IEEE Transactions on Biomedical Engineering最新文献

{"title":"Integrating Low-loss High-permittivity Ceramic Helmet with RF Head Array Coil to Largely Enhance Coil Sensitivity and SNR for Human Brain MRI at 7T.","authors":"Soo Han Soon, Xin Li, Xiao-Hong Zhu, Qing X Yang, Michael T Lanagan, Kamil Ugurbil, Wei Chen","doi":"10.1109/TBME.2026.3737200","DOIUrl":"https://doi.org/10.1109/TBME.2026.3737200","url":null,"abstract":"<p><p>Use of proper high-permittivity material could improve radiofrequency (RF) coil efficiency and imaging signal-to-noise ratio (SNR) for human MRI applications. We designed, constructed, and evaluated RF transmit and receive head array coils integrated with a low-loss high permittivity ceramic (HPC) helmet with the relative permittivity (ϵ_r) of 200 and four-segment helmet structure for human brain MRI application at 7 Tesla (T). Both electromagnetic simulation and experimental MRI results consistently demonstrated that the novel HPC helmet largely increased the RF reception (B₁^-) and transmission (B₁⁺) magnetic field strength by 82% and 19%, respectively, and net imaging SNR gain by 64% across the entire human brain compared to the baseline condition using the same RF coil configuration without the HPC helmet. The findings conclude that the new HPC helmet effectively enhances SNR primarily by boosting the RF coil detection sensitivity while introducing only a small noise increase (11%). The HPC helmet moderately reduced coil to-coil coupling of the receive array elements. Compared to the baseline condition, HPC helmet demonstrated an approximate 10 dB reduction in noise correlation and 2% to 7% increase in the 97^th percentile of 1/g-factors with various acceleration factors, ultimately facilitating high-quality parallel imaging reconstruction. The four-segment HPC helmet showed a nearly 50% lower specific absorption rate (SAR) than the monolithic HPC helmet. This work demonstrates the advancement of the integrated HPC helmet and RF coil technology to largely improve the imaging quality at 7T for whole human brain MRI applications, and potentially for translation.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149038379","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":"Characterizing Higher-order Functional Brain Networks with Simplicial Complexes and Temporal Dynamics.","authors":"Xinyu Li, Pengyu Cheng, Chenlong Wang, Yunlu Cai, Jiaying Yan, Zhaohui Li, Xi Zhang","doi":"10.1109/TBME.2026.3737173","DOIUrl":"10.1109/TBME.2026.3737173","url":null,"abstract":"<p><p>While conventional approaches characterize brain functional connectivity using static, pairwise interactions, higher-order interactions among multiple regions and their temporal organization are also essential for understanding complex brain dynamics. Here, we propose a temporal higher-order functional brain network framework (THFBN) that integrates simplicial complexes with temporal dynamical modeling to infer directed, time-varying higher-order interactions from stereoelectroencephalography (SEEG) recordings in epilepsy. We show that higher-order interactions across time reveals structured network reconfigurations that are not captured by static higher-order representations alone, giving rise to rapid yet structured network reconfigurations that are not captured by static or pairwise representations. Applying THFBN to epileptic seizures reveals a spatially heterogeneous redistribution of higher-order positive and negative predictive influences across seizure stages, with the strongest variability observed in temporal and insular regions, while positive-negative balance remains generally stable despite substantial changes in influence strength. Together, these findings demonstrate that integrating temporal dynamics with higher-order interactions reveals structured network reconfigurations during seizure evolution that are not captured by conventional connectivity models, and provide a principled framework for studying dynamic higher-order brain organization in epilepsy.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149030352","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":"Unsupervised Subtyping of Focal Epilepsy IEDs via Graph-Based MEG Networks.","authors":"Qijian Zheng, Yafei Liu, Zhongyi Xie, Ke Dong, Zhi Liu, Yujie Chen, Jian Sun, Chunyan Cao, Feng Liu","doi":"10.1109/TBME.2026.3736741","DOIUrl":"10.1109/TBME.2026.3736741","url":null,"abstract":"<p><strong>Objective: </strong>Interictal epileptiform discharges (IEDs) are clinically important biomarkers in focal epilepsy, yet their source-level network organization and subtype structure remain insufficiently characterized. This study aimed to identify reproducible IED network subtypes from source-level magnetoencephalography (MEG) connectivity and to examine whether these subtypes capture clinically meaningful information related to MEGIEDlocalization.</p><p><strong>Methods: </strong>Fifty-five patients with focal epilepsy were retrospectively included. After manual IED identification, source-level MEG signals were reconstructed and converted into 148 × 148 phase-locking value connectivity matrices. These matrices provided the basis for graph-based representation learning, in which DeepWalk, Node2Vec, Graph Convolutional Network (GCN), Graph Attention Network (GAT), and Multi-View Graph Representation Learning (MVGRL) models were paired with K-means or spectral clustering to identify latent interictal epileptiform discharge (IED) network subtypes. The selected subtype structure was then examined in a downstream clinical prediction task. Specifically, a subtype enhanced model encoded each patient as a bag of IEDconnectivity matrices, summarized IED-level embeddings through attention based multiple-instance pooling, and integrated subtype composition through soft prototype routing for MEG IED localization classification.</p><p><strong>Results: </strong>A total of 8,509 IEDs were identified. MVGRL combined with K-means achieved the best clustering performance and revealed four IED network subtypes. These subtypes differed in global network efficiency, nodal organization, and associations with MRI-derived cortical thickness. Overall network efficiency followed the order Subtype 4, Subtype 1, Subtype 2, and Subtype 3. Matrix-level validation showed that the four subtypes could be classified from graph theoretical features with an overall SVM accuracy of 93.90%. In patient-level validation, the subtype-enhanced model achieved 58.8% balanced accuracy for temporal-versus-other MEG IED localization classification, outperforming both the graph-feature random forest baseline and the deep multiple-instance baseline without subtype routing.</p><p><strong>Conclusion: </strong>This study demonstrates that leveraging the spatiotemporal dynamics of IEDs through unsupervised clustering enhances the understanding of shared/unique network features across focal epilepsy types.</p><p><strong>Significance: </strong>These findings provide an imaging-based framework for characterizing the heterogeneity of interictal epileptiform network states in focal epilepsy. Further validation against independently established clinical seizure-onset and epileptogenic zone labels is required before the framework can be interpreted as a clinical seizure-localization tool.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149024720","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":"Multimodal Portable Donor Organ Assessment Device.","authors":"Yuvaraj Manchukan, Mohammad Mir, Aneri Patel, Shadi Sarami, Jun Tae Huh, Brandon A Guenthart, Gordana Vunjak-Novakovic, Jinho Kim","doi":"10.1109/TBME.2026.3736322","DOIUrl":"10.1109/TBME.2026.3736322","url":null,"abstract":"<p><strong>Objective: </strong>The objective of this study is to develop a portable, handheld multimodal sensing device for rapid, localized, and quantitative assessment of donor organ tissue. A human-sized swine lung was used as a representative model system to evaluate device feasibility and performance.</p><p><strong>Methods: </strong>The device integrated ultrasound imaging, indentation-based stiffness measurement, bioelectrical impedance spectroscopy, and visible near-infrared optical sensing within a single system. Device validation was conducted using engineered tissue phantoms, rat lungs, and ex vivo swine lungs.</p><p><strong>Results: </strong>The sensing modalities effectively differentiated variations among tissue phantoms. In rat lungs, the optical sensor detected temporal changes in reflected light following euthanasia (It = 0 = 1.00 ± 0.02; It = 90min = 0.41 ± 0.01). In swine lungs, the device distinguished healthy and injured tissue. Specifically, injured lung tissue exhibited increased mechanical stiffness (healthy: 0.35 ± 0.08 kPa vs. injured: 9.77 ± 1.21 kPa) and reduced bioimpedance (proximal lung: 671.3 ± 161.6 vs. 596.3 ± 121.9 Ω; distal lung: 696.3 ± 43.7 vs. 237.4 ± 58.1 Ω) with healthy tissue. Spatial mapping of bioimpedance further revealed regional heterogeneity associated with tissue injury.</p><p><strong>Conclusion: </strong>These findings demonstrate the feasibility of integrating multiple sensing modalities into a single portable platform for localized and quantitative tissue assessment.</p><p><strong>Significance: </strong>This study establishes proof of concept for portable multimodal tissue assessment and provides a foundation for future evaluation in clinically relevant transplantation workflows, including ex vivo lung perfusion systems and donor organ assessment applications.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149030349","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":"OrthoTSS: A Zero-Calibration Framework for Cross-Subject P300 Decoding Via Orthogonality-Guided Task-Subject Separation.","authors":"Xingrong Du, Xiaoying Xu, Hong Zeng, Huayu Zhang, Jie Yu, Hanjie Deng, Aiguo Song, Wenbin Zhang, Dapeng Chen","doi":"10.1109/TBME.2026.3735299","DOIUrl":"10.1109/TBME.2026.3735299","url":null,"abstract":"<p><strong>Objective: </strong>Cross-subject P300 decoding remains challenging for zero-calibration brain-computer interfaces (BCIs), as inter-subject variability must be reduced while preserving weak task-relevant neural information.</p><p><strong>Methods: </strong>We propose Orthogonality-guided Task-Subject Separation (OrthoTSS), integrating a multi-scale spatio-temporal frontend, dual bidirectional Mamba streams, and orthogonality regularization. The task stream performs target/non-target decoding, while an auxiliary domain stream models subject-related variability during training. The regularization reduces linear cross-stream coupling and promotes functional differentiation without assuming complete disentanglement.</p><p><strong>Results: </strong>Under leave-one-subject-out (LOSO) evaluation, OrthoTSS achieved 76.35% balanced accuracy on PhysioNet ERP and 86.75% on Naturalistic Search FRP. It showed favorable performance against representative architectural, recent cross-subject, and domain-generalization baselines. Ablation and representation analyses further indicated reduced cross-stream similarity and relative branch specialization while preserving task-discriminative structure.</p><p><strong>Conclusion: </strong>OrthoTSS improves cross-subject P300 decoding by promoting functional differentiation between task-related and subject-related representations while retaining discriminative neural information.</p><p><strong>Significance: </strong>This task-preservation-oriented framework provides offline evidence toward practical zero-calibration P300 BCI decoding.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149038438","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":"Efficient Neural Decoding of Multi-Finger Forces for Continuous Human-Robot Interaction Control.","authors":"Long Meng, Xiaogang Hu","doi":"10.1109/TBME.2026.3736046","DOIUrl":"10.1109/TBME.2026.3736046","url":null,"abstract":"<p><strong>Objective: </strong>Accurate prediction of multi-finger forces can advance human-robotic interactions, enhancing the dexterity and precision of robotic assistance in various applications. Surface electromyogram (sEMG), which captures electrical signals during muscle activation, is a preferred information source for such predictions. Nevertheless, existing techniques are hindered by either inaccurate decoding capacity or intricate processes.</p><p><strong>Methods: </strong>To address these challenges, our study introduced a novel motor unit (MU)-level probabilistic refinement framework to simultaneously and continuously predict extension and flexion forces of individual fingers. Specifically, we extracted MUs using high-density sEMG data during sequential multi-finger tasks, where numerous non-target MUs compromise decoding accuracy. To overcome this, we quantified the likelihood of each MU being finger-specific by mapping its average firing rates during plateau contractions into a normalized probability distribution using a Softmax-based strategy. An optimization-driven probability threshold was then applied to retain only those MUs with the most reliable and finger-specific contributions for final decoding.</p><p><strong>Results: </strong>The results showed that our new decoder accurately predicted sequential multi-finger forces during dexterous finger flexion and extension tasks. Our approach outperformed current state-of-the-art neural decoding methods (p < 0.05), with high computational efficiency and robustness over different task variations.</p><p><strong>Conclusion: </strong>Our MU-level probabilistic refinement effectively reduced non-target MU interference and enables accurate, robust, and efficient continuous decoding of individual-finger forces across flexion and extension.</p><p><strong>Significance: </strong>The outcomes can potentially promote advances in real-time force control during human-robot interactions, thereby fostering more intuitive and efficient collaborations between humans and machines.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149030273","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":"Exploring the Muscle-Tendon Biomechanics and Load Mitigation of a Passive Shoulder Exoskeleton During Overhead Work.","authors":"Jin Tian, Chunzhi Yi, Chao Gan, Chifu Yang, Yuli Fang, Haiqi Zhu, Baichun Wei","doi":"10.1109/TBME.2026.3735723","DOIUrl":"10.1109/TBME.2026.3735723","url":null,"abstract":"<p><strong>Objective: </strong>The biomechanical mechanisms, assistance-level effects, and load-mitigation effects of shoulder exoskeletons during overhead work remain insufficiently understood. This study aims to explain experimental outcomes using simulation-derived muscle-tendon metrics, compare assistive torque profiles within the tested range, and evaluate their biomechanical effects on load mitigation.</p><p><strong>Methods: </strong>OpenSim musculoskeletal simulations were conducted for eight participants performing overhead work. Partial least squares (PLS) regression was performed using EMG data from a previously published experimental dataset to characterize multivariate associations between simulation-derived muscle-tendon metrics and experimentally observed muscular outcomes. In addition, five participants were newly recruited in the present study for independent validation for the prediction from simulation, i.e. the muscle activation reduction under 0%, 30%, 50% and 60% conditions.</p><p><strong>Results: </strong>PLS analyses showed muscle- and condition-dependent latent associations between simulation-derived muscle-tendon metrics and experimental outcomes. Force- and tendon-related variables showed higher contributions in several fitted PLS models; these contributions were interpreted within the shared covariance structure of the mechanically coupled predictors rather than as independent physiological effects. Simulation results showed that 70% assistance produced the lowest mean total shoulder muscle activation, while no statistically significant differences were detected among the 50%, 60%, and 70% conditions. The 60% condition was retained as a representative experimentally tested condition for subsequent biomechanical-demand analyses. Under this condition, the cumulative activation index, force-weighted activation index, and cumulative joint loading index demonstrated reductions in cumulative muscle activation, force-weighted muscular demand, and shoulder joint loading, respectively. Experimental EMG decreased monotonically with increasing assistance up to 60%, which was the highest experimentally tested assistance level.</p><p><strong>Conclusion: </strong>This study demonstrates that simulation-derived muscle-tendon metrics can characterize experimental responses and support evidence-based comparison of assistive strategies for reducing biomechanical demand and shoulder loading during the tested overhead work.</p><p><strong>Significance: </strong>The proposed simulation-experiment pipeline supports evidence-based exoskeleton design for reducing the biomechanical load during overhead work.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149024740","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":"TinyMyo: a Tiny Foundation Model for Flexible EMG Signal Processing at the Edge.","authors":"Matteo Fasulo, Giusy Spacone, Thorir Mar Ingolfsson, Yawei Li, Luca Benini, Andrea Cossettini","doi":"10.1109/TBME.2026.3736157","DOIUrl":"10.1109/TBME.2026.3736157","url":null,"abstract":"<p><strong>Objective: </strong>Surface electromyography (EMG) is a non-invasive sensing modality used in biomechanics, rehabilitation, prosthetic control, and human-machine interfaces. However, achieving robust generalization across subjects, recording systems, and acquisition protocols remains challenging. While foundation models (FMs) are gaining traction for EMG, existing approaches remain limited to single downstream tasks and lack deployability. We address these limitations.</p><p><strong>Methods: </strong>We present TinyMyo, a lightweight FM based on a Transformer encoder architecture. The model is pre-trained in a self-supervised manner using masked reconstruction on public datasets. With 3.6 M parameters, TinyMyo is designed to support multiple downstream tasks through minimal task-specific head adaptations.</p><p><strong>Results: </strong>We demonstrate generalization across hand gesture classification, hand kinematic regression, speech production and recognition, with performance comparable or surpassing the state of the art (SoA), and model size below 5 M parameters. We achieve SoA results compared to previous FM-based works on the NinaPro DB5 (87.98%), UCI-EMG (97.1%), and EPN-612 (96.6%) datasets. We demonstrate the first-time deployment of an EMG FM on an ultra-low power microcontroller (GAP9), with an inference time of 0.785 s, energy of 44.91 mJ and power envelope of 57.18 mW, reduced to 0.496 s and 0.089 s inference times with a smaller (1.9 M) model variant on 5 and 1-s data windows.</p><p><strong>Conclusion: </strong>TinyMyo demonstrates that compact, self-supervised EMG FM can guarantee generalization across multiple downstream tasks while remaining compatible with low-power edge devices.</p><p><strong>Significance: </strong>TinyMyo is the first EMG FM for ultra-low-power edge-devices, enabling scalable and energy-efficient sensing for motor intent decoding, neuromuscular assessment, and human-machine interaction.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149038417","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":"Sonomyography-Based Decoding of Attempted Hand Movements in Individuals with Spinal Cord Injury.","authors":"Manikandan Shenbagam, Nikhil Chowdhary, Priyanka Vijay, Chitra Kataria, Biswarup Mukherjee","doi":"10.1109/TBME.2026.3735727","DOIUrl":"10.1109/TBME.2026.3735727","url":null,"abstract":"<p><strong>Objective: </strong>Restoring hand function is critical for individuals with spinal cord injury (SCI). This study develops and evaluates a sonomyography-based human-machine interface (HMI) for decoding attempted hand movements across individuals with different levels and severities of cervical SCI, using a sparse-ultrasound processing pipeline designed for 8 channel A-mode ultrasound systems.</p><p><strong>Methods: </strong>A custom convolutional neural network, SonoSCINet, was developed using sparse B-mode ultrasound data and evaluated in non-disabled individuals and individuals with SCI, and compared with traditional machine learning classifiers. Feature selection and two-dimensional UMAP were used to examine handcrafted and learned representations. The same network architecture and processing pipeline were retrained and evaluated on an 8-channel A-mode ultrasound system in individuals with cervical SCI spanning different neurological injury levels and functional impairments.</p><p><strong>Results: </strong>SonoSCINet outperformed traditional models for gesture classification using sparse B-mode data (non-disabled: $91.3 pm 12.79%$; SCI: $82.03 pm 23.48%$). With the 8-channel A-mode system, SonoSCINet achieved a mean accuracy of $94.17 pm 4.75%$, numerically higher than LDA-based methods ($86.69 pm 6.27%$; not statistically significant), with improved robustness and gesture separability in the learned latent space. A real-time feasibility study in individuals with SCI further supports online implementation of the proposed pipeline for HMI applications.</p><p><strong>Conclusion: </strong>SonoSCINet, trained with a subject-specific pipeline, enables accurate decoding of hand gestures across sparse B-mode and A-mode ultrasound systems and across a range of SCI impairment levels, supporting the feasibility of a unified processing framework across modalities.</p><p><strong>Significance: </strong>These results support sonomyography combined with deep learning as a promising direction for portable A-mode ultrasound HMIs in individuals with SCI, with real-time feasibility of the underlying acquisition pipeline demonstrated using a lightweight classifier.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149024779","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":"Continuous-Time Transformer with Adaptive Mutation-Locking for Early Prediction of Acute Kidney Injury.","authors":"Weizhi Nie, Haolin Chen, Huifang Hao, Yuting Su, Keliang Xie, Bo Yang","doi":"10.1109/TBME.2026.3735426","DOIUrl":"10.1109/TBME.2026.3735426","url":null,"abstract":"<p><strong>Objective: </strong>Accurate early prediction of acute kidney injury is critical, yet existing deep learning models operate as opaque black boxes that rely on uninterpretable correlations. Therefore, this study develops a transparent temporal attribution architecture to uncover precise pathophysiological triggers.</p><p><strong>Methods: </strong>We propose the CT-Former framework, which integrates continuous time neural networks with a transformer to preserve irregular patient trajectories without biased data imputation. Furthermore, a two-stage temporal attribution decoupling module purifies the global attention field. This module generates an explicit temporal attribution graph to identify sudden critical time steps. The network locks onto these acute mutational triggers before fusing this temporal signal with the background context via an adaptive gate.</p><p><strong>Results: </strong>Evaluated on a MIMIC-IV cohort of 18419 patients, the CT-Former achieves the highest predictive performance among the evaluated baselines across all prediction time windows. Additionally, the learned temporal attribution matrix demonstrates structural consistency with independent interpretability evaluations.</p><p><strong>Conclusion: </strong>The CT-Former establishes a transparent temporal decision pathway that delivers exceptional predictive robustness while entirely avoiding spurious statistical correlations.</p><p><strong>Significance: </strong>This explicitly interpretable framework represents a promising step toward interpretable prognostic tools for critically ill patients.</p>","PeriodicalId":13245,"journal":{"name":"IEEE Transactions on Biomedical Engineering","volume":"PP ","pages":""},"PeriodicalIF":4.4,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149001763","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}