not applicable to capture clinically important intermediate resistance levels, motivating the use of a ternary classification model. METHODS: For the model development we used the Self-Consistent Extreme Classifier. One-versus-rest and one-versus-one ternary approaches were applied to sequences related resistance data from Stanford University HIV Drug Resistance Database (StDB). RESULTS: For the final classifiers we selected the most appropriate models with 0.913 sensitivity, 0.894 specificity, 0.741 precision and 0.953 area under ROC, all values provided in average. We tested our approach in a clinical task and performed prospective validation for eight sequences of HIV protease and reverse transcriptase obtained from treatment-naive HIV-positive male patients. We performed a prediction and compared the results with the therapeutic outcome, in particular, with the viral load decline at 24 weeks. CONCLUSIONS: The results of the prospective validation are generally consistent with the results of the therapeutic outcome and confirm the possibility of using the developed approach for the selection of the most appropriate therapeutic regimens.
Advancing the vision of "reliability metadata": From conceptual refinement to clinical validation.
Yu Z, Cheng W
Comput Methods Programs Biomed · 2026 May · PMID 41666594 · Publisher ↗
Abstract loading — click title to view on PubMed.
Artificial intelligence approaches for non-invasive diabetes prediction using ECG signals: A systematic review.
Balakrishnan K, Velusamy D, Ramasamy K … +4 more , Hinkle HE, Hudson HJ, Pachori RB, Khan H
Comput Methods Programs Biomed · 2026 May · PMID 41643489 · Publisher ↗
Diabetes is a major global health challenge, with many individuals remaining undiagnosed due to the limitations of traditional screening methods. Artificial intelligence (AI)-based electrocardiogram (ECG) analysis offers...
Diabetes is a major global health challenge, with many individuals remaining undiagnosed due to the limitations of traditional screening methods. Artificial intelligence (AI)-based electrocardiogram (ECG) analysis offers a promising, non-invasive approach for the early detection of diabetes. This systematic review aims to critically evaluate machine learning (ML) and deep learning (DL) models developed for non-invasive prediction of diabetes and prediabetes using ECG signals. A comprehensive literature search was conducted across PubMed, Embase, Web of Science, IEEE Xplore, and ACM Digital Library in accordance with PRISMA 2020 guidelines. Twenty-five studies met the inclusion criteria. Extracted data included ECG input types, model architectures, preprocessing methods, feature sets, validation strategies, and performance metrics. Most studies used small, single-site, cross-sectional datasets, with sample sizes ranging from 24 to over 190,000 individuals. ECG preprocessing methods varied widely, including filtering, normalization, and decomposition. Features were extracted from time, frequency, morphological, and non-linear domains, though formal feature selection was applied inconsistently. ML and DL models reported high internal accuracy (>90%) but most lacked external validation and subgroup performance assessments. Notably, no study specifically focused on rural or underserved populations, and only one provided open-source code. AI-based ECG analysis demonstrates strong potential for detecting diabetes; however, current research is limited by generalizability issues, lack of standardized methods, poor external validation, and insufficient transparency. Future studies should prioritize rigorous validation, reproducibility, fairness audits, and applications in rural and underserved settings to ensure equitable and clinically viable deployment of these models.
Detecting optimal biomarkers in ovarian cancer cells from high-dimensional mRNA expression data using machine learning.
Thelagathoti RK, Jiang C, Chandel DS … +5 more , Tom WA, Sarmiento C, Krzyzanowski G, Olou A, Fernando MR
Comput Methods Programs Biomed · 2026 May · PMID 41643488 · Publisher ↗
BACKGROUND AND OBJECTIVE: Reliable detection of robust biomarkers from high-dimensional transcriptomic data remains a major challenge in computational oncology. Traditional approaches often suffer from overfitting and po...
BACKGROUND AND OBJECTIVE: Reliable detection of robust biomarkers from high-dimensional transcriptomic data remains a major challenge in computational oncology. Traditional approaches often suffer from overfitting and poor generalization due to the high dimensionality of genomic data and limited sample sizes. This study aims to identify an optimal, biologically meaningful subset of mRNA biomarkers capable of distinguishing ovarian cancer samples from healthy controls using an integrated machine learning-based feature selection framework. METHODS: We analyzed mRNA expression data encompassing approximately 63,000 transcripts from ovarian cancer and control samples derived from cell lines. A hybrid feature selection pipeline combining statistical filtering, recursive elimination, and regularization was implemented under stratified cross-validation to derive stable biomarkers. Model validation was performed using Logistic Regression, Random Forest, XGBoost, and Support Vector Machine classifiers, while experimental validation was conducted through droplet digital PCR (ddPCR). Statistical analyses included ANOVA, t-tests, and pathway enrichment. RESULTS: The pipeline identified 80 discriminative mRNA biomarkers with exceptionally high classification performance (accuracy = 1.00, sensitivity = 1.00, specificity = 1.00 for top models). ddPCR confirmed consistent expression patterns, with significant downregulation of ADAMTS12, FN1, and ABI3BP and overexpression of EPCAM, COX6C, and TMT1B in ovarian cancer. Pathway enrichment revealed involvement in DNA repair, RNA processing, protein translation, immune regulation, and metabolic reprogramming. CONCLUSIONS: This hybrid feature selection framework applied to patient derived cell lines, effectively reduces dimensionality, enhances biomarker reliability, and uncovers biologically interpretable mRNA signatures associated with ovarian cancer, demonstrating potential for diagnostic and therapeutic applications.
HCAR1 antagonist screening based on boundary-selected negative sampling strategy and multi-level graph neural network.
Fan M, He D, Liu Q … +7 more , Liu Q, Wang F, Li H, Wang H, Shan S, Zhang J, Hou Y
Comput Methods Programs Biomed · 2026 May · PMID 41633067 · Publisher ↗
BACKGROUND AND OBJECTIVE: Hydroxycarboxylic acid receptor 1 (HCAR1), also known as the lactate receptor, is closely associated with tumorigenesis and cancer progression due to its aberrant activation, making it an attrac...
BACKGROUND AND OBJECTIVE: Hydroxycarboxylic acid receptor 1 (HCAR1), also known as the lactate receptor, is closely associated with tumorigenesis and cancer progression due to its aberrant activation, making it an attractive therapeutic target for cancer treatment. Accurate prediction of HCAR1 antagonists is therefore crucial for tumor immunotherapy. However, traditional drug screening suffers from high costs and suboptimal performance caused by imbalanced datasets and incomplete molecular representations, contributing to the scarcity of clinically available HCAR1 antagonists. METHODS: A balanced HCAR1 target activity dataset was constructed using a boundary-selected negative sampling strategy. Subsequently, a multi-level graph neural network (Multi-GNN) was proposed for HCAR1 target activity prediction, integrating multiple molecular representations, including fingerprints, molecular graphs, and fragment-level features. RESULTS: Experimental results demonstrate that the proposed model outperforms eight state-of-the-art methods in comparative evaluations. Furthermore, approximately ten million compounds were screened using the trained Multi-GNN model in combination with physicochemical filtering and molecular docking, yielding five candidate compounds. Finally, in vitro cAMP antagonistic activity assays identified a promising HCAR1 inhibitor with an IC of 22.39 μM. CONCLUSIONS: This study introduces a novel artificial intelligence-based framework for HCAR1-targeted drug discovery and highlights potential lead compounds for further development.
Correlative analysis between ocular surface features and carotid plaque : A multimodal machine learning framework.
Zhang S, Hu D, Luo L … +1 more , Cao J
Comput Methods Programs Biomed · 2026 May · PMID 41621344 · Publisher ↗
BACKGROUND AND OBJECTIVE: The diagnosis of carotid plaques plays an important role in revealing cardiovascular and cerebrovascular diseases, thus attracting widespread research attention. However, most medical examinatio...
BACKGROUND AND OBJECTIVE: The diagnosis of carotid plaques plays an important role in revealing cardiovascular and cerebrovascular diseases, thus attracting widespread research attention. However, most medical examinations rely heavily on specialists and carotid ultrasound images, which are time-consuming, radiative, expensive and limited in tracking disease progression. To alleviate these deficiency, inspired by the human blood supply sequence, a detailed study on the association between carotid plaque and ocular surface image features is proposed in the paper. METHODS: This paper systematically verifies the correlation between carotid plaque and ocular surface image through a multi-dimensional feature analysis approach incorporating texture, frequency domain features, and color characteristics. The analysis combines feature selection, confidence evaluation, and distribution property studies to establish robust associations. Besides, multiple machine learning classifiers are used to evaluate the robustness of the extracted features, with subgroup validation conducted across different subsets, systematically assessing the influence of age and gender factors. RESULTS: The proposed method achieves high prediction accuracy on 8875 individuals from Hangzhou Wuyunshan Hospital (Hangzhou Institute for Health Promotion), with electronic health record (EHR) features showing the strongest association (Odds Ratios [ORs]: 4.35 [3.90-4.86] in males; 2.92 [2.60-3.27] in females). Experimental results demonstrate that age, male gender, and ocular surface image features - including EHR, local binary patterns (LBP), gray-level gradient co-occurrence matrix (GLGCM), and gray-level co-occurrence matrix (GLCM) - show strong associations with carotid plaque, where LBP and EHR features are selected most frequently. CONCLUSIONS: Ocular surface image analysis offers a practical and non-invasive method for carotid plaque screening. The observed feature associations and strong predictive performance highlight its potential for clinical applications, especially in large-scale population screening.
Decoding metabolic reprogramming heterogeneity across bladder cancer stages using single-cell and spatial multi-omics approaches.
Zhao J, Zhao J, Huang W … +6 more , Lin W, Jie K, Wu Z, Li B, Fan L, Wang X
Comput Methods Programs Biomed · 2026 Apr · PMID 41621229 · Publisher ↗
BACKGROUND: Reprogramming of metabolic pathways represents a central indicator in cancer pathogenesis, but the metabolic heterogeneity of bladder cancer (BLCA) at different stages is not well understood. This study aims...
BACKGROUND: Reprogramming of metabolic pathways represents a central indicator in cancer pathogenesis, but the metabolic heterogeneity of bladder cancer (BLCA) at different stages is not well understood. This study aims to analyze metabolic reprogramming in BLCA across stages and its impact on patient survival. METHODS: Single-cell sequencing data were used to examine metabolic heterogeneity of epithelial cells and cell subpopulation differentiation in BLCA at various clinical stages. Spatial transcriptome data were analyzed for copy number variability and riboflavin metabolism in BLCA epithelial cells. Bulk RNA sequencing data from BLCA patients were used for riboflavin pathway expression analysis and prognostic biomarker identification. The effects of three biomarkers (ENPP1, ACP1, and RFK) on BLCA risk were validated using RT-qPCR, Mendelian randomization and co-localization analysis. RESULTS: Epithelial cells exhibited significant metabolic heterogeneity during bladder cancer progression. Compared to normal control stage, riboflavin metabolic activity progressively increased with disease stage, as validated by spatial transcriptomics and bulk RNA-seq. High expression of ENPP1, ACP1, and RFK (riboflavin pathway) strongly correlated with poor overall survival. RT-qPCR confirmed their high expression in tumours, increasing with stage. Mendelian randomisation/co-localisation indicated these genes localise to bladder epithelium, and their genetic variation associates negatively with BLCA risk. CONCLUSION: Increased riboflavin metabolism is likely to be an important marker of malignant progression in BLCA. The ENPP1, ACP1 and RFK genes in this pathway may serve as valuable prognostic biomarkers for BLCA, with potential implications for early diagnosis, monitoring disease progression, and guiding personalized treatment strategies.
Application of artificial intelligence in colonoscopy imaging for polyp analysis-A systematic review.
Amirmohammadi E, Shalbaf A, Esteki A … +4 more , Sadeghi A, Moghadam AR, Moghtaderi M, Moghadam PK
Comput Methods Programs Biomed · 2026 Apr · PMID 41621228 · Publisher ↗
The colon is a major component of the digestive system, so early detection of colorectal polyps is essential in preventing colorectal cancer, which is a leading cause of cancer-related death worldwide. While colonoscopy...
The colon is a major component of the digestive system, so early detection of colorectal polyps is essential in preventing colorectal cancer, which is a leading cause of cancer-related death worldwide. While colonoscopy remains the gold standard for polyp detection, its diagnostic accuracy is highly operator-dependent. Recent advances in Deep Learning (DL), a branch of Artificial Intelligence (AI), have shown substantial potential to improve colonoscopy image analysis by enhancing the accuracy, consistency, and objectivity of polyp detection, segmentation, and classification. Artificial intelligence-based systems have significantly reduced inter-observer variability and increased diagnostic efficiency, ultimately transforming the landscape of colorectal lesion assessment. This survey provides a comprehensive and critical analysis of the current status of deep learning applications in colorectal polyp analysis. We systematically review state-of-the-art methodologies across various DL architectures-including Convolutional Neural Networks (CNNs), transformer-based models, and hybrid approaches-and examine their performance on publicly available benchmark datasets. Additionally, we highlight the strengths and limitations of existing techniques, explore the clinical relevance of AI-assisted tools, and identify prevailing challenges such as data imbalance, real-time deployment, and generalizability across diverse populations and colonoscopy devices. By consolidating key advances and outlining future research directions, this review aims to serve as a valuable resource for researchers, clinicians, and developers seeking to leverage deep learning to enhance colorectal polyp detection, diagnosis, and clinical decision-making.
Explainable reinforcement learning for glucose monitoring based on shapley value analysis.
Adjevi A, Abdirashid AM, Aktaş F … +2 more , Ucar MHB, Solak S
Comput Methods Programs Biomed · 2026 May · PMID 41619710 · Publisher ↗
BACKGROUND AND OBJECTIVE: Effective diabetes management requires continuous regulation of blood glucose in response to complex factors such as diet, activity, stress, and medication. Advances in continuous glucose monito...
BACKGROUND AND OBJECTIVE: Effective diabetes management requires continuous regulation of blood glucose in response to complex factors such as diet, activity, stress, and medication. Advances in continuous glucose monitoring and machine learning have improved short-term glucose prediction. However, preprocessing of signals like insulin, carbohydrate intake, heart rate, and activity to better capture metabolic dynamics remains underexplored. Similarly, the integration of predictive models with preventive strategies for guiding interventions is still limited. METHODS: We propose a research-only decision-support framework combining signal preprocessing, CNN-based glucose prediction, Shapley Additive Explanations (SHAP) values attribution, and an Actor-Critic Reinforcement Learning (RL) agent. Exponential decay models preprocess inputs, a compact CNN forecasts short-term glucose levels, and SHAP values highlights the most influential input features; however, these attributions reflect associative patterns in the data and do not establish or map to causal clinical mechanisms. These SHAP-derived attributions guide the RL agent, which issues bounded one-step behavioral adjustments. Because SHAP-guided RL remains stochastic and uncertain, the proposed system is exploratory and not clinically safe, serving solely as a simulation framework. RESULTS: Using the OhioT1DM dataset, the model achieved state-of-the-art RMSE across prediction horizons with a compact size of 7̃4 KB per patient and training under one minute for 1000 epochs. Over 98% of predictions fell within Clarke Error Grid Zones A and B, confirming safe 5-20 min forecasts. The preventive component corrected hyper- and hypoglycemia in 2̃5% of cases within 10 min when predictions were near 80-120 mg/dL (±10 mg/dL). When deviations exceed ±10 mg/dL, the RL agent is unable to fully restore blood glucose to the target range within 10 min but can bring it as close as possible to the defined interval. CONCLUSIONS: This study presents a significant innovation by bridging predictive accuracy, adaptability, and transparency in diabetes management. The integration of a predictive model with Reinforcement Learning (RL) guided by SHAP values, which are typically used for interpretability but here are employed in the learning process, delivers a powerful decision support framework. This approach advances the field toward next-generation, personalized digital health tools.
Enhancing accuracy and explainability in colorectal lesion classification with attention-supervised Vision Transformers.
Carlini L, Di Stefano L, Lena C … +4 more , Massimi D, Rizkala T, Hassan C, De Momi E
Comput Methods Programs Biomed · 2026 Apr · PMID 41581326 · Publisher ↗
OBJECTIVE: Accurate assessment of colorectal lesion morphology during colonoscopy is essential for guiding treatment and estimating cancer risk. The Paris classification is widely adopted for this purpose but suffers fro...
OBJECTIVE: Accurate assessment of colorectal lesion morphology during colonoscopy is essential for guiding treatment and estimating cancer risk. The Paris classification is widely adopted for this purpose but suffers from substantial inter-observer variability, while Vision Transformers (ViTs) can base their decisions on diffuse, off-lesion attention patterns that are hard to interpret. This study investigates whether directly supervising ViT attention maps with expert lesion annotations can concurrently improve Paris classification performance and model explainability. METHOD: We propose a Lesion-Focused Attention Loss (L), an attention-supervised pretraining objective that uses expert polyp bounding boxes to focus last-layer [CLS] attention on annotated lesion regions, followed by standard cross-entropy fine-tuning. L is applied to six ViT architectures and evaluated on the public SUN dataset for binary (0-I vs. 0-II) and three-class (0-Ip, 0-Is, 0-IIa) Paris classification. Performance is assessed using frame-wise accuracy and the AttIn, we additionally perform an ablation study against a Grad-CAM consistency baseline. RESULTS: Attention-supervised pretraining yields consistent gains in both accuracy and lesion-focused attention. Across the six ViTs, adding L improves three-class accuracy by up to 7 percentage points. In a detailed ablation on ViT-B/16, L outperforms a Grad-CAM consistency baseline by about 5-13 percentage points across the 2-class and 3-class tasks, and χ tests confirm a significant association between high AttIn and correct predictions. CONCLUSION: Direct supervision of ViT attention with L leverages expert knowledge to jointly boost Paris classification accuracy and spatial interpretability, and compares favourably with Grad-CAM-based explanation regularisation. The source code and dataset splits are publicly available at https://github.com/LucaCarlini/SUNDatasetPretraining.
Effect of Doppler ultrasound and high-frame-rate ultrasound particle image velocimetry derived inlet boundary conditions on wall shear stress parameters in the stented superficial femoral artery.
Rutten L, van de Velde L, Pol L … +3 more , Jain K, Reijnen MMPJ, Versluis M
Comput Methods Programs Biomed · 2026 Apr · PMID 41581325 · Publisher ↗
BACKGROUND AND OBJECTIVES: Hemodynamic predictions by computational fluid dynamics (CFD) strongly depend on inlet boundary conditions (IBC). One-dimensional Doppler ultrasound (DUS) is typically used for estimating flow...
BACKGROUND AND OBJECTIVES: Hemodynamic predictions by computational fluid dynamics (CFD) strongly depend on inlet boundary conditions (IBC). One-dimensional Doppler ultrasound (DUS) is typically used for estimating flow IBCs, despite its sensitivity to the operator, ultrasound hardware and assumptions in flow rate computation. An alternative is two-dimensional high-frame-rate ultrasound particle image velocimetry (echoPIV). This study investigated the differences between DUS and echoPIV-derived IBCs and their effect on wall shear stress parameters in the stented superficial femoral artery. METHODS: CFD simulations using DUS and echoPIV-derived IBCs were performed for three patients with a superficial femoral artery stenosis that were treated with a stent. Spatiotemporal velocity profiles were compared at 0 - 50 mm from the inlet. Differences were quantified with the root-mean-square error (RMSE). Regions of low time-averaged wall shear stress (TAWSS) and high oscillatory shear index (OSI) using a literature-based threshold of 0.4 Pa and 0.2, respectively, and an IBC-specific threshold (lower third and upper third, respectively) were determined. Co-localization was quantified using the Jaccard similarity index. RESULTS: The DUS and echoPIV-derived IBCs differed in flow rate and velocity profile, with the largest difference found at peak systole (RMSE: > 50 cm/s). Using the literature-based threshold, similarity in low TAWSS was high for two patients (0.85 - 0.88) and low for one (0.57). Agreement in high OSI was low in two patients (0.45 - 0.48) and high in one patient (0.83). The IBC-specific threshold increased the agreement for both low TAWSS and high OSI (≥0.75). CONCLUSIONS: Differences in DUS and echoPIV-derived IBCs affected the TAWSS and OSI magnitudes. Regions of low TAWSS and high OSI corresponded well using an IBC-specific threshold. The literature-based threshold resulted in lower similarity values and different interpretations of restenosis risk that may cause differences in follow-up intensity or medical management.
Robust multimodal mental workload classification: A cross-physiological condition machine learning approach.
Pontiggia A, Quiquempoix M, Fabries P … +10 more , Beauchamps V, Jacques C, Guillard M, Van Beers P, Malle C, Gomez-Merino D, Koulmann N, Chennaoui M, Sauvet F, HYPSOM Investigator Group
Comput Methods Programs Biomed · 2026 Apr · PMID 41576780 · Publisher ↗
BACKGROUND AND OBJECTIVE: Aircraft pilots can be faced with a high mental workload (MW) combined with moderate hypoxia and sleep restriction. We aimed to assess the cross-validation of a machine learning-based MW predict...
BACKGROUND AND OBJECTIVE: Aircraft pilots can be faced with a high mental workload (MW) combined with moderate hypoxia and sleep restriction. We aimed to assess the cross-validation of a machine learning-based MW predictive model under hypoxia and/or sleep restriction. Secondly, we developed a robust predictive model using multimodal physiological parameters to improve the validity across different physiological conditions. METHODS: Seventeen healthy participants were randomly exposed to three 12-minute periods of increased MW (low, medium, and high) in a 4-condition crossover design: sleep restriction (SR, <3 h Total Sleep Time, TST) vs. habitual sleep (HS, >6 h TST), hypoxia (HY, 2 h, FO=13.6%, ∼3500 m) vs. normoxia (NO, FO=21%). MW levels were designed using the Multi-Attribute Test Battery (MATB)-II with an additional auditory Oddball-like task. Six machine learning classifiers were compared. Features selection (from EEG, ECG, respiratory and eye tracking sensors) was performed using backward Recursive Feature Elimination (RFE). RESULTS: The best models for 1-minute MW levels classification on HSNO were K-Nearest Neighbors (KNN, F1 score = 80.3 ± 8.9%), Support Vector Machine (SVM, 77.8 ± 10.3%) and Random Forest (RF, 75.7 ± 9.1%). Exposure to sleep restriction and/or hypoxia decreased models' performance (F1 <35%). KNN and RF models, in particular those including EEG and eye tracking, trained on All-Conditions performed well across conditions (F1 scores = 77.4 ± 7.8% and 70.7 ± 10.2%). CONCLUSION: Our results highlight the need for training MW models under different physiological constraints and using multimodal datasets to improve robustness. (NCT05563688).
CellViT: Energy-efficient and adaptive cell segmentation and classification using foundation models.
Hörst F, Rempe M, Becker H … +3 more , Heine L, Keyl J, Kleesiek J
Comput Methods Programs Biomed · 2026 Apr · PMID 41576779 · Publisher ↗
BACKGROUND AND OBJECTIVE: Deep learning-based cell segmentation and classification methods in digital pathology are critical for diagnostics but are hampered by models that require extensive annotated datasets, are compu...
BACKGROUND AND OBJECTIVE: Deep learning-based cell segmentation and classification methods in digital pathology are critical for diagnostics but are hampered by models that require extensive annotated datasets, are computationally expensive, and lack adaptability to new cell types. This creates a significant bottleneck in research and clinical workflows. This study introduces CellViT, a data-efficient and lightweight framework for generalized cell segmentation that allows for rapid adaptation to novel cell taxonomies with minimal data. METHODS: CellViT leverages a Vision Transformer with a frozen pretrained foundation model for segmentation. It simultaneously extracts deep cell embeddings from the transformer tokens during the forward pass at no extra computational cost. To adapt to new cell types, only a lightweight classifier is trained on these embeddings, bypassing the need to retrain the segmentation model. We also demonstrate an automated workflow to generate training data from registered H&E and immunofluorescence (IF) slides. The framework was validated on seven public datasets. RESULTS: The framework achieves remarkable zero-shot segmentation results and data efficiency. On the CoNSeP dataset for colon cancer, we achieved superior results with only 10% of the training data. On all other datasets, we outperformed competing methods or at least approached their performance, all in one model. The classifier approach, based on zero-shot segmentation models, drastically reduces computational costs, with training times of minutes versus hours for baseline models, decreasing CO emission by 96.93%. Models trained on automatically generated labels from IF-staining achieved performance comparable to (lymphocytes, ΔF:-0.042) or even exceeding (plasma cells, ΔF:+0.108) those trained on expert-annotated datasets. CONCLUSIONS: CellViT provides a robust and efficient open-source framework that addresses key limitations in computational pathology by decoupling segmentation from classification. Its ability to adapt to new cell types with minimal data and its support for automated dataset generation from IF slides significantly reduce the reliance on time-consuming expert annotation. This work provides a foundational tool to accelerate research, enhance diagnostic workflows, and enable deeper cohort analysis. The code is available under https://github.com/TIO-IKIM/CellViT-plus-plus and as a PyPI package.
In silico modelling of aortic annuloplasty: hemodynamic assessment through in vitro experiments and in vivo MRI.
Zattoni M, Bontempi L, Ringgaard S … +4 more , Luraghi G, Benhassen LL, Johansen P, Colombo M
Comput Methods Programs Biomed · 2026 Apr · PMID 41576778 · Full text
Aortic annuloplasty (AA) is an innovative surgical technique for aortic root (AR) enlargement. It is performed by implanting sutures, bands, or rings, either externally or internally the AR, hereby reducing its diameter....
Aortic annuloplasty (AA) is an innovative surgical technique for aortic root (AR) enlargement. It is performed by implanting sutures, bands, or rings, either externally or internally the AR, hereby reducing its diameter. This study evaluates the impact of AA approaches on AR hemodynamic by employing a porcine-specific workflow combining in vivo magnetic resonance imaging (MRI), in vitro experiments and in silico fluid-structure interaction (FSI) simulations investigating external single ring AA. CAD models of native and post-annuloplasty ARs were segmented from in vivo porcine MRI data and served as the basis for fabricating 3D-printed resin phantoms and implementing computational digital twins. The former were tested on a pulsatile flow-loop, whereas the latter were integrated in FSI simulations, with time-dependent boundary conditions based on the resultant experimental pressure waveforms. Additionally, a proof-of-concept validation of the in silico model against in vivo data is proposed. Computational results of the two cases were compared in terms of fluid velocity, vorticity, helicity, and wall shear stresses, providing a step towards understanding the complex interactions between the AR and blood flow dynamics. Results suggested that the presence of the ring increased the systolic jet flow and post-valve velocities (three-fold increase), reduced the backward, vortical flow during diastole (∼ 9% decrease), and induced modifications in bulk flow and wall shear stresses distribution. Furthermore, the development of an animal-specific digital twin of a post-AA AR represents a significant advancement in the field, providing a valuable tool for future research and for clinical applications to aid AA decision-making process.
A diagnosis tool for early detection and classification of heart disease in individuals using transformer mechanisms.
Arif S, Son SH, Kim HY … +2 more , Kim SC, Lee JY
Comput Methods Programs Biomed · 2026 Apr · PMID 41576777 · Publisher ↗
BACKGROUND: Heart disease remains a leading cause of mortality, making accurate and efficient prediction tools essential for the general population. In medical diagnosis, deep learning-based approaches have shown signifi...
BACKGROUND: Heart disease remains a leading cause of mortality, making accurate and efficient prediction tools essential for the general population. In medical diagnosis, deep learning-based approaches have shown significant potential in identifying complex patterns within clinical data. METHODS: This paper proposes a transformer-based model, named the Heart-Doctor for self-detection and classification of heart disease in the general population, addressing the critical challenge of early diagnosis and real-life risk assessment. The proposed model, inspired by the transformer architecture, processes 16 key attributes through a multi-layer transformer encoder, leveraging residual connections and deep feature extraction for improved classification. The proposed method utilizes electronic health records (EHR) from Chungbuk National University Hospital (CBNUH), incorporating symptom-based features along with common clinical attributes such as diagnoses and medical history to enhance predictive performance. The proposed model's effectiveness is evaluated by implementing and comparing it with Convolutional Neural Network (CNN) and Random Forest (RF) classifiers. RESULTS: Experimental results demonstrate that the proposed model achieves 99% accuracy, compared to RF (98.79%) and CNN (97.64%), with superior precision, recall, and F1-scores, making it a highly effective tool for multi-class heart disease classification. To ensure real-world applicability, the study also includes the development of an Android-based application that integrates the model for real-time risk assessment, thereby enabling healthcare professionals to make timely and data-driven clinical decisions. CONCLUSION: Consequently, the performance of the proposed transformer-based diagnosis model outperforms other models for heart disease classification, allowing individuals to detect their heart disease symptoms early and independently in real life.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe