The DOI matches its Crossref record; 183 of the 183 references it cites were found in a registry or an index; 110 of the 167 sources its claims cite were checked for support.
Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine
Open accessAbstract
Cognitive disorders, with dementia as a primary exemplar, present profound diagnostic and therapeutic challenges due to their complex pathologies and heterogeneous presentations. Artificial intelligence (AI), particularly when applied to multimodal neuroimaging and clinical data, offers a powerful approach to advancing precision medicine in this domain. This comprehensive review first examines foundational AI algorithms, including artificial neural networks for feature extraction, multimodal fusion strategies (e.g. early, intermediate, and late fusion) for data integration, and explainable AI (XAI) techniques to enhance clinical transparency. The core focus is on the application of these multimodal AI frameworks across the dementia care continuum, encompassing improved differential diagnosis, early detection through presymptomatic biomarkers, development of predictive models for disease progression, and optimization of patient stratification for clinical trials. Despite significant advances, persistent challenges remain, including limited generalizability across populations and protocols, data scarcity for non-Alzheimer’s dementias and prodromal stages—exacerbated by demographic biases—and barriers to interpretability. We discuss solutions such as federated learning for privacy-preserving data sharing and advanced XAI techniques. Finally, we outline pivotal future directions, including intelligent sensor fusion for discovering novel early biomarkers, hybrid AI architectures combining generative and discriminative models, innovations for handling missing modalities, and robust multicenter data integration frameworks. By synthesizing these advances, this review highlights the role of multimodal AI in advancing precise diagnosis, early prediction, and therapeutic development for neurodegenerative and vascular cognitive disorders, while identifying key translational challenges for precision medicine.
Overview
CoChat Summary
TL;DR
This comprehensive review synthesizes AI algorithms and multimodal neuroimaging frameworks for cognitive disorders, spanning differential diagnosis, early detection, progression prediction, and clinical trial stratification. It highlights substantial advances alongside persistent challenges in generalizability, data scarcity and demographic bias, and interpretability, and outlines solutions such as federated learning, explainable AI, and missing-modality fusion.
Key findings
- Multimodal integration generally outperforms unimodal approaches: combining MRI, FDG-PET, and CSF yielded 5–10% accuracy gains over MRI alone (Kohannim et al., 2010), while Aβ-PET/tau-PET/FDG-PET individually achieved 81%/85%/88% accuracy versus 95% when all three were integrated (Dang et al., 2023).
- AI supports differential diagnosis and early detection: MC-004 EEG discriminated DLB from AD with 72.2% sensitivity, 85.7% specificity, and 79.5% accuracy; an FDG-PET deep learning model predicted AD diagnosis 6.3 years before clinical manifestation with 82% specificity and 100% sensitivity; a 10-etiology multimodal model achieved AUC 0.94 for NC/MCI/dementia discrimination and AUC 0.96 for etiology differentiation (Xue et al., 2024).
- Progression prediction and stratification show moderate to strong performance: MCI-to-AD conversion models typically peak at AUC 0.80–0.85; hippocampal covariance network signatures improved early MCI prediction from AUC 0.72 to 0.79 with 86% sensitivity; and an AI model recommending cognitive-enhancing therapy was associated with a 0.60-point MMSE reduction versus 2.80 points in nonadherent patients.
- Key challenges include generalization failure—neuroimaging classifiers dropped from 73% accuracy within training hospitals to 55% at unseen sites—and data scarcity/demographic underrepresentation; federated learning raised balanced accuracy from 0.51 to 0.80 for smaller speech-based datasets, while XAI methods such as Grad-CAM and SHAP and intermediate fusion are proposed to improve trust and handle missing modalities.
Methods
- Design: comprehensive review of AI algorithms and multimodal neuroimaging frameworks applied to cognitive disorders, including dementia as the primary exemplar.
- Data or sample: no new primary cohort; the paper synthesizes published studies, tables, and cited literature across differential diagnosis, early detection, progression prediction, and patient stratification.
- Main analysis: narrative synthesis of artificial neural networks, CNNs, GCNs, GANs, multimodal fusion strategies (early, intermediate, late), explainable AI techniques, clinical applications, challenges, and future directions.
Limitations
- The paper does not state limitations of its own review methodology.
- Evident from the text: as a comprehensive review, it does not report a systematic search strategy, inclusion criteria, or quantitative meta-analysis, so its conclusions depend on the cited literature and may be subject to selection or synthesis bias.
- The review itself notes field-level limitations: limited generalizability across populations and protocols, data scarcity for non-Alzheimer’s dementias and prodromal stages, demographic biases, and barriers to interpretability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures

![Figure 2: Multimodal AI neuroimaging framework for cognitive disorders. The current AI-powered framework integrates multimodal neuroimaging data [e.g. structural MRI (sMRI), functional MRI (fMRI), diffusion tensor imaging (DTI), positron emission tomography (PET), and electroencephalography (EEG) data], which are primarily used for the early detection of cognitive impairment, the differential diagnosis of dementia, the prediction of disease progression trajectories, and population risk stratification. By enabling full-spectrum intelligent diagnostics and therapeutics, this approach optimizes intervention systems for cognitive disorders.](https://figures.cochat.ai/paper-figures/3fdaa10bfc474cc1b27db15d3eceaf97/multimodal-neuroimaging-and-ai-integration-in-cognitive-disorders-advances-figure-2.jpg)

Verification
Claims the paper makes
90 Valid · 4 Not valid · 44 Not checked · 67 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“exemplified by Korolev’s 3D-CNN reaching 80% accuracy with raw MRI data despite limited samples”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > AI algorithms and the framework applied to neuroimaging > Artificial neural networks
What the sources say
- Partly supportedcites Korolev et al., 2017 sourceExperimental — accuracy not yet measured
Table 1 reports approximately 80% accuracy for the ResNet on the binary Alzheimer’s disease-versus-normal-cohort task, and the Data section describes a dataset of 231 images. But the Data section says the MRI images were aligned, skull-stripped, and normalized—not raw. The 80% result also does not describe accuracy across all six classification tasks. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“GANs can be used to stratify 27 402 undiagnosed individuals into subgroups with shared variant patterns linked to genetic profiles and cognitive trajectories”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Application of multimodal AI in cognitive impairment management > Prediction of disease progression
What the sources say
- Supportedcites Skampardoni et al., 2024 source
The Introduction says the study used Smile-GAN to identify brain-structure subgroups. The Findings and Results report 3 subgroups among 27,402 people without diagnosed cognitive impairment, with distinct genetic associations and trajectories of cognitive decline. The subgroups were defined by brain measures, with genetic variants associated afterward. (LLM, conf=0.90)
plainstates an associationn = 27,402- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“a multi-cohort study leveraging ML algorithms and survival analysis methods established visuospatial dysfunction as a novel biomarker for predicting the progression of PD-MCI and subjective cognitive decline”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Application of multimodal AI in cognitive impairment management > Prediction of disease progression
What the sources say
- Partly supportedcites Loo et al., 2025 sourceExperimental — accuracy not yet measured
The Results section identifies Benton JLO visuospatial performance as a top predictor of PD-MCI and time-to-PD-MCI, and ranks it highly for SCD classification. It does not establish visuospatial dysfunction as a predictor of time-to-SCD progression or call it a novel biomarker; the Discussion says the finding is consistent with previous research. No page numbers are available in our copy. (LLM, conf=0.88)
boostedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“baseline CSF biomarkers, MRI measures, and APOE status can predict symptom emergence at 5-, 7-, and 10-year intervals at the individual level”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Application of multimodal AI in cognitive impairment management > Prediction of disease progression
What the sources say
- Supportedcites Albert et al., 2018 source
The Discussion states that baseline biomarkers predicted, at the individual level, who would develop clinical symptoms at 5, 7, or 10 years. The Results describe models using APOE status, MRI measures, and CSF biomarkers at those intervals; the models also included demographic and cognitive measures. (LLM, conf=0.94)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“cognitively normal older adults seeking clinical evaluations experience elevated MCI risks, particularly when subtle neuropsychological declines coincide with imaging-verified atrophy”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Application of multimodal AI in cognitive impairment management > Prediction of disease progression
What the sources say
- Supportedcites Y. Chen et al., 2017 source
The Discussion reports a substantially higher MCI conversion risk among clinically recruited, cognitively normal older adults than among community-recruited adults. In Results > Predictors of conversion (page not provided), worse baseline executive function and smaller total brain volume were independently associated with conversion in a joint model. The paper supports these as coexisting risk indicators, without claiming an interaction between them. (LLM, conf=0.88)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
112 more claims · 47 with their cited sources checked
Create a free account to see all 205 claims and to verify the 7 references still open.
- It may state two separate claims as one. 73
- Probably not a substantial claim. 21
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Application of multimodal AI in cognitive impairment management > Differential diagnosis of dementia
“Kohannim et al. (2010) demonstrated that combining MRI, FDG-PET, and cerebrospinal fluid (CSF) biomarkers yielded 5–10% accuracy gains over MRI alone.”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Application of multimodal AI in cognitive impairment management > Differential diagnosis of dementia
“Unimodal (MRI alone): AD vs CN = 79.07%, MCI vs CN = 71.21%; multimodal (MRI + FDG-PET + CSF): AD vs CN = 90.70%, MCI vs CN = 75.76%”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > Future directions and prospects > Multicentric data integration and missing modality innovation
“In contrast, intermediate fusion architectures circumvent the need for explicit data generation altogether (Guarrasi et al., 2025).”
§ Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine > AI algorithms and the framework applied to neuroimaging > Multimodal fusion strategies
“A representative example is the Joint Image Synthesis and Classification (JISCL) framework, which employs a multiscale GAN to bidirectionally synthesize MR and PET images (Z. Chen et al., 2025).”
Citations
Cites
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.74 · Cited by 24,513
A Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707
Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., Downey, P., Elliott, P., Green, J., Landray, M., et al. (2015) · PLOS Medicine · doi:10.1371/journal.pmed.1001779 · Cited by 12,728
Self-citationFrontotemporal lobar degeneration: A consensus on clinical diagnostic criteria
Neary, D., Snowden, J. S., Gustafson, L., Passant, U., Stuss, D., Black, S., Freedman, M., Kertesz, A., Robert, P. H., Albert, M., et al. (1998) · Neurology · doi:10.1212/wnl.51.6.1546 · Cited by 5,047
Petersen, R. C., Stevens, J. C., Ganguli, M., Tangalos, E. G., Cummings, J. L., & DeKosky, S. T. (2001) · Neurology · doi:10.1212/wnl.56.9.1133 · Cited by 2,065
Advances in natural language processing
Hirschberg, J., & Manning, C. D. (2015) · Science · doi:10.1126/science.aaa8685 · Cited by 1,712
Age-associated cognitive decline
Deary, I. J., Corley, J., Gow, A. J., Harris, S. E., Houlihan, L. M., Marioni, R. E., Penke, L., Rafnsson, S. B., & Starr, J. M. (2009) · British Medical Bulletin · doi:10.1093/bmb/ldp033 · Cited by 1,295
Harmonization of multi-site diffusion tensor imaging data
Fortin, J.-P., Parker, D., Tunç, B., Watanabe, T., Elliott, M. A., Ruparel, K., Roalf, D. R., Satterthwaite, T. D., Gur, R. C., Gur, R. E., et al. (2017) · NeuroImage · doi:10.1016/j.neuroimage.2017.08.047 · Cited by 1,261
Mueller, S. G., Weiner, M. W., Thal, L. J., Petersen, R. C., Jack, C. R., Jagust, W., Trojanowski, J. Q., Toga, A. W., & Beckett, L. (2005) · Alzheimer's & Dementia · doi:10.1016/j.jalz.2005.06.003 · Cited by 1,233
Deep multimodal learning: a survey on recent advances and trends
Ramachandram, D., & Taylor, G. W. (2017) · IEEE Signal Processing Magazine · doi:10.1109/msp.2017.2738401 · Cited by 1,139
Cited by
AI in Neurological Health Care: Qualitative Study of Patient and Public Perceptions
Bedenik, T., Fennelly, O., & Bennett, K. (2026) · Journal of Medical Internet Research · doi:10.2196/87806
Murthy, M. K. (2026) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2026.111921
Shateri, A., & Rezaei, S. (2026) · Romanian Journal of Neurology · doi:10.37897/rjn.2026.3.7
Tele-Neurology Meets Artificial Intelligence: Current Progress, Limitations, and Emerging Horizons
Treteanu, A. R., Herdes, H., Cobuz, G., Niță, M., Visoiu, I.-E., Hostiuc, A., Gregorini, M., Lorusso, L., Sîrbu, C. A., & Stănescu, A. M. A. (2026) · AI · doi:10.3390/ai7070272
“/large-language-model assistantsforpatientandcaregiver engagement[87–89];IoTand smart-homedigitalbiomarkersfor earlydementiadetection[61]; multimodalandfoundationAI models[93].”
Peng, L., He, H., Huang, Z., & Wang, Y. (2026) · Frontiers in Psychiatry · doi:10.3389/fpsyt.2026.1833948
Metadata
- Authors
- Dang, M., Liu, B., Chen, Y., & Zhang, Z.
- Affiliations
- Beijing Normal University (CN)
- Corresponding author
- Chen, Yaojing
- Zhang, Zhanjun
- Venue
- Psychoradiology
- Publisher
- Oxford University Press (OUP)
- Published
- 2026
- Volume
- 6
- Pages
- kkag007-kkag007
- ISSN
- 2634-4416 (ISSN-L)
- PMC
- PMC13010822
- OpenAlex
- W7135047425
- Cited by
- 10
- References
- 169
- Citations by year
- 2026: 10
- FWCI
- 10.02
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- Psychoradiology (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Funders
- Science and Technology Innovation 2030 Major Projects (2022ZD0211600)
- National Key Research and Development Program of China (2023YFC3605400)
- China Postdoctoral Science Foundation (GZC20252268)
- Beijing Nova Program
- Topics
- Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Keywords
- cognitive disorders, dementia, multimodal neuroimaging, artificial intelligence, precision medicine, multimodal fusion, explainable AI, artificial neural networks, presymptomatic biomarkers, differential diagnosis, disease progression prediction, patient stratification, federated learning, demographic bias, non-Alzheimer's dementia, neurodegenerative diseases, missing modalities
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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: Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine. https://app.cochat.ai/papers/10.1093/psyrad/kkag007
- The paper
- Dang, M., Liu, B., Chen, Y., & Zhang, Z. (2026). Multimodal neuroimaging and AI integration in cognitive disorders: advances, challenges, and future directions for precision medicine. Psychoradiology. https://doi.org/10.1093/psyrad/kkag007
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.