pith. sign in

arxiv: 2604.23964 · v1 · submitted 2026-04-27 · 💻 cs.LG · cs.AI

Task-guided Spatiotemporal Network with Diffusion Augmentation for EEG-based Dementia Diagnosis and MMSE Prediction

Xiaoyu Zheng , Xu Tian , Bin Jiao , Kunbo Cui , Hanhe Lin , Lu Shen , Jin Liu This is my paper

Pith reviewed 2026-05-08 04:35 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords EEGdementiaMMSEspatiotemporal networkdiffusion augmentationtask-guidedAlzheimer's diseasemulti-task learning
0
0 comments X

The pith

Task-guided spatiotemporal network separates dementia diagnosis and MMSE prediction tasks in EEG to boost performance.

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

The paper aims to solve feature entanglement in multi-task learning where EEG signals are used both to diagnose dementia types and to predict Mini-Mental State Examination scores. It introduces a network that fuses multi-band features, augments data via diffusion, applies gated spatiotemporal attention, and uses a task-guided query to extract features specific to each task. This design is evaluated on the XY02 dataset for binary and three-class dementia classification as well as regression of the MMSE score, with additional checks on DS004504 for generalization. A reader would care if this leads to more accurate, non-invasive tools for assessing cognitive decline. If the modules work as intended, joint modeling of related medical tasks from the same data source becomes more effective without one degrading the other.

Core claim

The authors develop TGSN that combines multi-band feature fusion, pre-trained diffusion augmentation, gated spatiotemporal attention, and a task-guided query module. This architecture captures spectral, spatial, and temporal EEG information while ensuring task-specific extraction to avoid interference between classification of dementia subtypes and prediction of MMSE scores. Experiments show it surpasses state-of-the-art methods with accuracies of 97.78% for AD/FTD and 83.93% for AD/FTD/VCI on XY02, plus RMSE values of 1.93 and 2.38 for MMSE, and demonstrates cross-dataset generalization.

What carries the argument

The task-guided query module, which enables task-specific feature extraction from the shared spatiotemporal representations to mitigate interference between diagnosis and score prediction objectives.

Load-bearing premise

The gains in accuracy and error reduction result specifically from the task separation provided by the query module and the diversity added by diffusion augmentation, as opposed to dataset peculiarities or baseline underperformance on the XY02 and DS004504 sets.

What would settle it

If an independent research group applies the same TGSN architecture to a new dementia EEG dataset and fails to observe similar accuracy gains or error reductions relative to standard methods, the claims of superior task handling would be challenged.

Figures

Figures reproduced from arXiv: 2604.23964 by Bin Jiao, Hanhe Lin, Jin Liu, Kunbo Cui, Lu Shen, Xiaoyu Zheng, Xu Tian.

Figure 1
Figure 1. Figure 1: Illustration of the challenge and our solution. (a) Con￾ventional multi-task learning suffers from feature entanglement when handling heterogeneous tasks. (b) The proposed TGSN employs a task-guided query module for task-specific feature extraction, thereby mitigating inter-task interference. for EEG-based dementia diagnosis and MMSE prediction, as shown in view at source ↗
Figure 2
Figure 2. Figure 2: Framework of the proposed TGSN. a. Multi-band feature fusion (MBFF) module: Performs EEG preprocessing, band decomposition, and feature extraction. b. Pre-trained data augmentation (PTDA) module: Employs a pre-trained strategy to generate EEG features, enhancing data diversity. c. Gated spatiotemporal attention (GSA) module: Models long-range spatial dependencies and temporal dynamics from multi￾channel EE… view at source ↗
Figure 3
Figure 3. Figure 3: t-SNE visualization of feature distributions for real and generated EEG features across three classes. characteristics observed in each group, while maintaining appropriate diversity. Overall, these results demonstrate that the PTDA produces high-quality EEG features that preserve the statistical properties of real data and enhance sample diversity. B. Quantitative analysis of GSA A sensitivity analysis wa… view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity analysis of model performance with varying numbers of GSA layers. Fp1 Fp2 F7 F3 Fz F4 F8 T3 C3 Cz C4 T4 T5 P3 Pz P4 T6 A1 A2 O1 O2 (a) AD/FTD (b) AD/VCI (c) FTD/VCI (d) AD/FTD/VCI Diagnosis attention Fp MMSE attention 1 Fp2 F7 F3 Fz F4 F8 T3 C3 Cz C4 T4 T5 P3 Pz P4 T6 A1 A2 O1 O2 Fp1 Fp2 F7 F3 Fz F4 F8 T3 C3 Cz C4 T4 T5 P3 Pz P4 T6 A1 A2 O1 O2 Fp1 Fp2 F7 F3 Fz F4 F8 T3 C3 Cz C4 T4 T5 P3 Pz P4 T… view at source ↗
Figure 5
Figure 5. Figure 5: Topographic maps of attention weights from the TGQ module. Columns represent diagnosis and MMSE attention, while rows correspond to (a) AD/FTD, (b) AD/VCI, (c) FTD/VCI, and (d) AD/FTD/VCI tasks. The maps illustrate the spatial distribution of task￾specific attention across EEG channels. Electrodes are grouped by anatomical regions: Fp1, Fp2, F3, F4, F7, F8, and Fz correspond to the frontal region; C3, C4, … view at source ↗
read the original abstract

Patients with dementia typically exhibit cognitive impairment, which is routinely assessed using the Mini-Mental State Examination (MMSE). Concurrently, their underlying neurophysiological abnormalities are reflected in Electroencephalography (EEG), providing a basis for joint modeling. However, traditional multi-task approaches suffer from feature entanglement, which leads to inter-task interference when handling heterogeneous objectives.To address this challenge, we propose a task-guided spatiotemporal network (TGSN) with diffusion augmentation for EEG-based dementia diagnosis and MMSE prediction. Specifically, TGSN integrates a multi-band feature fusion module to capture complementary spectral information from EEG. Meanwhile, a pre-trained data augmentation module utilizing a diffusion process is introduced toincrease sample diversity. To model the complex spatiotemporal patterns of EEG, we propose a gated spatiotemporal attention module that captures long-range spatial dependencies and temporal dynamics. Moreover, we design a task-guided query module to achieve task-specific feature extraction, thereby mitigating task interference. The effectiveness of TGSN is evaluated on the XY02 dataset. Experimental results demonstrate that the proposed network outperforms several state-of-the-art methods, achieving classification accuracies of 97.78\% for Alzheimer's Disease (AD)/Frontotemporal Dementia (FTD) and 83.93\% for AD/FTD/Vascular Cognitive Impairment (VCI), which exceed the best baselines by 16.39\% and 8.28\%, respectively. In parallel, it reduces the RMSE for MMSE prediction to 1.93 and 2.38, achieving significant error reductions of 1.44 and 1.43 compared to the best baselines. Additionally, validation on the DS004504 dataset demonstrates strong cross-dataset generalization...

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.

Referee Report

2 major / 1 minor

Summary. The paper proposes a task-guided spatiotemporal network (TGSN) with diffusion augmentation for joint EEG-based dementia classification (AD/FTD and AD/FTD/VCI) and MMSE prediction. It combines multi-band feature fusion, a pre-trained diffusion augmentation module, gated spatiotemporal attention, and a task-guided query module to reduce inter-task interference, claiming accuracies of 97.78% and 83.93% (outperforming baselines by 16.39% and 8.28%) plus RMSE reductions to 1.93 and 2.38 on the XY02 dataset, with additional cross-dataset validation on DS004504.

Significance. If the performance gains hold under rigorous subject-independent validation without leakage, the approach could advance multi-task EEG modeling for dementia by addressing feature entanglement and demonstrating practical cross-dataset generalization, which is valuable given the clinical importance of non-invasive dementia assessment.

major comments (2)
  1. [Abstract / Experimental evaluation] The abstract and experimental claims report large accuracy and RMSE improvements but provide no details on cross-validation strategy, subject-independent splits, statistical tests, ablation studies, or error bars. This directly affects the load-bearing performance claims in the abstract and results sections, as non-independent splits are a known source of leakage in EEG studies with high inter-subject variability.
  2. [Methods (diffusion augmentation) / Results] No information is given on whether the diffusion augmentation module was pre-trained on the same XY02 or DS004504 data used for the main TGSN training and evaluation. If overlap exists, this would introduce leakage that could explain the reported 16.39% and 8.28% margins without reflecting genuine task separation from the task-guided query module.
minor comments (1)
  1. [Abstract] Typo in abstract: 'toincrease sample diversity' should read 'to increase sample diversity'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comments point by point below and will revise the paper to improve transparency on validation procedures and data handling.

read point-by-point responses

  1. Referee: [Abstract / Experimental evaluation] The abstract and experimental claims report large accuracy and RMSE improvements but provide no details on cross-validation strategy, subject-independent splits, statistical tests, ablation studies, or error bars. This directly affects the load-bearing performance claims in the abstract and results sections, as non-independent splits are a known source of leakage in EEG studies with high inter-subject variability.

    Authors: We agree that explicit details on the evaluation protocol are essential. The full manuscript employs subject-independent 5-fold cross-validation on XY02 (ensuring no subject overlap between train and test folds) along with cross-dataset testing on DS004504. Results are averaged over multiple runs with standard deviation error bars, and paired statistical tests (Wilcoxon signed-rank) confirm significance of improvements. Ablation studies isolating the task-guided query module appear in Section 4.3. We will expand the abstract, methods, and results sections to explicitly describe the subject-independent splits, statistical tests, error bars, and ablations. revision: yes

  2. Referee: [Methods (diffusion augmentation) / Results] No information is given on whether the diffusion augmentation module was pre-trained on the same XY02 or DS004504 data used for the main TGSN training and evaluation. If overlap exists, this would introduce leakage that could explain the reported 16.39% and 8.28% margins without reflecting genuine task separation from the task-guided query module.

    Authors: The diffusion augmentation module was pre-trained on a disjoint public EEG corpus with no overlap to the XY02 or DS004504 subjects or recordings used for TGSN training/evaluation. This design avoids leakage while increasing sample diversity. Ablation results in the manuscript already isolate the contribution of the task-guided query module from the augmentation. We will revise the methods section to explicitly state the pre-training data source, confirm the lack of overlap, and add a note on leakage prevention. revision: yes

Circularity Check

0 steps flagged

No circularity detected; claims rest on empirical evaluation of a proposed architecture

full rationale

The manuscript describes an empirical machine-learning architecture (TGSN) consisting of defined modules (multi-band feature fusion, diffusion-based augmentation, gated spatiotemporal attention, task-guided query) and reports classification accuracies plus RMSE values as direct experimental outcomes on the XY02 and DS004504 datasets. No first-principles derivations, closed-form predictions, or parameter-fitting steps are presented that could reduce by construction to the inputs themselves. Performance margins are framed as comparisons against external baselines rather than quantities forced by self-referential definitions or load-bearing self-citations. The derivation chain is therefore self-contained as a standard supervised-learning pipeline whose validity is tested externally on held-out data.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 2 invented entities

The central claims rest on standard deep-learning assumptions plus several new modules whose effectiveness is asserted without external grounding. Free parameters include all network weights and diffusion hyperparameters fitted during training. Invented entities are the task-guided query module and gated spatiotemporal attention module.

free parameters (2)
  • Network hyperparameters and weights
    All trainable parameters of the multi-band fusion, attention, and task-guided modules are fitted to the EEG data.
  • Diffusion model parameters
    Pre-trained diffusion augmentation module contains parameters fitted on EEG or related data.
axioms (2)
  • domain assumption EEG signals contain sufficient discriminative information for both dementia subtype classification and MMSE regression
    Invoked in the problem formulation and evaluation design.
  • domain assumption The XY02 and DS004504 datasets are representative and free of major labeling or acquisition biases
    Required for the generalization and cross-dataset claims.
invented entities (2)
  • Task-guided query module no independent evidence
    purpose: To extract task-specific features and reduce inter-task interference
    New module introduced to address feature entanglement; no independent evidence outside the reported experiments.
  • Gated spatiotemporal attention module no independent evidence
    purpose: To capture long-range spatial dependencies and temporal dynamics in EEG
    Novel combination presented as core modeling component; effectiveness asserted via end-to-end results.

pith-pipeline@v0.9.0 · 5621 in / 1810 out tokens · 27776 ms · 2026-05-08T04:35:17.686729+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 1 canonical work pages

  1. [1]

    Diagnosis and manage- ment of dementia,

    Z. Arvanitakis, R. C. Shah, and D. A. Bennett, “Diagnosis and manage- ment of dementia,”JAMA, vol. 322, no. 16, pp. 1589–1599, 2019

    2019

  2. [2]

    Distinct network topology in Alzheimer’s disease and behavioral variant fron- totemporal dementia,

    A. S. L. Ng, J. Wang, K. K. Ng, J. S. X. Chong, X. Qian, J. K. W. Lim, Y . J. Tan, A. C. W. Yong, R. J. Chander, S. Hameedet al., “Distinct network topology in Alzheimer’s disease and behavioral variant fron- totemporal dementia,”Alzheimer’s Research & Therapy, vol. 13, pp. 1–16, 2021

    2021

  3. [3]

    Navigating the dementia landscape: Biomark- ers and emerging therapies,

    S. Maheshwari, A. Singh, V . A. Ansari, T. Mahmood, R. Wasim, J. Akhtar, and A. Verma, “Navigating the dementia landscape: Biomark- ers and emerging therapies,”Ageing Research Reviews, vol. 94, p. 102193, 2024

    2024

  4. [4]

    Comparing clinical features of behavioral variant frontotemporal dementia and Alzheimer’s disease using network analysis,

    G. J. Goodwin, S. Mehrzad, J. L. Cummings, B. N. Renn, J. W. Kinney, and S. E. John, “Comparing clinical features of behavioral variant frontotemporal dementia and Alzheimer’s disease using network analysis,”Alzheimer’s & Dementia, vol. 21, no. 6, p. e70361, 2025

    2025

  5. [5]

    Mini-Mental State Examination (MMSE) for the early detection of dementia in people with mild cognitive impairment (MCI),

    I. Arevalo-Rodriguez, N. Smailagic, M. Roqu ´e-Figuls, A. Ciapponi, E. Sanchez-Perez, A. Giannakou, O. L. Pedraza, X. B. Cosp, and S. Cullum, “Mini-Mental State Examination (MMSE) for the early detection of dementia in people with mild cognitive impairment (MCI),” Cochrane Database of Systematic Reviews, no. 7, 2021

    2021

  6. [6]

    Mini-Mental State Examination (MMSE) for the detection of dementia in clinically unevaluated people aged 65 and over in community and primary care populations,

    S. T. Creavin, S. Wisniewski, A. H. Noel-Storr, C. M. Trevelyan, T. Hampton, D. Rayment, V . M. Thom, K. J. Nash, H. Elhamoui, R. Milliganet al., “Mini-Mental State Examination (MMSE) for the detection of dementia in clinically unevaluated people aged 65 and over in community and primary care populations,”Cochrane Database of Systematic Reviews, no. 1, 2016

    2016

  7. [7]

    Emerging trends in early demen- tia diagnosis: An analysis on advanced machine learning approaches,

    B. Gami, M. Agrawal, and R. Katarya, “Emerging trends in early demen- tia diagnosis: An analysis on advanced machine learning approaches,” ACM Computing Surveys, vol. 58, no. 4, pp. 1–35, 2025

    2025

  8. [8]

    Transformer-based spatial-temporal feature learning for eeg decoding.arXiv preprint arXiv:2106.11170, 2021

    Y . Song, X. Jia, L. Yang, and L. Xie, “Transformer-based spatial-temporal feature learning for EEG decoding,”arXiv preprint arXiv:2106.11170, 2021

    work page arXiv 2021

  9. [9]

    Quantitative EEG as a predictive biomarker for Parkinson disease dementia,

    B. Klassen, J. Hentz, H. Shill, E. Driver-Dunckley, V . Evidente, M. Sab- bagh, C. Adler, and J. Caviness, “Quantitative EEG as a predictive biomarker for Parkinson disease dementia,”Neurology, vol. 77, no. 2, pp. 118–124, 2011

    2011

  10. [10]

    Multi-scale adaptive graph learning based on multi-wave EEG data for dementia diagnosis,

    J. Wei, B. Jiao, H. Lin, X. Tian, L. Shen, and J. Liu, “Multi-scale adaptive graph learning based on multi-wave EEG data for dementia diagnosis,”Big Data Mining and Analytics, vol. 8, no. 5, pp. 1023– 1043, 2025

    2025

  11. [11]

    Alzheimer’s disease and frontotemporal dementia: A robust classification method of EEG signals and a comparison of validation methods,

    A. Miltiadous, K. D. Tzimourta, N. Giannakeas, M. G. Tsipouras, T. Afrantou, P. Ioannidis, and A. T. Tzallas, “Alzheimer’s disease and frontotemporal dementia: A robust classification method of EEG signals and a comparison of validation methods,”Diagnostics, vol. 11, no. 8, p. 1437, 2021

    2021

  12. [12]

    Early detection of Alzheimer’s disease from EEG signals using Hjorth parameters,

    M. S. Safi and S. M. M. Safi, “Early detection of Alzheimer’s disease from EEG signals using Hjorth parameters,”Biomedical Signal Process- ing and Control, vol. 65, p. 102338, 2021

    2021

  13. [13]

    A novel electroencephalog- raphy based approach for Alzheimer’s disease and mild cognitive im- pairment detection,

    B. Oltu, M. F. Aks ¸ahin, and S. Kibaro ˘glu, “A novel electroencephalog- raphy based approach for Alzheimer’s disease and mild cognitive im- pairment detection,”Biomedical Signal Processing and Control, vol. 63, p. 102223, 2021

    2021

  14. [14]

    Multi-threshold recurrence rate plot: a novel methodology for EEG analysis in Alzheimer’s disease and frontotemporal dementia,

    H. Zheng, X. Xiong, and X. Zhang, “Multi-threshold recurrence rate plot: a novel methodology for EEG analysis in Alzheimer’s disease and frontotemporal dementia,”Brain Sciences, vol. 14, no. 6, p. 565, 2024

    2024

  15. [15]

    Prediction of Mini-Mental State Examination scores via machine learning for Alzheimer’s disease,

    S. Luo, X. Yao, Z. Yuan, and L. Zhou, “Prediction of Mini-Mental State Examination scores via machine learning for Alzheimer’s disease,” inInternational Conference on Image, Vision and Intelligent Systems, vol. 13. Springer, 2022, pp. 878–883

    2022

  16. [16]

    Ranking convolutional neural network for Alzheimer’s disease Mini-Mental State Examination prediction at multiple time-points,

    H. Qiao, L. Chen, and F. Zhu, “Ranking convolutional neural network for Alzheimer’s disease Mini-Mental State Examination prediction at multiple time-points,”Computer Methods and Programs in Biomedicine, vol. 213, p. 106503, 2022

    2022

  17. [17]

    Emotion recognition from EEG based on multi-task learning with capsule network and attention mechanism,

    C. Li, B. Wang, S. Zhang, Y . Liu, R. Song, J. Cheng, and X. Chen, “Emotion recognition from EEG based on multi-task learning with capsule network and attention mechanism,”Computers in Biology and Medicine, vol. 143, p. 105303, 2022

    2022

  18. [18]

    MTLFuseNet: a novel emotion recognition model based on deep latent feature fusion of EEG signals and multi-task learning,

    R. Li, C. Ren, Y . Ge, Q. Zhao, Y . Yang, Y . Shi, X. Zhang, and B. Hu, “MTLFuseNet: a novel emotion recognition model based on deep latent feature fusion of EEG signals and multi-task learning,”Knowledge- Based Systems, vol. 276, p. 110756, 2023

    2023

  19. [19]

    Neural biomarker diagnosis and prediction to mild cognitive impairment and Alzheimer’s disease using EEG technology,

    B. Jiao, R. Li, H. Zhou, K. Qing, H. Liu, H. Pan, Y . Lei, W. Fu, X. Wang, X. Xiaoet al., “Neural biomarker diagnosis and prediction to mild cognitive impairment and Alzheimer’s disease using EEG technology,” Alzheimer’s Research & Therapy, vol. 15, no. 1, p. 32, 2023

    2023

  20. [20]

    Diagnosis of Alzheimer’s disease and frontotemporal dementia from electroencephalography signals,

    D.-G. Lee and S.-B. Lee, “Diagnosis of Alzheimer’s disease and frontotemporal dementia from electroencephalography signals,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 33, pp. 2160–2169, 2025

    2025

  21. [21]

    A novel multi-modal machine learning based approach for automatic classification of EEG recordings in dementia,

    C. Ieracitano, N. Mammone, A. Hussain, and F. C. Morabito, “A novel multi-modal machine learning based approach for automatic classification of EEG recordings in dementia,”Neural Networks, vol. 123, pp. 176–190, 2020

    2020

  22. [22]

    Smart-data-driven system for Alzheimer disease detection through electroencephalographic signals,

    T. Ara ´ujo, J. P. Teixeira, and P. M. Rodrigues, “Smart-data-driven system for Alzheimer disease detection through electroencephalographic signals,”Bioengineering, vol. 9, no. 4, p. 141, 2022

    2022

  23. [23]

    Multi-feature fusion learning for Alzheimer’s disease prediction using EEG signals in resting state,

    Y . Chen, H. Wang, D. Zhang, L. Zhang, and L. Tao, “Multi-feature fusion learning for Alzheimer’s disease prediction using EEG signals in resting state,”Frontiers in Neuroscience, vol. 17, p. 1272834, 2023

    2023

  24. [24]

    STEADYNet: Spatiotemporal EEG analysis for dementia detection using convolutional neural network,

    P. H. Kachare, S. B. Sangle, D. V . Puri, M. M. Khubrani, and I. Al- Shourbaji, “STEADYNet: Spatiotemporal EEG analysis for dementia detection using convolutional neural network,”Cognitive Neurodynam- ics, vol. 18, no. 5, pp. 3195–3208, 2024

    2024

  25. [25]

    A deep learning model for the detection of various dementia and MCI pathologies based on resting-state electroencephalography data: A retrospective multicentre study,

    Y . Watanabe, Y . Miyazaki, M. Hata, R. Fukuma, Y . Aoki, H. Kazui, T. Araki, D. Taomoto, Y . Satake, T. Suehiroet al., “A deep learning model for the detection of various dementia and MCI pathologies based on resting-state electroencephalography data: A retrospective multicentre study,”Neural Networks, vol. 171, pp. 242–250, 2024

    2024

  26. [26]

    SpectroCVT-Net: A convolutional vision transformer architecture and channel attention for classifying Alzheimer’s disease using spectro- grams,

    M. A. Bravo-Ortiz, E. Guevara-Navarro, S. A. Holgu ´ın-Garc´ıa, M. Rivera-Garcia, O. Cardona-Morales, G. A. Ruz, and R. Tabares- Soto, “SpectroCVT-Net: A convolutional vision transformer architecture and channel attention for classifying Alzheimer’s disease using spectro- grams,”Computers in Biology and Medicine, vol. 181, p. 109022, 2024

    2024

  27. [27]

    Understanding the role of self-attention in a Transformer model for the discrimination of SCD from MCI using resting-state EEG,

    E. Sibilano, D. Buongiorno, M. Lassi, A. Grippo, V . Bessi, S. Sorbi, A. Mazzoni, V . Bevilacqua, and A. Brunetti, “Understanding the role of self-attention in a Transformer model for the discrimination of SCD from MCI using resting-state EEG,”IEEE Journal of Biomedical and Health Informatics, vol. 28, no. 6, pp. 3422–3433, 2024

    2024

  28. [28]

    Ranking the importance of spatiotemporal win- dows of EEG signals results in a better Alzheimer’s disease prediction,

    R. Taub and Y . Savir, “Ranking the importance of spatiotemporal win- dows of EEG signals results in a better Alzheimer’s disease prediction,” in2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, 2024, pp. 1–6

    2024

  29. [29]

    Weakly supervised deep learning for brain disease prognosis using MRI and incomplete clinical scores,

    M. Liu, J. Zhang, C. Lian, and D. Shen, “Weakly supervised deep learning for brain disease prognosis using MRI and incomplete clinical scores,”IEEE Transactions on Cybernetics, vol. 50, no. 7, pp. 3381– 3392, 2019

    2019

  30. [30]

    Machine learning in the prediction of post-stroke cogni- tive impairment: a systematic review and meta-analysis,

    X. Li, Z. Chen, H. Jiao, B. Wang, H. Yin, L. Chen, H. Shi, Y . Yin, and D. Qin, “Machine learning in the prediction of post-stroke cogni- tive impairment: a systematic review and meta-analysis,”Frontiers in Neurology, vol. 14, p. 1211733, 2023

    2023

  31. [31]

    Predicting cognitive impairment in outpatients with epilepsy using machine learning techniques,

    F. Lin, J. Han, T. Xue, J. Lin, S. Chen, C. Zhu, H. Lin, X. Chen, W. Lin, and H. Huang, “Predicting cognitive impairment in outpatients with epilepsy using machine learning techniques,”Scientific Reports, vol. 11, no. 1, p. 20002, 2021

    2021

  32. [32]

    An ensemble learning model for continuous cognition assessment based on resting-state EEG,

    J. Sun, Y . Sun, A. Shen, Y . Li, X. Gao, and B. Lu, “An ensemble learning model for continuous cognition assessment based on resting-state EEG,” npj Aging, vol. 10, no. 1, p. 1, 2024

    2024

  33. [33]

    Multi-task learning for dense prediction tasks: A survey,

    S. Vandenhende, S. Georgoulis, W. Van Gansbeke, M. Proesmans, D. Dai, and L. Van Gool, “Multi-task learning for dense prediction tasks: A survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 7, pp. 3614–3633, 2021

    2021

  34. [34]

    Multi-task learning for Alzheimer’s disease diagnosis and Mini-Mental State Examination score prediction,

    J. Liu, X. Tian, H. Lin, H.-D. Li, and Y . Pan, “Multi-task learning for Alzheimer’s disease diagnosis and Mini-Mental State Examination score prediction,”Big Data Mining and Analytics, vol. 7, no. 3, pp. 828–842, 2024

    2024

  35. [35]

    GMSS: Graph-based multi-task self-supervised learning for EEG emotion recognition,

    Y . Li, J. Chen, F. Li, B. Fu, H. Wu, Y . Ji, Y . Zhou, Y . Niu, G. Shi, and W. Zheng, “GMSS: Graph-based multi-task self-supervised learning for EEG emotion recognition,”IEEE Transactions on Affective Computing, vol. 14, no. 3, pp. 2512–2525, 2022. 12 GENERIC COLORIZED JOURNAL, VOL. XX, NO. XX, XXXX 2026

    2022

  36. [36]

    A dual-branch spatio-temporal-spectral transformer feature fusion network for EEG- based visual recognition,

    J. Luo, W. Cui, S. Xu, L. Wang, X. Li, X. Liao, and Y . Li, “A dual-branch spatio-temporal-spectral transformer feature fusion network for EEG- based visual recognition,”IEEE Transactions on Industrial Informatics, vol. 20, no. 2, pp. 1721–1731, 2023

    2023

  37. [37]

    Gated parallel feature fusion multi-task learning for motor imagery EEG classification,

    X. Wang, V . Liesaputra, and Z. Huang, “Gated parallel feature fusion multi-task learning for motor imagery EEG classification,”Expert Sys- tems with Applications, vol. 293, p. 128678, 2025

    2025

  38. [38]

    Explainable identification of dementia from transcripts using transformer networks,

    L. Ilias and D. Askounis, “Explainable identification of dementia from transcripts using transformer networks,”IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 8, pp. 4153–4164, 2022

    2022

  39. [39]

    Differentiating between Alzheimer’s disease and frontotemporal dementia based on the resting-state multilayer EEG network,

    Y . Si, R. He, L. Jiang, D. Yao, H. Zhang, P. Xu, X. Ma, L. Yu, and F. Li, “Differentiating between Alzheimer’s disease and frontotemporal dementia based on the resting-state multilayer EEG network,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 31, pp. 4521–4527, 2023

    2023

  40. [40]

    Complexity analysis of the biomedical signal using fuzzy entropy measurement,

    H.-B. Xie, W.-T. Chen, W.-X. He, and H. Liu, “Complexity analysis of the biomedical signal using fuzzy entropy measurement,”Applied Soft Computing, vol. 11, no. 2, pp. 2871–2879, 2011

    2011

  41. [41]

    The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified peri- odograms,

    P. Welch, “The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified peri- odograms,”IEEE Transactions on Audio and Electroacoustics, vol. 15, no. 2, pp. 70–73, 2003

    2003

  42. [42]

    SegFormer: Simple and efficient design for semantic segmentation with transformers,

    E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, and P. Luo, “SegFormer: Simple and efficient design for semantic segmentation with transformers,”Advances in Neural Information Processing Systems, vol. 34, pp. 12 077–12 090, 2021

    2021

  43. [43]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in Neural Information Processing Systems, vol. 30, 2017

    2017

  44. [44]

    A dataset of EEG recordings from: Alzheimer’s disease, Frontotemporal dementia and Healthy subjects,

    A. Miltiadous, K. D. Tzimourta, T. Afrantou, P. Ioannidis, N. Grigo- riadis, D. G. Tsalikakis, P. Angelidis, M. G. Tsipouras, E. Glavas, N. Giannakeaset al., “A dataset of EEG recordings from: Alzheimer’s disease, Frontotemporal dementia and Healthy subjects,”OpenNeuro, vol. 1, p. 88, 2023

    2023

  45. [45]

    A novel CNN-based framework for Alzheimer’s disease detection using EEG spectrogram representations,

    K. Stefanou, K. D. Tzimourta, C. Bellos, G. Stergios, K. Markoglou, E. Gionanidis, M. G. Tsipouras, N. Giannakeas, A. T. Tzallas, and A. Miltiadous, “A novel CNN-based framework for Alzheimer’s disease detection using EEG spectrogram representations,”Journal of Person- alized Medicine, vol. 15, no. 1, p. 27, 2025

    2025

  46. [46]

    Decoding dementia: Classifying Alzheimer’s and frontotemporal dementia with EEG and Riemannian geometry,

    A. V . D. Kerchove, T. Mlinari ˇc, B. Verovnik, Z. I. Barinaga, and M. V . Hulle, “Decoding dementia: Classifying Alzheimer’s and frontotemporal dementia with EEG and Riemannian geometry,”Alzheimer’s & Demen- tia, vol. 21, p. e105377, 2025

    2025

  47. [47]

    Multi-dimensional EEG analysis reveals distinct neurophysiological patterns in Alzheimer’s and frontotemporal dementia,

    G. Cai, Y . Shi, J. Ma, X. Zhang, and W. Wu, “Multi-dimensional EEG analysis reveals distinct neurophysiological patterns in Alzheimer’s and frontotemporal dementia,”Journal of NeuroEngineering and Rehabili- tation, vol. 23, p. 81, 2026

    2026

  48. [48]

    Visualizing data using t-SNE

    L. Van der Maaten and G. Hinton, “Visualizing data using t-SNE.” Journal of Machine Learning Research, vol. 9, no. 11, 2008

    2008