IEEE Transactions on Biomedical Engineering最新文献

{"title":"VIGIL: Vision-Language Guided Multiple Instance Learning Framework for Ulcerative Colitis Histological Healing Prediction.","authors":"Zhengxuan Qiu, Bo Peng, Lingrui Zhang, Xiaoying Tang, Jiankun Wang, Qin Guo","doi":"10.1109/TBME.2026.3723583","DOIUrl":"10.1109/TBME.2026.3723583","url":null,"abstract":"<p><strong>Objective: </strong>Ulcerative colitis (UC) is characterized by chronic inflammation with alternating remission-relapse cycles, necessitating precise histological healing (HH) evaluation to improve clinical outcomes. To address the limitations of annotation-intensive deep learning methods and suboptimal multi-instance learning (MIL) in HH prediction, we propose VIGIL, the first vision-language-guided MIL framework that integrates white-light endoscopy (WLE) and endocytoscopy (EC).</p><p><strong>Methods: </strong>VIGIL introduces a dual-branch MIL module, KS-MIL, based on top-K typical frame selection and similarity-weighted metric learning to effectively model relationships among frame features. By incorporating diagnostic report text and a specially designed multi-level alignment and supervision strategy between image-text pairs, VIGIL establishes joint vision-language guidance during training to capture richer disease-related semantic information. Furthermore, VIGIL employs a multi-modal masked relation fusion (MMRF) strategy to uncover latent diagnostic correlations between WLE and EC representations.</p><p><strong>Results: </strong>Comprehensive experiments on a real-world clinical dataset demonstrate VIGIL's superior performance, achieving 92.69% accuracy and 94.79% AUC, outperforming existing state-of-the-art methods.</p><p><strong>Conclusion: </strong>The proposed VIGIL framework establishes an effective vision-language guided MIL paradigm for UC HH prediction, reducing annotation burdens while improving prediction reliability.</p><p><strong>Significance: </strong>The research outcomes provide new insights for non-invasive UC diagnosis and hold theoretical significance and clinical value for advancing intelligent healthcare development.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148758634","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":"Seismo-Mat: mattress-embedded inertial sensors for unobtrusive monitoring of heart rate, respiratory rate, and bed occupancy during sleep.","authors":"Chiara Romano, Manish Sharma, Daniela Lo Presti, Sergio Silvestri, Emiliano Schena, Carlo Massaroni","doi":"10.1109/TBME.2026.3722918","DOIUrl":"10.1109/TBME.2026.3722918","url":null,"abstract":"<p><strong>Objective: </strong>Continuous monitoring of heart rate (HR) and respiratory rate (RR) during sleep may support longitudinal assessment and the identification of clinically relevant trends, while current gold-standard techniques such as polysomnography are intrusive and unsuitable for prolonged home use. This study aims to develop and assess the feasibility of an unobtrusive system for overnight monitoring of HR, RR, and preliminary bed occupancy detection in real-world settings.</p><p><strong>Methods: </strong>A novel nearable system is presented, based on multiple mattress-embedded inertial sensors capturing the mechanical vibrations and low-frequency movements associated with cardiac and breathing activity. A dedicated signal processing pipeline detects periods of stillness, selects the most informative sensor and axis, and estimates HR and RR, while bed occupancy is identified using machine learning techniques.</p><p><strong>Results: </strong>The system was assessed through controlled laboratory experiments and overnight home recordings against a reference device. Across different body postures, mean absolute errors were below 5 beats per minute for HR and below 2 breaths per minute for RR. In the home recordings, 96% of the analysis windows fell within 5 beats per minute for HR using the gyroscope, against 83% using the accelerometer.</p><p><strong>Conclusion: </strong>The results demonstrate the feasibility of multi-parameter sleep monitoring using mattress embedded inertial sensors in both the laboratory setting and unsupervised home conditions.</p><p><strong>Significance: </strong>The proposed system estimates HR, RR, and bed occupancy simultaneously from mattress-embedded sensors and offers new insights into the role of inertial measurement units in HR and RR monitoring during sleep.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148720326","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":"Noninvasive Intracranial Pressure Estimation Using Subspace System Identification and Machine Learning Algorithms: A Learning-to-Rank Approach.","authors":"Anni Zhao, Ayca Ermis, Jeffrey Robert Vitt, Sergio Brasil, Wellingson Paiva, Magdalena Kasprowicz, Malgorzata Burzynska, Robert Hamilton, Runze Yan, Ofer Sadan, J Claude Hemphill, Lieven Vandenberghe, Xiao Hu","doi":"10.1109/TBME.2026.3723349","DOIUrl":"10.1109/TBME.2026.3723349","url":null,"abstract":"<p><strong>Objective: </strong>Accurate noninvasive estimation of intracranial pressure (ICP) remains a major challenge in critical care. We developed a machine learning algorithm that integrates system identification and rankingconstrained optimization to estimate mean ICP from noninvasive signals.</p><p><strong>Methods: </strong>A machine learning framework was proposed to obtain estimated mean ICP values using arbitrary noninvasive signals. The subspace system identification algorithm is employed to identify cerebral hemodynamics models for ICP simulation using arterial blood pressure (ABP), cerebral blood velocity (CBv), and R-wave to R-wave interval (R-R interval) signals in a comprehensive database. A mapping function to describe the relationship between the features of noninvasive signals and the estimation errors is learned using innovative ranking constraints through convex optimization. Patients across multiple clinical settings were randomly split into testing and training datasets for performance evaluation of the mapping function.</p><p><strong>Results: </strong>The results indicate that about 31.88% of testing entries achieved estimation errors within 2 mmHg and 34.07% of testing entries between 2 mmHg and 6 mmHg from the nonlinear mapping with constraints.</p><p><strong>Conclusion: </strong>Our results demonstrate the feasibility of the proposed noninvasive ICP estimation approach.</p><p><strong>Significance: </strong>This work presents a proof-of-concept framework for noninvasive ICP estimation in patients with acute brain injury and related conditions. Although the findings demonstrate the feasibility of the proposed framework, important limitations remain, particularly regarding estimation accuracy, cohort diversity, and validation across clinical settings. Additional algorithm development, expanded multicenter datasets, and rigorous prospective validation are therefore required to improve performance and establish clinical utility.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148721912","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":"Dissecting Spectral Granger Causality Through Partial Information Decomposition.","authors":"Luca Faes, Gorana Mijatovic, Riccardo Pernice, Salvatore Spaccaforno, Daniele Marinazzo, Sebastiano Stramaglia, Yuri Antonacci","doi":"10.1109/TBME.2026.3722943","DOIUrl":"10.1109/TBME.2026.3722943","url":null,"abstract":"<p><strong>Objective: </strong>Granger causality (GC), a popular statistical method for the inference of directional influences between time series measured from a complex network, is sensitive to high-order (non-pairwise) interactions which fundamentally shape the collective network dynamics. This work introduces Partial Decomposition of Granger Causality (PDGC), a tool to elicit redundant and synergistic causal interactions among the subsystems of physiological networks.</p><p><strong>Methods: </strong>We use the framework of partial information decomposition to dissect the multivariate GC from a set of driver random processes to a target process into unique effects carried exclusively by each driver, redundant effects carried identically by more drivers, and synergistic effects carried jointly by some drivers but not by any of them individually. Computation is based on multivariate state-space models expanded in the frequency domain to assess PDGC both in specific bands and in the time domain after whole-band integration.</p><p><strong>Results: </strong>The validation on benchmark simulations demonstrates that the measures of unique, redundant, and synergistic GC reflect the underlying causal mechanisms and are computationally reliable. The application to arterial pressure, respiration, cerebral blood velocity and heart period variability reveals striking differences in the response to postural stress of patients prone to neurally-mediated syncope compared to healthy controls.</p><p><strong>Conclusion: </strong>PDGC elicits unprecedented modes of physiological interaction, related to the sympathetic control of low-frequency cardiovascular and cerebrovascular oscillations, characterizing distinctive patterns of autonomic dysfunction.</p><p><strong>Significance: </strong>The extraction of high-order causality patterns from the spectral GC allows dissecting the mechanisms of causal influence underlying multivariate physiological interactions, potentially providing new biomarkers for the assessment of patho-physiological states.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148722356","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":"Free-Text Smartphone Keystroke Dynamics for Cognitive Impairment Pre-Screening with LLM-Assisted Text Analysis.","authors":"Jaehun Uhm, Jaekyum Lee, Ji Hwan Yun, Hong Yoon Kim, Byeongjin Choe, Seoyoung Yoon, Hyobin Lee, Woohyun Baek, Mira Yu, Sungjun Cho, Sung-Yoon Kang","doi":"10.1109/TBME.2026.3722901","DOIUrl":"10.1109/TBME.2026.3722901","url":null,"abstract":"<p><strong>Objective: </strong>To evaluate whether free-text smartphone keystroke dynamics, augmented by LLM-assisted text analysis, can support low-burden pre-screening for mild cognitive impairment (MCI) under more natural typing conditions than fixed-text copy tasks.</p><p><strong>Methods: </strong>We developed a framework that combines timing-based keystroke features with LLM-assisted text-derived features while reducing reliance on bespoke language-specific natural language processing. Using free-text smartphone typing data from aKorean-speaking cohort comprising cognitively normal (CN) participants and participants with protocol-confirmed MCI, weaggregated participant-level features across repeated sessions and evaluated three proposed digital biomarker configurations under a common XGBoost classifier.</p><p><strong>Results: </strong>The proposed free text keystroke-derived biomarkers showed meaningful utility for participant-level cross-sectional MCI pre-screening. Timing-based biomarkers and LLM-assisted biomarkers achieved ROC-AUCs of 0.737 and 0.703, respectively, and their combination improved the ROC-AUC to 0.771.</p><p><strong>Conclusion: </strong>Free-text smartphone keystroke derived biomarkers can capture clinically relevant signals associated with cognitive impairment pre-screening, with timing-based biomarkers providing a strong core signal and LLM-assisted analysis contributing complementary information when automated text processing is acceptable.</p><p><strong>Significance: </strong>This work presents a practical biomedical engineering route toward low-burden adjunctive pre-screening for cognitive impairment on personal smartphones using ecologically collected typing behavior. Because the present evaluation was conducted in a Korean-speaking cohort, broader multilingual generalizability remains to be established.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148724546","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":"ECG-Based Spatiotemporal Characterization of Delayed and Heterogeneous Ventricular Activation in Brugada Syndrome.","authors":"Sofia Romagnoli, Alba Isabel-Roquero, Alba Martin-Yebra, Flavio Palmieri, Pedro Gomis, Pablo Laguna, Elena Arbelo, Ana Minchole","doi":"10.1109/TBME.2026.3723174","DOIUrl":"10.1109/TBME.2026.3723174","url":null,"abstract":"<p><strong>Objective: </strong>Brugada syndrome (BrS) has been associated with delayed and heterogeneous electrical activation, particularly involving the RVOT. This study proposes a spatiotemporal characterization of ventricular activation using the high-precordial 12-lead electrocardiogram (ECG).</p><p><strong>Methods: </strong>Delayed ventricular activation was quantified by a lead-dependent propagation progression time (PPT), defined as the time instant within the QRS complex at which 30%, 50%, or 70% of cumulative QRS energy is reached. PPTs were calculated every 30 minutes and characterized temporally, by their 24-hour median and standard deviation, and spatially, by their maximum inter-lead differences. The method was evaluated on 162 subjects (118 patients, 44 controls).</p><p><strong>Results: </strong>BrS patients exhibited significantly longer PPTs across all leads than controls. The largest difference was observed at 70% QRS energy (median PPT in lead $rm V2_{2IC}$: BrS 74[63;87]ms vs control 60[56;66]ms, p-value< 0.001). Increased spatial dispersion of PPT was observed in male patients (maximum inter-lead difference: male 20[12;26]ms vs female 12[9;20]ms, p-value< 0.01), potentially reflecting sex-related differences in ventricular conduction patterns. The intra-subject temporal variability of PPT at 70% QRS energy can distinguish between controls and BrS patients with induced type-I ECG pattern (AUC>0.75 in leads $rm V2_{2IC}$ and $rm V1_{3IC}$).</p><p><strong>Conclusion: </strong>PPT analysis captures the spatiotemporal dynamics of delayed and heterogeneous ventricular activation associated with BrS and provides a quantitative non-invasive characterization of conduction abnormalities.</p><p><strong>Significance: </strong>Developing a non-invasive ECG-based approach for the assessment of ventricular conduction dynamics may complement current non-invasive ECG-based diagnostic tools in BrS, even in patients without spontaneous type-I pattern manifestation.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148725456","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":"Fabrication and Biocompatibility Testing of Thin-Film Metamaterial Neurostimulation Leads for Magnetic Resonance Imaging.","authors":"Francesca Marturano, Tayeb Zaidi, Aditya Tummala, Pamela Pantazopoulos, Joshua P Aronson, Laleh Golestanirad, Ilknur Ay, Giorgio Bonmassar","doi":"10.1109/TBME.2026.3723341","DOIUrl":"10.1109/TBME.2026.3723341","url":null,"abstract":"<p><strong>Objective: </strong>Chronically implantable neurostimulation leads, such as those used for deep brain stimulation, remain only partially MRI-compatible due to the presence of wire conductors that can induce hazardous radiofrequency (RF) heating. As a result, current systems are limited to MRI-conditional use, restricting patients' access to routine clinical imaging. Here, we propose a novel metamaterial based Resistive Tapered Stripline (RTS) architecture as a candidate solution for next-generation DBS leads.</p><p><strong>Methods: </strong>The RTS design incorporates impedance transitions along the lead, producing metamaterial-like behavior that increases inductance and promotes RF current reflection and dissipation while preserving low-frequency electrical conduction required for the therapeutic stimulation. Ultrathin titanium-gold microstrip segments with sharp impedance transitions were fabricated via physical vapor deposition on a non-conductive, non-magnetic substrate wire, with electrical properties guided by electromagnetic simulations. Material biocompatibility was evaluated in a rodent model following ISO 10993-6 guidelines.</p><p><strong>Results: </strong>Scattering-parameter measurements demonstrated high RF reflection (>64% of incident power), low transmission, and symmetric electrical behavior, confirming effective suppression of RF coupling. Histological analyses of microgliosis, astrocytosis, neurodegeneration, demyelination, and cellular composition showed no significant differences between RTS wires and commercially available DBS leads, confirming material biocompatibility.</p><p><strong>Conclusion: </strong>We developed and tested a novel technology specifically designed as a candidate for future MRI compatible implantable DBS leads.</p><p><strong>Significance: </strong>This technology could represent a promising advancement toward MRI safer implantable neurostimulation systems, enabling broader access to high-quality MRI while maintaining device performance and long-term biocompatibility. Its versatile materials and manufacturing approach could be potentially extended to other chronically active implantable leads.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148722268","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":"Latent Koopman Dynamics of Brain Structure-Function Coupling for Identifying Adolescent Prenatal Drug Exposure.","authors":"Badhan Mazumder, Lei Wu, Sir-Lord Wiafe, Vince D Calhoun, Dong Hye Ye","doi":"10.1109/TBME.2026.3722610","DOIUrl":"10.1109/TBME.2026.3722610","url":null,"abstract":"<p><strong>Objective: </strong>Prenatal drug exposure (PDE) has been linked to persistent alterations in adolescent brain development, yet the mechanisms by which structural and functional networks jointly contribute to later neurodevelopmental vulnerability remain unclear. Existing approaches typically analyze brain connectivity as static or treat modalities independently, limiting insight into how structure-function interactions are jointly characterized in relation to individual cognitive profiles.</p><p><strong>Methods: </strong>We present NeuroKoop++, a graph neural network-based multimodal framework that characterizes brain structure-function coupling through subject-level latent dynamical modeling. Modality-specific graph neural network (GNN) encoders extract structural connectivity (SC) and functional network connectivity (FNC) representations, which are integrated via bidirectional cross-attention and subsequently advanced through a spectrally constrained Koopman operator conditioned on subject-specific cognitive scores.</p><p><strong>Results: </strong>Applied to the ABCD cohort of 10,199 adolescents, NeuroKoop++ outperformed leading state-of-the-art multimodal approaches with statistically significant improvements and revealed interpretable signatures of altered large-scale brain network organization associated with PDE history.</p><p><strong>Conclusion: </strong>Modeling structure-function coupling as a cognition-modulated latent dynamical process yields robust and interpretable gains for PDE classification over existing fusion approaches.</p><p><strong>Significance: </strong>This work offers a principled computational pathway toward identifying neurodevelopmental biomarkers of PDE, with broader applicability to multimodal neuroimaging in pediatric and adolescent brain health research.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148706520","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":"The use of a simple and wireless Lenz lens for in ovo MR imaging at ultra-high field.","authors":"Chang-Hoon Choi, Sandra Krause, Suk-Min Hong, Jorg Felder, N Jon Shah","doi":"10.1109/TBME.2026.3722271","DOIUrl":"10.1109/TBME.2026.3722271","url":null,"abstract":"<p><p>A Lenz lens is a detachable, passive structure used to enhance the RF-field (B1-field), significantly improving local sensitivity by concentrating the passively induced B1-field from a confined region, and thereby increasing signal-to-noise ratio (SNR). Attaining higher sensitivity or SNR is crucial in MRI, as it directly impacts image quality, specifically spatial resolution and acquisition time. In this study, we present the design of a novel yet simple and wireless Lenz lens with a diameter of approximately 10 mm in the focusing part, intended to improve MR image quality at 7 T. The performance of the proposed lens was evaluated in in ovo MR measurements, with the specific aim of improving the MR image quality of a tumour implanted on the chorioallantoic membrane of a chick embryo, a region that is challenging to image using conventional MRI coils due to its location. The use of the lens resulted in a significant increase (approximately 75%) in SNR within the region-of-interest covered by the lens, whereas only negligible changes were observed in the areas outside its coverage, relative to that conducted without the lens.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148706528","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":"Adaptive Multi-Prototype Open-Set Recognition Network for Medical Image Classification.","authors":"Xingyu Cai, Haiyan Yang, Jiayi Chen, Jun Shi, Jun Wang","doi":"10.1109/TBME.2026.3721904","DOIUrl":"10.1109/TBME.2026.3721904","url":null,"abstract":"<p><strong>Objective: </strong>Existing prototype learning methods for open-set recognition (OSR) often use a fixed number of prototypes to represent each class, which struggle to model the inherent intra class variations widely existing in practical scenarios. This limitation is particularly pronounced in medical image applications, where high intra-class heterogeneity makes accurate OSR challenging. To address this issue, we propose a novel framework called Adaptive Multi-Prototype Network with Pretrained Swin Transformer (PSW-AMPN) for OSR on medical images. Specifically, a sparse gated attention module is devised to compute attention scores based on prototype-sample relations, thereby adaptively suppressing redundant sub-prototypes and selectively activating discriminative ones for each class in an end to-end optimization process. By jointly optimizing classification loss and the regularization for open space risk based on multiple prototypes, PSW-AMPN effectively captures complex intra-class structures and enhances class discriminability. Furthermore, PSW-AMPN uses a Pretrained Swin Transformer and a lightweight projector as feature extractor to effectively capture both local and global features. Extensive experiments demonstrate that our approach significantly outperforms existing baselines and achieves state-of-the-art performance on multiple medical image OSR tasks.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-08-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148689042","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}