DEMENTIA: a hybrid attention-based multimodal and multi-task learning framework with expert knowledge for Alzheimer’s disease assessment from speech
The DOI matches its Crossref record; 53 of the 53 references it cites were found in a registry or an index.
Abstract
The prevalence of Alzheimer's disease (AD) is rising annually, imposing a severe burden on patients and society. Therefore, assisted AD assessment is crucial. The decline in language function and the cognitive impairment it reflects are key external manifestations of AD. Many studies have utilized speech analysis to achieve convenient, non-invasive, and low-cost AD detection. Although state-of-the-art researches achieve high-precision AD detection using multimodal information, these studies often ignore interactions between different modalities and lack explanations for complex models. To address this, we propose a multi-task learning (MTL) AD assessment model that combines hybrid attention with multimodal representations. The model fuses audio, text, and expert knowledge to fully capture intra- and inter-modal interactions, achieving simultaneous AD detection and cognitive state prediction, along with comprehensive explainability analyses of the model and various modalities. Results show that the proposed method is sufficiently sensitive in assessing AD, achieving 89.58% accuracy and 91.67% recall for the classification task and a root mean square error of 4.31 for the regression task with good generalization performance. Multimodal representations with expert knowledge and MTL contribute to AD assessment performance. Explainability analyses indicate that, compared to healthy controls, AD patients exhibit slower speech rates, reduced syntactic complexity, and a greater tendency to use pause fillers and pronouns. Therefore, Our study validates the effectiveness of the proposed method, addressing trust issues in clinical practice for assisted decision-making and further advancing the development of speech as a promising biomarker for early AD screening and cognitive decline monitoring.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“SHAP was implemented in 4 studies spanning dementia detection [9], auditory verbal hallucination assessment [52], psychiatric disorder classification [51], and TBI [53].”
From Explainable and Interpretable AI for Voice and Speech Analysis in Clinical Care: Systematic Review (2026) · cites this paper as [9]
In the citing paper
“Zhang et al [9] and Fu et al [48] performed cross-dataset evaluation not as consistency analysis for explainability but as external validation for the underlying model.”
From Explainable and Interpretable AI for Voice and Speech Analysis in Clinical Care: Systematic Review (2026) · cites this paper as [9]
In the citing paper
“In dementia detection, SHAP and LIME analysis identified noun phrase rate, empty word rate, and hesitation ratio as influential linguistic and fluency features indicative of decreased lexical complexity and increased disfluency in AD speech, alongside voice-related features reflecting lower vocal energy [9].”
From Explainable and Interpretable AI for Voice and Speech Analysis in Clinical Care: Systematic Review (2026) · cites this paper as [9]
Citations
Cites
A Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707
Effective Approaches to Attention-based Neural Machine Translation
Luong, T., Pham, H., & Manning, C. D. (2015) · Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d15-1166 · Cited by 8,382 · cited 1×
“…as: MHA t → a = LN AT 1t → a , AT 2t → a , . . . , AT h t → a W t → a + O a Next, the above CM attention output between text and audio is concatenated to obtain fusion output F at ∈ R T × 256 , which is then fed into the global attention layer to calculate the Luong’s general attention scores [42].”
“Why Should I Trust You?”: Explaining the Predictions of Any Classifier
Ribeiro, M., Singh, S., & Guestrin, C. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations · doi:10.18653/v1/n16-3020 · Cited by 6,682
Dementia prevention, intervention, and care
Livingston, G., Sommerlad, A., Orgeta, V., Costafreda, S. G., Huntley, J., Ames, D., Ballard, C., Banerjee, S., Burns, A., Cohen-Mansfield, J., et al. (2017) · The Lancet · doi:10.1016/s0140-6736(17)31363-6 · Cited by 6,072
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019) · arXiv · doi:10.48550/arxiv.1910.01108 · Cited by 4,653
Multi-Task Learning for Dense Prediction Tasks: A Survey
Vandenhende, S., Georgoulis, S., Van Gansbeke, W., Proesmans, M., Dai, D., & Van Gool, L. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3054719 · Cited by 840
Linguistic Features Identify Alzheimer’s Disease in Narrative Speech
Fraser, K. C., Meltzer, J. A., & Rudzicz, F. (2015) · Journal of Alzheimer's Disease · doi:10.3233/jad-150520 · Cited by 797 · cited 1×
“Thus early cognitive decline in AD can be manifested through subtlechangesinspeechpatterns,providingauniquewindowfor observinganindividual’scognitivestate[11],[12].”
Vermunt, L., Sikkes, S. A. M., van den Hout, A., Handels, R., Bos, I., van der Flier, W. M., Kern, S., Ousset, P., Maruff, P., Skoog, I., et al. (2019) · Alzheimer's & Dementia · doi:10.1016/j.jalz.2019.04.001 · Cited by 570 · cited 1×
“On the other hand, the damage caused by AD is progressive and irreversible, with no effective treatment currently available [4], [5].”
Do as AI say: susceptibility in deployment of clinical decision-aids
Gaube, S., Suresh, H., Raue, M., Merritt, A., Berkowitz, S. J., Lermer, E., Coughlin, J. F., Guttag, J. V., Colak, E., & Ghassemi, M. (2021) · npj Digital Medicine · doi:10.1038/s41746-021-00385-9 · Cited by 543 · cited 1×
“Addressing the explainability and clinicians’ trust in AI decisions is an es-sentialprerequisiteforthesuccessfulapplicationofAIinclinicalpractice[2],[21].”
Chomsky, N. (2006) · Cambridge University Press eBooks · doi:10.1017/cbo9780511791222 · Cited by 430
Research works printed but not linked: 2
- Gal, 2017, Concrete dropout, 3584
- Maaten, 2008, Visualizing high-dimensional data t-SNE, J. Mach. Learn. Res., 9, 2579
Cited by
Transformer-Based Deep Learning Approaches for Speech-Based Dementia Detection: A Systematic Review
Mobtahej, P., Gouron, S., Javaheri, R., Le, D. N., El‐Khoury, A., Naing, T. H., & Sajjadi, S. A. (2025) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2025.3595781 · Cited by 9
Jin, Q., Jia, Q., Zhou, X., Jin, D., Song, X., Xie, Z., Alimujiang, A., Li, Y., Huang, J., & Ma, C. (2025) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2025.3597467 · Cited by 5
“Inaddition,weconductedanevaluationstudywithdomain expertstofurthervalidatethereliabilityoftheSHAPanalysis [40].”
Jamshed, M., Shahzad, A., & Kiseon, K. I. M. (2026) · IEEE Transactions on Neural Systems and Rehabilitation Engineering · doi:10.1109/tnsre.2026.3651786 · Cited by 2
“Speech and language-based biomarkers derived from spontaneous or task-driven verbal responses have been shown to reflect cognitive decline through alterations in lexical diversity, fluency, and prosody [4], [5].”
Lu, J., Xu, W. (., & Wen, Y. (2026) · Building and Environment · doi:10.1016/j.buildenv.2026.114603 · Cited by 1
Layer-Aware Early Fusion of Acoustic and Linguistic Embeddings for Cognitive Status Classification
Novotny, K., Moro-Velázquez, L., & Mekyska, J. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11463687 · Cited by 1
“Side-by-side evaluations report mixed outcomes for early or late combination, whereas methods that let one stream guide the other (such as cross-attention between speech and text) tend to be stronger and easier to inspect [4, 5, 6, 7].”
Metadata
- Authors
- Zhang, Z., Wang, T., Hu, Z., Yang, L.-Z., & Li, H.
- Affiliations
- Chinese Academy of Sciences (CN)
- Anhui Province Key Laboratory of Medical Physics and Technology, Institute of Health and Medical Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, China
- Venue
- IEEE Journal of Biomedical and Health Informatics
- Published
- 2025-04
- Volume
- 29
- Issue
- 4
- Pages
- 2957-2968
- ISSN
- 2168-2194 (ISSN-L)
- 2168-2208
- OpenAlex
- W4405521324
- Cited by
- 17
- References
- 53
- Citations by year
- 2026: 15 · 2025: 2
- FWCI
- 4.98
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Journal of Biomedical and Health Informatics (publishedVersion)
- PubMed (publishedVersion)
- Funders
- National Natural Science Foundation of China (82371931)
- Natural Science Foundation of Anhui Province (2408085MC056)
- HFIPS Director's Fund (YZJJ202207-TS)
- Anhui Provincial Key Research and Development Plan (202204295107020004)
- Topics
- Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Keywords
- Alzheimer's disease detection, speech analysis, multimodal fusion, hybrid attention mechanism, multi-task learning, expert knowledge integration, cognitive state prediction, explainable AI, speech rate, syntactic complexity, pronoun usage, speech biomarkers, early screening, cognitive decline monitoring
- MeSH terms
- Machine Learning, Aged, Aged, 80 and over, Alzheimer Disease, Female, Humans, Male, Speech
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: DEMENTIA: a hybrid attention-based multimodal and multi-task learning framework with expert knowledge for Alzheimer’s disease assessment from speech. https://app.cochat.ai/papers/10.1109/jbhi.2024.3509620
- The paper
- Zhang, Z., Wang, T., Hu, Z., Yang, L.-Z., & Li, H. (2025). DEMENTIA: a hybrid attention-based multimodal and multi-task learning framework with expert knowledge for Alzheimer’s disease assessment from speech. IEEE Journal of Biomedical and Health Informatics. https://doi.org/10.1109/jbhi.2024.3509620
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.