IEEE Transactions on Neural Systems and Rehabilitation Engineering最新文献

{"title":"Continuous prediction of lower limb joint parameters for robotic rehabilitation based on horizontal position of centre of mass.","authors":"Jinjian Jiang, Mozafar Saadat, Guowei Liu, Marco Maddalena, S Mehdi Rezaei","doi":"10.1109/TNSRE.2025.3619996","DOIUrl":"https://doi.org/10.1109/TNSRE.2025.3619996","url":null,"abstract":"<p><p>Continuous prediction of joint parameters is replacing discrete gait phase control to be the mainstream in rehabilitation robot control field. The sensor-based methods which use inertial measurement units (IMU), surface electromyography (sEMG) and so on are widely used in gaining joint parameters for robot control. However, those methods introduce many sensors attached to patients and affect the walking during training. To reduce the number of sensors needed, a method is proposed to use centre of mass (CoM) horizontal position to predict angles, angular velocities and accelerations of ankle, knee, and hip joints. Long short-term memory (LSTM) is a kind of recurrent neural network (RNN) widely used in predicting time series data. To gain the most suitable model to predict joint parameters of each joint, the performances of Autoencoding-LSTM (combining encoder-decoder with LSTM), CNN-LSTM (combining convolutional neural network with LSTM) and stacked LSTM with different input window sizes on predicting joint parameters of each joint are compared, and the models with optimal performances are selected as a model pack to achieve high quality prediction of joint parameters. The number of sensors needed is reduced by 50% with the accuracy equal to those methods using IMU or sEMG sensors. And the results additionally show that different models perform variously on predicting different joint parameters of different joints.</p>","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"PP ","pages":""},"PeriodicalIF":5.2,"publicationDate":"2025-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145274548","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":"Reinforcement Learning Identifies Age-Related Balance Strategy Shifts","authors":"Huiyi Wang;Jozsef Kovecses;Guillaume Durandau","doi":"10.1109/TNSRE.2025.3619868","DOIUrl":"10.1109/TNSRE.2025.3619868","url":null,"abstract":"Falls are one of the leading causes of non-disease death and injury in the elderly, partly due to the loss of muscle mass in a musculoskeletal disorder named sarcopenia. Studying the impact of this muscle weakness on standing balance through direct human experimentation poses ethical dilemmas, involves high costs, and fails to fully capture the internal dynamics of the muscle. To address these limitations, we employ neuromusculoskeletal modeling to explore the impact of sarcopenia on balance. In this study, we introduce a novel full-body MSK model comprising both the torso and lower limbs, with 290 muscle actuators controlling 23 degrees of freedom and supporting varying levels of sarcopenia. Using reinforcement learning coupled with curriculum learning and muscle synergy representations, we trained an agent to perform standing balance on a backward-sliding plate and compared its behavior to human experiments. Our results demonstrate that, without pre-recorded experimental data, both healthy and sarcopenic agents can reproduce ankle and hip balancing strategies consistent with experimental findings. Furthermore, we show that as the degree of sarcopenia increases, the agent adapts its balancing strategy based on the platform’s acceleration. The full code is open-sourced and can be found in this repository: <uri>https://github.com/cherylwang20/StandingBalance</uri>","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"33 ","pages":"4078-4088"},"PeriodicalIF":5.2,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11197589","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145258122","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optical-Tattoo Sensing for Non-Contact Control of Bionic Limbs: A Conceptual Framework","authors":"Saeed Bahrami Moqadam;Ahmad Saleh Asheghabadi;Farzaneh Norouzi","doi":"10.1109/TNSRE.2025.3618615","DOIUrl":"10.1109/TNSRE.2025.3618615","url":null,"abstract":"Conventional pattern recognition (PR) methods in bionic hand systems are reliant on contact-based sensors and remain vulnerable to the inherent instability of biological signals. This study presents an alternative method that uses a novel non-contact PR approach to classify the motion of individual fingers and hand grasping gestures. This method does not depend on biosignals; instead, it utilises optical sensing. To enhance optical differentiation during muscle contraction, interference-pigment tattoos were applied to the targeted areas of the skin of the muscular areas. In this approach, muscle activity in the forearm’s flexor medialis region is captured through red-green-blue (RGB) colour information and reflected light intensity (LI), which is then processed by a low execution time (ET) and an accurate recognition system. Two integrated sensors were used to precisely detect light reflections associated with a group of muscle activities. To minimise feeding data to the system, the RGB signals were clustered by light intensity using the k-nearest neighbours (KNN) algorithm, and then both time-domain and wavelet features were extracted and classified using four classifiers, including linear discriminant analysis (LDA), support vector machine (SVM), multilayer perceptron (MLP), and convolutional neural network (CNN). In the golden-integrated tattoo design, the system demonstrated a mean accuracy of 97.74%±0.8% across twelve intact participants and a mean accuracy of 95.23%±1.2% across six wrist disarticulation amputees, both classified as athletes and non-athletes. This underscores its strong potential as an alternative method to enhance the accuracy, intuitiveness, and reliability of prosthetic hand control systems.","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"33 ","pages":"4171-4183"},"PeriodicalIF":5.2,"publicationDate":"2025-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11195833","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145244468","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Raster Scanning Can Improve Task Performance in Simulated Prosthetic Vision","authors":"Haozhe Zac Wang;Yan Tat Wong","doi":"10.1109/TNSRE.2025.3617891","DOIUrl":"10.1109/TNSRE.2025.3617891","url":null,"abstract":"Current challenges exist for cortical visual prostheses in presenting complex visual scenes. One of the major constraints is the number of electrodes that can be stimulated simultaneously, due to issues with electrical interaction between electrodes and the resulting complications in visual perception. To overcome this, studies have presented the outline of objects sequentially. However, this method has only been tested with simple visual stimuli, such as letters. We combined the strengths of both simultaneous and sequential presentation of phosphenes via a novel stimulation protocol named raster scanning. We tested this method using simulated prosthetic vision with a Virtual Reality headset and evaluated participants’ visual abilities over three tasks. We recorded head movement data to investigate the various strategies participants employed to explore the visual scene. We found that raster scanning could improve task accuracy and reduce response time across three tasks. Moreover, raster scanning required less head movement to complete tasks. These results suggest that raster scanning binds visual cues more efficiently than head scanning alone. Our findings underscore the importance of sequentially presenting visual information and confirm the potential of cortical visual prostheses to provide functional vision, even under the current spatial and temporal constraints.","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"33 ","pages":"4159-4170"},"PeriodicalIF":5.2,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11193827","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145238525","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Post-Stroke Fine Hand Motion Intention Recognition Based on sEMG Decomposition and Residual Spiking Neural Networks","authors":"Jinting Ma;Lifen Wang;Yiyun Tan;Jintao Chen;Naiwen Zhang;Lihai Tan;Guanglin Li;Minghong Sui;Naifu Jiang;Guo Dan","doi":"10.1109/TNSRE.2025.3616378","DOIUrl":"10.1109/TNSRE.2025.3616378","url":null,"abstract":"Fine motor dysfunction of the hand severely impacts activities of daily living in stroke survivors. Accurate decoding of motion intentions from surface electromyography (sEMG) is critical for enabling survivors to participate actively in robot-assisted rehabilitation. Motion intention recognition methods using motor unit spike trains (MUSTs) derived from sEMG decomposition have demonstrated superior performance compared to conventional sEMG-based methods. However, these methods inadequately leverage the inherent spatiotemporal sparse coding efficiency of MUSTs and the full potential of sEMG decomposition remains underutilized in post-stroke populations. This study proposes a hand motion intention recognition framework integrating sEMG decomposition with a residual spiking neural network (Res-SNN). sEMG signals were recorded from 14 neurotypical individuals and 7 stroke survivors performing 35 fine hand and wrist movements. The performance of Res-SNN was evaluated separately in neurotypical and post-stroke cohorts, and compared with a traditional sEMG-based deep residual network (ResNet) and a MUST-based convolutional SNN (CSNN). Results indicate that Res-SNN achieved classification accuracies above 0.95 for both cohorts, significantly surpassing those of ResNet (neurotypical: <inline-formula> <tex-math>$0.84pm 0.08$ </tex-math></inline-formula>; post-stroke: <inline-formula> <tex-math>$0.90pm 0.04$ </tex-math></inline-formula>). While Res-SNN showed comparable accuracy to CSNN in neurotypical subjects (<inline-formula> <tex-math>$0.99pm 0.01$ </tex-math></inline-formula> vs. <inline-formula> <tex-math>$0.96pm 0.08$ </tex-math></inline-formula>, <inline-formula> <tex-math>${P}={0}.{48}$ </tex-math></inline-formula>), it substantially outperformed CSNN in stroke survivors (<inline-formula> <tex-math>$0.95pm 0.03$ </tex-math></inline-formula> vs. <inline-formula> <tex-math>$0.71pm 0.16$ </tex-math></inline-formula>, <inline-formula> <tex-math>${P}lt 0.001$ </tex-math></inline-formula>). Moreover, Res-SNN exhibited low inference power consumption (5.41 mJ<inline-formula> <tex-math>$cdot $ </tex-math></inline-formula>s). By integrating sEMG decomposition with Res-SNN, this study provides a high-accuracy and energy-efficient solution for post-stroke intention recognition, advancing the application of neural decoding technologies and neuromorphic computing in human-machine interfaces.","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"33 ","pages":"4147-4158"},"PeriodicalIF":5.2,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11187391","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145238598","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficiently Identifying Non-FoG, Pre-FoG, Pre-FoG Transition, and FoG in Parkinson’s Disease Patients Using Window Acceleration and Spline Function Features","authors":"Yi-Ting Hwang;Yu-Ting Yeh;Hung-Jui Hsu;Cheng-Ping Huang;Yi-Syuan Ke;Ren-Kai Lai;Jie-Ling Yen;Bor-Shing Lin","doi":"10.1109/TNSRE.2025.3618400","DOIUrl":"10.1109/TNSRE.2025.3618400","url":null,"abstract":"Freezing of gait (FoG) is an episodic symptom that disrupts walking initiation in patients with Parkinson’s disease (PD). Identifying pre-FoG stages is crucial for patients with PD. However, in contrast to typical disease labels, defining the FoG and pre-FoG states during PD gait data collection is challenging. The sliding window method can be used to increase data volume; however, the labeling of windows according to fixed FoG or pre-FoG data point thresholds is insensitive to PD severity. Therefore, this study proposes a novel algorithm that dynamically defines labels on the basis of the collected gait data. Overlapping windows are used to augment these data, and sensitivity analysis is conducted to assess the effect of the overlap rate on classification. This study used accelerometer data collected from UCI, which included 10 high-risk of FOG participants and worn on their ankles, thighs, and trunk. Based on this experiment, our approach achieved a sensitivity of 89% and a specificity of 92% for identifying all FoG stages. Moreover, it exhibited a sensitivity of 96% and specificity of 88% for detecting the pre-FoG state (2 s prior to FoG). The aforementioned results were obtained with only 25 key features, thus reducing the computational demand. Furthermore, the risk of overfitting was low for an overlap rate below 25%. This study highlights the importance of dynamic label assignment for the accurate classification of FoG stages and provides the important features for FoG detection.","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"33 ","pages":"4100-4111"},"PeriodicalIF":5.2,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11194233","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145238562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Complex Network Analysis of Hippocampal Regulation in the Mouse Brain Network to Control Epileptic Seizures.","authors":"Xiaojun Zhou, Yuan Wang, Bailu Si","doi":"10.1109/TNSRE.2025.3616957","DOIUrl":"https://doi.org/10.1109/TNSRE.2025.3616957","url":null,"abstract":"<p><p>For focal epilepsy, modeling the virtual brain through large-scale network dynamics to customize treatments is currently a very promising approach. However, after obtaining the epileptic brain connectome of subjects, most researches were focused on exploring ways to help clinicians to better perform brain resections. From the perspective of complex networks, we explore the possibility of utilizing the strength of network coupling to treat seizures non-destructively. We use the Epileptor model to construct heterogeneous dynamic networks with epileptogenic zones and design global indices appropriate for this model to describe systemic seizures. Based on these, we explored the effects of epileptogenic proportion and global coupling strength on different artificial networks, and finally verified on a real Allen mouse connectome that the enhancement of coupling strength can effectively control epilepsy. Our simulations found that as the epileptogenic proportion increased, seizure propagation steadily improved for the small-world and the scale-free networks, while the random network jumped from a sustaining state of global suppression to a state of global bursting. As for the increase in global coupling strength, the small-world network maintained a steady spread, while both the random and scale-free networks had their seizures significantly controlled. Subsequently, we validated the suppression in the Allen mouse focal seizure by boosting the coupling strength a little in its hippocampal formation. Our study shows that the structural nature of networks significantly affects seizure propagation and synchronization. The topology of the random network is significantly anti-epileptic, and others are easy to maintain. Coupling strength is an effective way to control epilepsy in both random and scale-free networks. Thus, we give an idea of using the structural nature of networks to control seizures non-destructively, while this may also be the theoretical basis for other cognitive training therapies, such as emotional or exercise in controlling epilepsy by training projection strengths from different brain regions.</p>","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"PP ","pages":""},"PeriodicalIF":5.2,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145225187","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":"Efficacy and Neural Mechanisms of Robotic-Assisted Therapy in Upper Extremity Rehabilitation for Stroke Survivors: A Resting-State fMRI Study.","authors":"Yingnan Lin, Hewei Wang, Li Ding, Qingming Qu, Jianghong Fu, Yifang Lin, Jie Gu, Wenyu Wang, Xueli Shan, Sujing Xu, Jie Jia, Yanyan Huang","doi":"10.1109/TNSRE.2025.3616524","DOIUrl":"https://doi.org/10.1109/TNSRE.2025.3616524","url":null,"abstract":"<p><p>Robotic-assisted therapy (RAT) represents a promising adjunctive rehabilitation technology, however, its underlying neuroplastic mechanisms remain incompletely characterized. We aimed to elucidate the neuroplastic reorganization induced by RAT that mediates motor functional improvements in stroke survivors. Thirteen stroke survivors in the RAT group and 13 demographically/clinically matched in the conventional rehabilitation therapy (CRT) group underwent a 4-week rehabilitation intervention. Motor function was assessed using the Fugl-Meyer Assessment upper and lower extremity subscale (FMA-UE, FMA-LE) and modified Barthel Index (MBI) at pre- and post-intervention timepoints. Concurrently, resting-state functional MRI (rs-fMRI) data were acquired for amplitude of low-frequency fluctuation (ALFF) computation and seed-based functional connectivity (FC) analysis. Repeated measures ANOVA showed significant Group × Time interactions for both FMA-UE and FMA-LE (F(1,24) = 4.913, p＜0.05; F(1,24) = 4.778, p＜ 0.05). All motor outcomes displayed strong main effects of Time (all p < 0.001). Post hoc simple effects tests revealed significant within group gains in FMA UE for both RAT and CRT and in FMA LE for RAT only, with no between group differences at any single time point. Neuroimaging showed that increases in ALFF within the ipsilesional precentral gyrus correlated with improvements in both FMA-UE and FMA-LE. Compared with CRT, RAT strengthened interhemispheric functional connectivity between the precentral and postcentral gyri and between the precentral and supramarginal gyri. Together, these findings indicate that RAT promotes motor recovery by up regulating activity in the ipsilesional motor cortex and enhancing cross hemispheric sensorimotor integration, providing the direct evidence for mechanism of post stroke neural restitution.</p>","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"PP ","pages":""},"PeriodicalIF":5.2,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145206418","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":"An EEG-EMG-based Hybrid Brain-Computer Interface for Decoding Tones in Silent and Audible Speech.","authors":"Jiawei Ju, Yifan Zhuang, Chunzhi Yi","doi":"10.1109/TNSRE.2025.3616276","DOIUrl":"https://doi.org/10.1109/TNSRE.2025.3616276","url":null,"abstract":"<p><p>Speech recognition can be widely applied to support people with language disabilities by enabling them to communicate through brain-computer interfaces (BCIs), thus improving their quality of life. Despite the essential role of tonal variations in conveying semantic meaning, there have been limited studies focusing on the neural signatures of tones and their decoding. This paper systematically investigates the neural signatures of the four tones of Mandarin. It explores the feasibility of tone decoding in both silent and audible speech using a multimodal BCI based on electroencephalography (EEG) and electromyography (EMG). The time-frequency analysis of EEG has revealed significant variations in neural activation patterns across various tones and speech modes. For example, in the silent speech condition, temporal-domain analysis shows significant tone-dependent activation in the frontal lobe (ANOVA p = 0.000, Tone1 vs Tone2: p = 0.000, Tone1 vs Tone4: p = 0.000, Tone2 vs Tone3: p = 0.000, Tone3 vs Tone4: p = 0.001) and in channel F8 (ANOVA p=0.008, Tone1 vs Tone2: p=0.014, Tone2 vs Tone3: p=0.034). Spectral analysis shows significant differences between four tones in event-related spectral perturbation (ERSP) in the central region (p = 0.000) and channel C6 (p = 0.000). EMG analysis identifies a significant tone-related difference in activation of the left buccinator muscle (p = 0.023), and ERSP from the mentalis muscle also shows a marked difference across tones in both speech conditions (p = 0.00). Overall, tone-related neural differences were more pronounced in the audible speech condition than in the silent condition. For tone classification, RLDA and SVM classifiers achieved accuracies of 71.22% and 72.43%, respectively, using EEG temporal features in both speech modes. Additionally, the RLDA classifier with temporal features achieves binary tone classification accuracies of 90.92% (audible tones) and 91.00% (silent tones). The combination of EEG and EMG yields the highest speech modes decoding accuracy of 81.33%. These findings provide a potential strategy for speech restoration in tonal languages and further validate the feasibility of a speech brain-computer interface (BCI) as a clinically effective treatment for individuals with tonal language impairment.</p>","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"PP ","pages":""},"PeriodicalIF":5.2,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145206346","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":"MCI Detection From Odor-Evoked EEG Using a Multibranch Attention-Based Temporal–Spectral CNN","authors":"Farhan Riaz;Muhammad Muzammal;Christos Frantzidis;Imran Khan Niazi","doi":"10.1109/TNSRE.2025.3616523","DOIUrl":"10.1109/TNSRE.2025.3616523","url":null,"abstract":"Dementia is a progressive neurodegenerative condition often preceded by Mild Cognitive Impairment (MCI), which is marked by early-stage memory difficulties and reduced cognitive flexibility. Detecting MCI at an early stage is crucial for timely intervention and for improving long-term cognitive health and quality of life. In this paper, we aim to differentiate between normal subjects and those suffering from MCI based on odor-evoked brain potentials from EEG signals. To address this challenge, we used publicly available multichannel EEG data and calculated a set of temporal-spectral components using wavelets, spectral grouping, and canonical correlation. These features are fed separately into attention-based convolutional neural network (CNN) models, which are individually trained on each feature-set, leading to individual feature branches. Later, these branches are fed into a fully connected network for performing the classification task. Our experiments demonstrate that the proposed method outperforms other methods considered in this paper. Ablation studies also reveal the individual strength of each set of features adopted in this study, along with their combined strength when the entire feature set is used for classification.","PeriodicalId":13419,"journal":{"name":"IEEE Transactions on Neural Systems and Rehabilitation Engineering","volume":"33 ","pages":"4031-4043"},"PeriodicalIF":5.2,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11187358","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145206447","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}