Computers in biology and medicine最新文献

{"title":"Machine learning guided structural dynamics identifies translation elongation factor 1 (EEF1A1) as an immunological biomarker and marine natural products as therapeutic leads for rheumatoid arthritis with major depressive disorder","authors":"Santhiya Panchalingam ,&nbsp;Govindaraju Kasivelu ,&nbsp;Manikandan Jayaraman ,&nbsp;Jeyakanthan Jeyaraman","doi":"10.1016/j.compbiomed.2026.111480","DOIUrl":"10.1016/j.compbiomed.2026.111480","url":null,"abstract":"<div><div>Rheumatoid arthritis (RA) is a systemic autoimmune disease that predominantly affects synovial joints, especially those of the hands, elbows, wrists, knees, and shoulders. RA frequently co-occurs with major depressive disorder (MDD), amplifying disease burden and complicating clinical outcomes. This study employed a multi-step integrative bioinformatics and structural biology framework to identify candidate molecular biomarkers for RA and MDD. Differential gene expression analysis and weighted gene co-expression network analysis (WGCNA) were performed on the epitranscriptomic dataset. These analyses identified immune-regulatory gene modules that were significantly associated with both phenotypes. Least absolute shrinkage and selection operator (LASSO) regression was applied to select strong, statistically significant biomarkers. The methylated biomarker EEF1A1 was identified, and its structure predicted via AlphaFold, was subjected to <em>in silico</em> structure-based virtual screening (SBVS) against the Comprehensive Marine Natural Product Database (CMNPD). Four marine natural products (CMNPD17984, CMNPD27318, CMNPD26200, and CMNPD26011) showed significant binding affinity for EEF1A1. Furthermore, EEF1A1-MNP complexes were simulated for 150 ns using GROMACS, and PCA-based free energy landscape (FEL) analyses were performed to characterize the dynamic behavior and identify energy minima. This integrated computational approach provides a comprehensive platform for biomarker discovery and validation in RA and MDD, with potential applications in early diagnosis, therapeutic targeting, and precision medicine.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111480"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dementia severity index: A threshold-based approach to classifying dementia levels using resting state EEG","authors":"Shivani Ranjan ,&nbsp;Robin Badal ,&nbsp;Pramod Yadav ,&nbsp;Lalan Kumar","doi":"10.1016/j.compbiomed.2026.111505","DOIUrl":"10.1016/j.compbiomed.2026.111505","url":null,"abstract":"<div><h3>Background</h3><div>Alzheimer’s Disease (AD) and FrontoTemporal Dementia (FTD) are dementia conditions that often overlap clinically, leading to misdiagnoses. Traditional questionnaires are subjective and time-intensive, while neuroimaging is costly and less accessible. EEG-based methods offer a cost-effective alternative but primarily focus on spectral and source analyses, with a limited exploration into quantitative range identification for differentiating dementia states.</div></div><div><h3>Methods</h3><div>This study presents a threshold-based approach to dementia-level classification using resting-state EEG. In particular, an algorithm is presented for threshold computation followed by Dementia Severity Index (DSI) formulation. Two potential biomarkers for cognitive decline that capture band-specific alterations are explored. These biomarkers form the basis of the DSI, categorizing individuals into AD, FTD, or Healthy Control (HC). The classification performance of the proposed DSI is evaluated comprehensively using multiple machine learning classifiers and subject validation strategies.</div></div><div><h3>Results</h3><div>The proposed DSI-based approach achieves classification accuracies of 81.62% using kNN. The approach reliability is validated across three diverse EEG datasets and through threshold variation analysis. Furthermore, the relationship between EEG features and cognitive performance is analyzed using Spearman’s correlation. A significant correlation of 0.79 and 0.62 is obtained between predicted and actual MMSE.</div></div><div><h3>Conclusion</h3><div>The proposed DSI effectively differentiates AD, FTD, and HC, providing a robust threshold-based framework for dementia assessment. It enhances interpretability by assigning quantitative values to dementia states and reduces subjective reliance. This study offers a potential EEG-based biomarker suitable for clinical settings, offering minimal stress to patients during assessments.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111505"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075229","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transformer-based feature extraction approach for hematopoietic cancer subtype classification","authors":"Kwang Ho Park ,&nbsp;Younghee Lee ,&nbsp;Wei Ding ,&nbsp;Kwang Sun Ryu ,&nbsp;Keun Ho Ryu","doi":"10.1016/j.compbiomed.2026.111466","DOIUrl":"10.1016/j.compbiomed.2026.111466","url":null,"abstract":"<div><div>Accurate classification of hematopoietic cancer subtypes remains challenging due to the multipotent nature of hematopoietic cells and the absence of definitive genetic markers. To address this, we propose a Transformer-based Autoencoder that captures compact and biologically informative embeddings from gene expression data. Specifically, our method employs multi-head self-attention in the encoder to learn complex nonlinear interactions among genes, with a reconstruction decoder that enforces biological feature retention. We benchmarked our approach against four widely-used feature extraction methods—Principal Component Analysis, Non-negative Matrix Factorization, Autoencoder, and Variational Autoencoder—using transcriptomic data from five hematopoietic cancer subtypes in The Cancer Genome Atlas, totaling 2452 samples. Data were split 60:20:20 into training, validation, and test sets with stratification, and feature-extractor hyperparameters were chosen on the validation set<strong>.</strong> Each method produced 100-dimensional feature vectors, subsequently evaluated using eight multi-class classifiers: Light Gradient Boosting Machine, Extreme Gradient Boosting, Logistic Regression, Random Forest, Decision Tree, Support Vector Machine, and Neural Networks. On the independent test set, the Transformer-based Autoencoder embeddings combined with Light Gradient Boosting Machine achieved F1-score: 0.969, accuracy: 0.986, precision: 0.975, recall: 0.964, specificity: 0.996, G-mean: 0.980, and balanced accuracy: 0.954. For context, we additionally included a supervised tabular Transformer (FT-Transformer) as a reference; while strong, it is not directly comparable to our unsupervised feature extractor. To enhance interpretability and clinical relevance, we applied Shapley Additive exPlanations to identify the twenty most influential genes contributing to subtype discrimination. This analysis revealed key biomarkers related to endoplasmic reticulum function, antigen processing, and ribonucleic acid regulation. These findings demonstrate that transformer-based unsupervised feature extraction substantially improves predictive accuracy and yields valuable biological insights for complex hematologic malignancies. Overall, the study supports attention-driven representation learning for tabular biomedical data and motivates future work in generative/self-supervised representations for gene expression.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111466"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146009068","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"IntNet: Lightweight yet high-performance deep learning system for intuitive radar patterns analysis and human fall detection","authors":"Malek Y. Almallah ,&nbsp;Belal H. Sababha","doi":"10.1016/j.compbiomed.2026.111485","DOIUrl":"10.1016/j.compbiomed.2026.111485","url":null,"abstract":"<div><div>The growing trend of solitary living among the elderly and young, coupled with the high risk of falls leading to injuries and death, highlights the need for fall monitoring systems. Emphasizing individuals' privacy and comfort, these systems should rely on radar sensors instead of visual-based, acoustic-based, or wearable solutions. Current radar-based systems have yet to reach satisfactory real-world performance. This work proposes a radar-based fall detection system that offers superior performance in complex real-world scenarios while maintaining edge computing capabilities and utilizing minimal hardware resources. The proposed deep learning system achieved a recall of 98.99 % and a precision of 99.32 %. These unprecedented performance numbers are measured on the proposed dataset, which is the most real-life representative dataset in the literature. The system has 211.8k parameters and ∼8.84 M Floating Point Operations (FLOPs), achieving an edge computing capability. Moreover, the efficient model construction eliminates redundant computation in real-time operation. Furthermore, this work proposes a novel performance comparison methodology that can be used in all classification problems. This methodology compares performance metrics, which are calculated based on different datasets, with a high level of fairness.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111485"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Gender-based data bias and model fairness evaluation in benchmarked open-access disease prediction datasets","authors":"Shahadat Uddin ,&nbsp;Huan Liang ,&nbsp;Haolan Guo","doi":"10.1016/j.compbiomed.2026.111503","DOIUrl":"10.1016/j.compbiomed.2026.111503","url":null,"abstract":"<div><div>The widespread use of open-access datasets for validating machine learning (ML) models has raised critical concerns about data bias and model fairness, particularly in relation to gender. This study systematically investigates gender-based data bias in disease prediction datasets and evaluates the fairness of ML algorithms trained on them. A total of 74 datasets were selected from Kaggle and the UCI Machine Learning Repository, based on the inclusion of gender as a feature and classification labels. Data bias was quantified using Earth Mover's Distance to measure disparities in class-wise gender distributions, with statistical significance assessed via bootstrapping. Fairness was evaluated across seven ML algorithms (Decision Tree, Random Forest, Logistic Regression, Artificial Neural Networks, Support Vector Machine, K-Nearest Neighbours, and Naïve Bayes) using k-fold cross-validation and statistical tests. Two fairness definitions, Equalised Odds and Treatment Equality, were applied. Results showed that 35 datasets exhibited gender-based data bias, disproportionately affecting females. Heart disease datasets had the highest prevalence of data bias, while the lung cancer and mental health datasets were found to be bias-free. Fairness outcomes varied significantly across algorithms, with Decision Tree showing the fewest issues and Logistic Regression the most. Bias-free datasets consistently produced fewer fairness concerns, with statistically significant differences (p &lt; 0.01) across all algorithm groups. These findings highlight the importance of addressing gender-based data bias and selecting appropriate algorithms to improve fairness in ML applications. The study highlights the importance of addressing gender-based data bias in enhancing model fairness. It contributes to the development of equitable AI systems, thereby supporting data-driven decision-making in healthcare.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111503"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075228","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Influence of CT harmonization in longitudinal radiomics for NSCLC immunotherapy response prediction","authors":"Benito Farina ,&nbsp;Gonzalo Vegas-Sánchez-Ferrero ,&nbsp;Ana Delia Ramos-Guerra ,&nbsp;Carmelo Palacios Miras ,&nbsp;Andrés Alcazar Peral ,&nbsp;José Carmelo Albillos Merino ,&nbsp;Jon Zugazagoitia ,&nbsp;Germán R. Peces-Barba ,&nbsp;Luis Seijo Maceiras ,&nbsp;Luis Paz-Ares ,&nbsp;Ignacio Gil-Bazo ,&nbsp;Manuel Dómine Gómez ,&nbsp;Raul San José Estépar ,&nbsp;María J. Ledesma-Carbayo","doi":"10.1016/j.compbiomed.2026.111501","DOIUrl":"10.1016/j.compbiomed.2026.111501","url":null,"abstract":"<div><div>This study investigates the variability of radiomic features in longitudinal CT scans from a multi-institutional NSCLC cohort and introduces a harmonization pipeline to improve predictive modeling of immunotherapy response. Baseline and follow-up CT scans from NSCLC patients treated with anti-PD-1/PD-L1 agents were analyzed, with two institutions combined for model training and internal testing, and a third institution serving as an external test set. To address variability from imaging parameters—such as scanner manufacturer, slice thickness, and noise—we applied image harmonization followed by feature harmonization using NestedComBat. This approach substantially reduced feature dependence on acquisition confounders (from 78.8% to 12.8%) and improved feature robustness across institutions. We further assessed the temporal consistency of radiomic features across longitudinal scans using the intraclass correlation coefficient (ICC). Image harmonization yielded the largest gains in stability (mean <span><math><mi>Δ</mi></math></span>ICC = +0.021, p <span><math><mo>&lt;</mo></math></span> 0.001), while the combined approach also enhanced longitudinal reliability (<span><math><mi>Δ</mi></math></span>ICC = +0.014, p <span><math><mo>&lt;</mo></math></span> 0.001). Finally, harmonization improved predictive performance for 6-month immunotherapy response, increasing the AUC from 0.695 to 0.768 in the internal test and from 0.692 to 0.802 in the external test. These results demonstrate that combining image- and feature-level harmonization enhances the robustness and temporal consistency of radiomic features, potentially supporting more reliable and generalizable predictive modeling across diverse datasets and clinical settings.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111501"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Research on breast ultrasound images lesion localization and diagnosis based on knowledge-driven and data-driven methods","authors":"Jianqiang Li ,&nbsp;Lintao Song ,&nbsp;Xiaoling Liu ,&nbsp;Yiming Liu ,&nbsp;Tianbao Ma ,&nbsp;Jun Bai ,&nbsp;Qing Zhao ,&nbsp;Xi Xu","doi":"10.1016/j.compbiomed.2026.111465","DOIUrl":"10.1016/j.compbiomed.2026.111465","url":null,"abstract":"<div><div>Breast cancer poses the most significant threat to women’s health, yet early detection through screening can markedly reduce mortality. Ultrasound imaging, with its affordability, non-invasiveness, and efficacy in dense breast tissue, has emerged as a crucial tool for early screening. Recent advancements in computer vision have spurred the development of computer-aided diagnostic systems that focus on the automated localization and diagnosis of breast lesions. However, challenges such as speckle noise, blurred boundaries, and low contrast in ultrasound images impede accurate lesion detection. This review examines recent studies on breast ultrasound lesion localization and diagnosis, emphasizing model feature construction. It provides an overview of the task, available datasets, and evaluation metrics, and outlines selection criteria through a comprehensive literature analysis. The review categorizes models into three groups: domain knowledge-driven, data-driven, and hybrid approaches. It also discusses current challenges and future directions, aiming to enhance the accuracy of breast lesion localization and diagnosis.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111465"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146017584","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Amplitude symbolic analysis: a tool for the evaluation of the autonomic function complementary to traditional symbolic approach","authors":"Alberto Porta ,&nbsp;Beatrice Cairo ,&nbsp;Vlasta Bari ,&nbsp;Chiara Arduino ,&nbsp;Ilaria Burzo ,&nbsp;Beatrice De Maria ,&nbsp;Paolo Castiglioni ,&nbsp;Luc Quintin ,&nbsp;Aparecida Maria Catai ,&nbsp;Franca Barbic ,&nbsp;Raffaello Furlan","doi":"10.1016/j.compbiomed.2026.111473","DOIUrl":"10.1016/j.compbiomed.2026.111473","url":null,"abstract":"<div><div>Symbolic analysis (SA) infers cardiac control from spontaneous stationary sequences of heart period (HP) by estimating the probability of symbolic pattern classes. Unfortunately, SA does not assess the fraction of HP variability associated with symbolic pattern families. This study proposes amplitude SA (ASA) accounting for absolute changes between consecutive HPs. ASA leverages uniform 6-bin quantization to symbolize HP, the delay embedding procedure to form length-3 symbolic patterns and a traditional strategy to group symbolic patterns into four classes families according to number and sign of variations between adjacent symbols. ASA computes the fraction of variance associated with symbolic pattern classes. ASA was applied to HP variability derived from: 1) healthy subjects during pharmacological challenges (n = 9; age: 25–46 yrs, 9 males); 2) healthy subjects during graded postural stimuli (n = 19; age: 21–48 yrs, 8 males); 3) Parkinson disease (PD) patients (n = 12; age: 55–79 yrs, 8 males) and matched healthy controls (n = 12; age: 58–72 yrs, 7 males). We computed both global and local ASA markers and we compared them with SA indexes. Over stationary HP series we found that: i) ASA provides a general method to decompose HP variance according to symbolic pattern classes; ii) ASA is useful to describe cardiac control; iii) ASA indexes are complementary to SA markers; iv) ASA emphasizes the link of HP variability markers expressed in absolute units with vagal control; v) global and local ASA approaches provide similar information. SA and ASA should be utilized concomitantly for a deeper characterization of cardiac control from spontaneous HP fluctuations.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111473"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976663","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Peptide-nanoparticle platforms for antisense therapeutics: A coarse-grained modeling approach to brain delivery","authors":"Burcu Yesildag Uner ,&nbsp;Alper Demir ,&nbsp;Pingkun Zhou ,&nbsp;Ekim Z. Taskiran ,&nbsp;Tsjerk Wassenaar","doi":"10.1016/j.compbiomed.2026.111479","DOIUrl":"10.1016/j.compbiomed.2026.111479","url":null,"abstract":"<div><div>Traumatic brain injury (TBI) is a leading cause of long-term neurological deficits, often resulting in complex, unresolved molecular and cellular dysfunctions. Among these, gene–circuit disruptions—particularly those affecting neuroinflammation, oxidative stress, and mitochondrial dynamics—have emerged as critical mediators of post-traumatic neuropathology. In this study, we utilized artificial intelligence (AI)-driven proteomics and RNA sequence integration to map altered signaling pathways following TBI. Computational predictions identified specific gene–circuit nodes susceptible to therapeutic intervention, including redox-sensitive mitochondrial regulators and genes involved in the neuroimmune interface. Importantly, although our analyses are derived from rodent models, the conserved signaling pathways and regulatory circuits identified here provide a translational window with strong relevance to human TBI pathophysiology, thereby bridging preclinical findings with potential therapeutic application. Based on these insights, we designed a suite of responsive nanoparticle formulations optimized in silico for targeted delivery to dysregulated brain regions. These carriers incorporated ligands targeting disrupted circuits and incorporated redox-sensitive release mechanisms. Our platform demonstrates the feasibility of a closed-loop, data-guided strategy that integrates AI-based gene network profiling with rational nanocarrier design. This approach provides a scalable framework for precision neurotherapeutics, particularly for complex disorders such as TBI where conventional monotherapies have proven inadequate.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111479"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146046221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-task non-contact ballistocardiogram-based vital signs monitoring in acupuncture","authors":"Truong Tien Vo ,&nbsp;Quy Phuong Le ,&nbsp;Trong Nhan Nguyen ,&nbsp;Jaeyeop Choi ,&nbsp;Sudip Mondal ,&nbsp;Byeongil Lee ,&nbsp;Junghwan Oh","doi":"10.1016/j.compbiomed.2026.111461","DOIUrl":"10.1016/j.compbiomed.2026.111461","url":null,"abstract":"<div><div>The study introduces an innovative approach for efficient vital signs monitoring in acupuncture by combining multi-channel ballistocardiogram (BCG) signals and multi-task learning, taking advantage of the polyvinylidene fluoride (PVDF) film sensor and deep neural networks. The proposed system utilizes non-contact under-mattress BCG signals and deep learning for heart rate (HR), respiration rate (RR) estimation and lying posture detection. A custom-designed data-logger captures the signal from a BCG sensor located under the patient’s back for data acquisition, and integrates Gated Recurrent Unit (GRU) and Multi-head Self-Attention (MHSA) deep learning mechanisms for efficient HR, RR estimation and posture classification. In experiments with 25 participants, the proposed method achieved 98.7% accuracy for activity recognition and 97.6% for lying posture classification. In HR and RR estimation, the best case of mean absolute error (MAE) for HR achieves 0.77 beats per minute (bpm) in the right lateral posture, while the best value of MAE for RR is 0.43 breaths per minute (brpm) in the seated posture, compared to an FDA-approved device. The results demonstrate the high performance of multi-task learning for vital signs estimation and posture classification with our BCG-based system. This work establishes an innovative and practical pathway for medical assistance tools in non-contact monitoring and management.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"203 ","pages":"Article 111461"},"PeriodicalIF":6.3,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145997559","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}