arxiv: 2604.16554 · v1 · submitted 2026-04-17 · 💻 cs.CV · cs.AI

PA-TCNet: Pathology-Aware Temporal Calibration with Physiology-Guided Target Refinement for Cross-Subject Motor Imagery EEG Decoding in Stroke Patients

Xiangkai Wang , Yun Zhao , Dongyi He , Qingling Xia , Gen Li , Nizhuan Wang , Ningxiao Peng , Bin Jiang This is my paper

Pith reviewed 2026-05-10 08:22 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords stroketemporalmotorpa-tcnetcalibrationcross-subjectdecodingdynamics

0 comments

The pith

PA-TCNet improves cross-subject motor imagery EEG decoding accuracy in stroke patients to 66.56% and 72.75% on two datasets by pathology-aware rhythmic state modeling and physiology-constrained pseudo-label refinement.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Stroke damages brain areas used for movement, creating slow abnormal waves and large differences between patients that break standard EEG decoding methods. PA-TCNet tackles this with two parts. The first part uses a Mamba state model to split EEG data into slow rhythmic background tied to the disease and quick changing signals, then feeds the combined context into the network to track the unusual timing patterns better. The second part builds templates from sensorimotor brain regions using source data and applies rules based on expected physiological responses to adjust guessed labels for target patients on the fly, making the adaptation more stable. On leave-one-subject-out tests from two stroke EEG collections, the approach reached average accuracies of 66.56 percent and 72.75 percent while beating prior methods. The work shows that explicitly handling disease-related timing changes and adding body-based constraints can make cross-patient BCI systems more practical for post-stroke motor recovery training.

Core claim

Leave-one-subject-out experiments on two independent stroke EEG datasets, XW-Stroke and 2019-Stroke, yielded mean accuracies of 66.56% and 72.75%, respectively, outperforming state-of-the-art baselines.

Load-bearing premise

That the PRSM module's decomposition of EEG into slowly varying rhythmic context and fast perturbations accurately captures lesion-related abnormal dynamics without introducing new artifacts, and that the PGTC module's physiological consistency constraints reliably refine pseudo-labels across highly heterogeneous stroke patients.

Figures

Figures reproduced from arXiv: 2604.16554 by Bin Jiang, Dongyi He, Gen Li, Ningxiao Peng, Nizhuan Wang, Qingling Xia, Xiangkai Wang, Yun Zhao.

**Figure 1.** Figure 1: Overall framework of PA-TCNet. The network first extracts local sensorimotor spatiotemporal patterns, then [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Figure 2: Architecture of the Pathology-aware Rhythmic State Mamba (PRSM) module. The module decomposes [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: t-SNE visualizations of feature distributions on the XW-Stroke (top) and 2019-Stroke (bottom) datasets at [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

**Figure 4.** Figure 4: Hyperparameter sensitivity analysis of PA-TCNet on XW-Stroke and 2019-Stroke. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗

**Figure 5.** Figure 5: Noise robustness of PA-TCNet under drift and spike perturbations. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗

read the original abstract

Stroke patient cross-subject electroencephalography (EEG) decoding of motor imagery (MI) brain-computer interface (BCI) is essential for motor rehabilitation, yet lesion-related abnormal temporal dynamics and pronounced inter-patient heterogeneity often undermine generalization. Existing adaptation methods are easily misled by pathological slow-wave activity and unstable target-domain pseudo-labels. To address this challenge, we propose PA-TCNet, a pathology-aware temporal calibration framework with physiology-guided target refinement for stroke motor imagery decoding. PA-TCNet integrates two coordinated components. The Pathology-aware Rhythmic State Mamba (PRSM) module decomposes EEG spatiotemporal features into slowly varying rhythmic context and fast transient perturbations, injecting the fused pathological context into selective state propagation to more effectively capture abnormal temporal dynamics. The Physiology-Guided Target Calibration (PGTC) module constructs source-domain sensorimotor region-of-interest templates, imposing physiological consistency constraints and dynamically refining target-domain pseudo-labels, thereby improving adaptation reliability. Leave-one-subject-out experiments on two independent stroke EEG datasets, XW-Stroke and 2019-Stroke, yielded mean accuracies of 66.56\% and 72.75\%, respectively, outperforming state-of-the-art baselines. These results indicate that jointly modeling pathological temporal dynamics and physiology-constrained pseudo-supervision can provide more robust cross-subject initialization for personalized post-stroke MI-BCI rehabilitation. The implemented code is available at https://github.com/wxk1224/PA-TCNet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

PA-TCNet pairs a Mamba decomposition for stroke-specific slow waves with physiology-constrained pseudo-label fixes, but the adaptation step still lacks the checks needed to trust the gains.

read the letter

The paper's main move is to split EEG features into slow rhythmic context and fast perturbations using a Mamba state model tuned for pathological patterns, then feed that into a module that builds source sensorimotor templates and applies consistency rules to clean target pseudo-labels. This gets reported mean accuracies of 66.56% and 72.75% in leave-one-subject-out on the two stroke datasets, beating the baselines they cite. The code release is a plus for anyone who wants to inspect the implementation directly. The approach is clearly aimed at the real clinical issue of lesion-driven temporal distortions and patient-to-patient differences that break standard domain adaptation in motor imagery BCI. The PRSM part at least tries to make the state propagation respect the abnormal dynamics instead of treating them as noise. That is a reasonable specialization. The PGTC part tries to anchor the target labels to physiological priors drawn from source data, which could in principle reduce the usual pseudo-label drift. The stress-test concern about PGTC holds up on the evidence given. Stroke lesions differ sharply in location and effect, so source templates are unlikely to be representative across the board. Without an ablation that turns PGTC off, without numbers on how often the refined labels match ground truth, and without per-subject spreads or statistical tests, it is hard to tell whether the calibration is helping or just smoothing over errors. The abstract and the reported results do not address that directly. This work is for people already working on cross-subject EEG decoding for stroke rehabilitation or similar clinical BCI settings. A reader who needs concrete modules and public code to experiment with would get something usable from it. It deserves a serious referee because the problem is well-motivated, the method is specific rather than generic, and the code is available, even though the validation of the adaptation component needs tightening.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes PA-TCNet, a pathology-aware temporal calibration framework for cross-subject motor imagery EEG decoding in stroke patients. It introduces the Pathology-aware Rhythmic State Mamba (PRSM) module, which decomposes EEG spatiotemporal features into slowly varying rhythmic context and fast transient perturbations for better capture of lesion-related abnormal dynamics, and the Physiology-Guided Target Calibration (PGTC) module, which builds source-domain sensorimotor ROI templates to impose physiological consistency constraints and refine target-domain pseudo-labels. Leave-one-subject-out experiments on the XW-Stroke and 2019-Stroke datasets report mean accuracies of 66.56% and 72.75%, respectively, outperforming state-of-the-art baselines. The code is made available at a GitHub repository.

Significance. If the reported gains hold under rigorous verification, the work could meaningfully advance BCI-based motor rehabilitation for stroke by explicitly addressing pathological slow-wave activity and inter-patient heterogeneity. The open release of implementation code is a clear strength that enables direct reproduction and extension.

major comments (3)

[Abstract] Abstract: The headline LOSO accuracies (66.56% and 72.75%) are presented without any accompanying information on baseline re-implementations, statistical significance testing, error bars, exact data splits, or handling of label noise, which directly undermines evaluation of the central claim that the two modules jointly improve generalization.
[Abstract / §3 (PGTC)] PGTC module description: The claim that source sensorimotor ROI templates and physiological consistency constraints reliably refine pseudo-labels across stroke patients rests on an untested assumption given pronounced lesion heterogeneity; no ablation isolating PGTC, no pseudo-label accuracy versus ground truth, and no per-subject variance are reported, leaving open the possibility that the constraints propagate rather than correct errors.
[Abstract / §3 (PRSM)] PRSM module: The decomposition of EEG into rhythmic context and fast perturbations is asserted to inject pathological context into selective state propagation, yet no quantitative analysis (e.g., state-transition statistics or ablation on the fusion step) demonstrates that this avoids introducing new artifacts in the presence of variable slow-wave abnormalities.

minor comments (2)

[Abstract] The abstract would benefit from stating the number of subjects and trials per dataset to allow immediate assessment of statistical power.
[Abstract] Ensure all acronyms (e.g., MI, BCI, ROI) are defined at first use and that the GitHub link is accompanied by a brief note on what is released (code, pretrained weights, or both).

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each of the major comments point-by-point below and will make the necessary revisions to improve the clarity, rigor, and completeness of the evaluation.

read point-by-point responses

Referee: [Abstract] Abstract: The headline LOSO accuracies (66.56% and 72.75%) are presented without any accompanying information on baseline re-implementations, statistical significance testing, error bars, exact data splits, or handling of label noise, which directly undermines evaluation of the central claim that the two modules jointly improve generalization.

Authors: We appreciate this observation. The detailed experimental protocol, including re-implementation of all baselines following their original publications under the same leave-one-subject-out (LOSO) setting, statistical significance testing via paired t-tests (with p-values reported in the results tables), standard deviations as error bars, and exact data splits (LOSO across stroke patients) are described in Section 4 of the manuscript. Label noise in the target domain is mitigated through the physiological consistency constraints in the PGTC module, as explained in Section 3.2. However, we agree that the abstract would benefit from a brief summary of these aspects. In the revised version, we will add a concise statement to the abstract noting the statistical significance of improvements and the use of identical evaluation protocols for baselines. revision: yes
Referee: [Abstract / §3 (PGTC)] PGTC module description: The claim that source sensorimotor ROI templates and physiological consistency constraints reliably refine pseudo-labels across stroke patients rests on an untested assumption given pronounced lesion heterogeneity; no ablation isolating PGTC, no pseudo-label accuracy versus ground truth, and no per-subject variance are reported, leaving open the possibility that the constraints propagate rather than correct errors.

Authors: We acknowledge the importance of validating the PGTC module's effectiveness given the heterogeneity of lesions in stroke patients. The current manuscript reports overall performance improvements when PGTC is integrated, but we agree that an isolated ablation study would provide stronger evidence. We will add such an ablation in the revised manuscript, comparing the full model against a variant without PGTC. Additionally, we will report per-subject accuracy variances and standard deviations in the main results table and supplementary material. Regarding pseudo-label accuracy versus ground truth, as this is an unsupervised domain adaptation setting, target labels are not available; however, we will include an analysis of pseudo-label consistency with source-domain predictions and discuss potential error propagation as a limitation. These additions will help demonstrate that the constraints primarily correct rather than propagate errors. revision: yes
Referee: [Abstract / §3 (PRSM)] PRSM module: The decomposition of EEG into rhythmic context and fast perturbations is asserted to inject pathological context into selective state propagation, yet no quantitative analysis (e.g., state-transition statistics or ablation on the fusion step) demonstrates that this avoids introducing new artifacts in the presence of variable slow-wave abnormalities.

Authors: We recognize that additional quantitative analysis would better support the design choices in the PRSM module. The decomposition is motivated by the need to separately model slow pathological rhythms and fast transients, with fusion into the Mamba state propagation. In the revision, we will include an ablation study on the fusion step (comparing decomposed vs. non-decomposed inputs) and provide supplementary visualizations or statistics on state transitions under varying slow-wave conditions from the datasets. This will illustrate that the approach captures lesion-related dynamics effectively, as supported by the superior performance over baselines that do not account for pathology. We will also discuss potential artifacts and how the physiology-guided constraints in PGTC help mitigate them. revision: yes

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 2 invented entities

The claim rests on the effectiveness of two newly introduced modules whose internal mechanisms are not derived from first principles but postulated to handle stroke-specific EEG issues; the neural architecture contains numerous fitted parameters whose values are not reported.

free parameters (1)

Neural network weights and hyperparameters (state dimension, learning rate, etc.)
Standard in deep learning models; optimized during training on source and target EEG data to achieve reported accuracies.

axioms (2)

domain assumption EEG spatiotemporal features can be meaningfully decomposed into slowly varying rhythmic pathological context and fast transient perturbations
Central to the PRSM module description in the abstract.
domain assumption Source-domain sensorimotor ROI templates provide valid physiological consistency constraints for refining target pseudo-labels
Invoked by the PGTC module to improve adaptation reliability.

invented entities (2)

Pathology-aware Rhythmic State Mamba (PRSM) module no independent evidence
purpose: Decompose EEG features and inject fused pathological context into selective state propagation
Newly proposed component to capture abnormal temporal dynamics in stroke patients.
Physiology-Guided Target Calibration (PGTC) module no independent evidence
purpose: Construct ROI templates and dynamically refine target-domain pseudo-labels
Newly proposed component to address unstable pseudo-labels in adaptation.

pith-pipeline@v0.9.0 · 5595 in / 1608 out tokens · 52914 ms · 2026-05-10T08:22:26.014433+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

32 extracted references

[1]

World stroke organization (wso): Global stroke fact sheet 2022.International 13 arXiv preprint Journal of Stroke, 17(1):18–29, 2022

Valery L Feigin, Michael Brainin, Bo Norrving, Sheila Martins, Ralph L Sacco, Werner Hacke, Marc Fisher, Jeyaraj Pandian, and Patrice Lindsay. World stroke organization (wso): Global stroke fact sheet 2022.International 13 arXiv preprint Journal of Stroke, 17(1):18–29, 2022. PMID: 34986727

2022
[2]

Wilson, Stephen J

Richard D. Wilson, Stephen J. Page, Michael Delahanty, Jayme S. Knutson, Douglas D. Gunzler, Lynne R. Sheffler, and John Chae. Upper-limb recovery after stroke: A randomized controlled trial comparing emg-triggered, cyclic, and sensory electrical stimulation.Neurorehabilitation and Neural Repair, 30(10):978–987, 2016. PMID: 27225977

2016
[3]

A comprehensive review of deep learning for motor imagery eeg: From healthy subjects to patients

Bin Jiang, Xiangkai Wang, Dongyi He, Siyu Cheng, Maoyu Liao, Duoqian Miao, Qingling Xia, Yun Zhao, and Gen Li. A comprehensive review of deep learning for motor imagery eeg: From healthy subjects to patients. Biomedical Signal Processing and Control, 117:109680, 2026

2026
[4]

Motor imagery combined with brain-computer interface for stroke patients: a meta-analysis.Frontiers in Neurology, V olume 17 - 2026, 2026

Yuhuang Lin, Yong Yuan, Jingjing Chen, and Xiangfu Lin. Motor imagery combined with brain-computer interface for stroke patients: a meta-analysis.Frontiers in Neurology, V olume 17 - 2026, 2026

2026
[5]

A comprehensive review of eeg-based brain–computer interface paradigms.Journal of Neural Engineering, 16(1):011001, jan 2019

Reza Abiri, Soheil Borhani, Eric W Sellers, Yang Jiang, and Xiaopeng Zhao. A comprehensive review of eeg-based brain–computer interface paradigms.Journal of Neural Engineering, 16(1):011001, jan 2019

2019
[6]

Pfurtscheller and C

G. Pfurtscheller and C. Neuper. Motor imagery and direct brain-computer communication.Proceedings of the IEEE, 89(7):1123–1134, 2001

2001
[7]

Eegnet: a compact convolutional neural network for eeg-based brain–computer interfaces.Journal of Neural Engineering, 15(5):056013, jul 2018

Vernon J Lawhern, Amelia J Solon, Nicholas R Waytowich, Stephen M Gordon, Chou P Hung, and Brent J Lance. Eegnet: a compact convolutional neural network for eeg-based brain–computer interfaces.Journal of Neural Engineering, 15(5):056013, jul 2018

2018
[8]

Deep learning with convolutional neural networks for eeg decoding and visualization.Human Brain Mapping, 38(11):5391–5420, 2017

Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Burgard, and Tonio Ball. Deep learning with convolutional neural networks for eeg decoding and visualization.Human Brain Mapping, 38(11):5391–5420, 2017

2017
[9]

Jiaheng Wang, Lin Yao, and Yueming Wang. Ifnet: An interactive frequency convolutional neural network for enhancing motor imagery decoding from eeg.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31:1900–1911, 2023

1900
[10]

Eeg conformer: Convolutional transformer for eeg decoding and visualization.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31:710–719, 2023

Yonghao Song, Qingqing Zheng, Bingchuan Liu, and Xiaorong Gao. Eeg conformer: Convolutional transformer for eeg decoding and visualization.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31:710–719, 2023

2023
[11]

Multi-scale convolutional transformer network for motor imagery brain-computer interface.Scientific Reports, 15(1):12935, Apr 2025

Wei Zhao, Baocan Zhang, Haifeng Zhou, Dezhi Wei, Chenxi Huang, and Quan Lan. Multi-scale convolutional transformer network for motor imagery brain-computer interface.Scientific Reports, 15(1):12935, Apr 2025

2025
[12]

Dbconformer: Dual-branch convolutional transformer for eeg decoding.IEEE Journal of Biomedical and Health Informatics, pages 1–14, 2025

Ziwei Wang, Hongbin Wang, Tianwang Jia, Xingyi He, Siyang Li, and Dongrui Wu. Dbconformer: Dual-branch convolutional transformer for eeg decoding.IEEE Journal of Biomedical and Health Informatics, pages 1–14, 2025

2025
[13]

Slimseiz: Efficient channel-adaptive seizure prediction using a mamba-enhanced network

Guorui Lu, Jing Peng, Bingyuan Huang, Chang Gao, Todor Stefanov, Yong Hao, and Qinyu Chen. Slimseiz: Efficient channel-adaptive seizure prediction using a mamba-enhanced network. In2025 IEEE International Symposium on Circuits and Systems (ISCAS), pages 1–5, 2025

2025
[14]

Unsupervised domain adaptation with synchronized self-training for cross- domain motor imagery recognition

Peiyin Chen, Xiaofeng Liu, Chao Ma, He Wang, Xiong Yang, Celso Grebogi, Xiao Gu, and Zhongke Gao. Unsupervised domain adaptation with synchronized self-training for cross- domain motor imagery recognition. IEEE Journal of Biomedical and Health Informatics, 29(5):3664–3677, 2025

2025
[15]

Ssas: Cross-subject eeg-based emotion recognition through source selection with adversarial strategy.Expert Systems with Applications, 308:130843, 2026

Yici Liu, Qi Wei Oung, and Hoi Leong Lee. Ssas: Cross-subject eeg-based emotion recognition through source selection with adversarial strategy.Expert Systems with Applications, 308:130843, 2026

2026
[16]

Ua-daan: An uncertainty- aware dynamic adversarial adaptation network for eeg-based depression recognition.IEEE Transactions on Affective Computing, 16(3):2130–2141, 2025

Jian Shen, Lechun You, Yu Ma, Zeguang Zhao, Huajian Liang, Yanan Zhang, and Bin Hu. Ua-daan: An uncertainty- aware dynamic adversarial adaptation network for eeg-based depression recognition.IEEE Transactions on Affective Computing, 16(3):2130–2141, 2025

2025
[17]

Rongrong Lu, Wenchang Deng, Tianhao Gao, Songhua Huang, Zhi Zhang, Yan Liu, and Sheng-hua Zhong. Mutual generation for cross-domain challenge in stroke patients’ motor imagery classification and functional recovery prediction.IEEE Journal of Biomedical and Health Informatics, pages 1–14, 2025

2025
[18]

Fa-eegnet: Enhancing eegnet with frequency- adaptive kernels for the classification of stroke rehabilitation movements

Milakul Hasana Mim, Debjoty Roy Pranto, and Md Mehedi Hasan. Fa-eegnet: Enhancing eegnet with frequency- adaptive kernels for the classification of stroke rehabilitation movements. In2026 5th International Conference on Electrical, Computer & Telecommunication Engineering (ICECTE), pages 1–6, 2026

2026
[19]

Ivana Kancheva, Sandra M. A. van der Salm, N. Ramsey, and M. Vansteensel. Association between lesion location and sensorimotor rhythms in stroke – a systematic review with narrative synthesis.Neurological Sciences, 44:4263 – 4289, 2023. 14 arXiv preprint

2023
[20]

Time-varying effective connectivity for describing the dynamic brain networks of post-stroke rehabilitation.Frontiers in Aging Neuroscience, 14, 2022

Fangzhou Xu, Yuandong Wang, Han Li, Xin Yu, Chongfeng Wang, Ming Liu, Lin Jiang, Chao Feng, Jianfei Li, De quan Wang, Zhiguo Yan, Yang Zhang, and Jiancai Leng. Time-varying effective connectivity for describing the dynamic brain networks of post-stroke rehabilitation.Frontiers in Aging Neuroscience, 14, 2022

2022
[21]

Rustamov, Joseph B

N. Rustamov, Joseph B. Humphries, A. Carter, and E. Leuthardt. Theta–gamma coupling as a cortical biomarker of brain–computer interface-mediated motor recovery in chronic stroke.Brain Communications, 4, 2022

2022
[22]

Müller-Putz, and Christa Neuper

Vera Kaiser, Ian Daly, Floriana Pichiorri, Donatella Mattia, Gernot R. Müller-Putz, and Christa Neuper. Relation- ship between electrical brain responses to motor imagery and motor impairment in stroke.Stroke, 43(10):2735– 2740, 2012

2012
[23]

Classification of motor imagery performance in acute stroke.Journal of Neural Engineering, 11(3):036001, apr 2014

Chayanin Tangwiriyasakul, Victor Mocioiu, Michel J A M van Putten, and Wim L C Rutten. Classification of motor imagery performance in acute stroke.Journal of Neural Engineering, 11(3):036001, apr 2014

2014
[24]

Ivana Kancheva, Sandra M. A. van der Salm, Nick F. Ramsey, and Mariska J. Vansteensel. Association between lesion location and sensorimotor rhythms in stroke – a systematic review with narrative synthesis.Neurological Sciences, 44(12):4263–4289, Dec 2023

2023
[25]

An maml-based lightweight neural network with domain adaptation for cross-subject and generalized eeg-based motor imagery classification.IEEE Sensors Letters, 10(2):1–4, 2026

Aaqib Raza and Mohd Zuki Yusoff. An maml-based lightweight neural network with domain adaptation for cross-subject and generalized eeg-based motor imagery classification.IEEE Sensors Letters, 10(2):1–4, 2026

2026
[26]

Buchl, Nathan A

Zafer Keser, Samuel C. Buchl, Nathan A. Seven, M. Markota, H. Clark, David T. Jones, G. Lanzino, Robert D. Brown, G. Worrell, and B. Lundstrom. Electroencephalogram (eeg) with or without transcranial magnetic stimulation (tms) as biomarkers for post-stroke recovery: A narrative review.Frontiers in Neurology, 13, 2022

2022
[27]

An eeg motor imagery dataset for brain computer interface in acute stroke patients.Scientific Data, 11(1):131, Jan 2024

Haijie Liu, Penghu Wei, Haochong Wang, Xiaodong Lv, Wei Duan, Meijie Li, Yan Zhao, Qingmei Wang, Xinyuan Chen, Gaige Shi, Bo Han, and Junwei Hao. An eeg motor imagery dataset for brain computer interface in acute stroke patients.Scientific Data, 11(1):131, Jan 2024

2024
[28]

Eeg data of motor imagery for stroke, 2 2019

Tianyu Jia. Eeg data of motor imagery for stroke, 2 2019

2019
[29]

Visualizing data using t-sne.Journal of Machine Learning Research, 9(86):2579–2605, 2008

Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne.Journal of Machine Learning Research, 9(86):2579–2605, 2008

2008
[30]

Dongyi He, Wai Ting Siok, and Nizhuan Wang. Toward robust, reproducible, and widely accessible intracranial language brain-computer interfaces: A comprehensive review of neural mechanisms, hardware, algorithms, evaluation, clinical pathways and future directions, 2026

2026
[31]

Yueyang Li, Weiming Zeng, Wenhao Dong, Di Han, Lei Chen, Hongyu Chen, Zijian Kang, Shengyu Gong, Hongjie Yan, Wai Ting Siok, and Nizhuan Wang. A tale of single-channel electroencephalography: Devices, datasets, signal processing, applications, and future directions.IEEE Transactions on Instrumentation and Measurement, 74:1–20, 2025

2025
[32]

A somatosensory erp-bci system for post-stroke hand function rehabilitation applications.IEEE Transactions on Instrumentation and Measurement, 75:1–12, 2026

Junlin Wang, Xiaodong Li, Xingyu Lu, Ningbo Fei, Wei Huang, and Yong Hu. A somatosensory erp-bci system for post-stroke hand function rehabilitation applications.IEEE Transactions on Instrumentation and Measurement, 75:1–12, 2026. 15

2026