Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.

Deep-learning based multi-modal models for brain age, cognition and amyloid pathology prediction

Open access
Wang, Chenxi, Zhang, Weiwei, Ni, Ming, Wang, Qiong, Liu, Chang, Dai, Linbin, Zhang, Mengguo, Shen, Yong, Gao, Feng · 2025 · Alzheimer's Research & Therapy

Abstract

BACKGROUND: Magnetic resonance imaging (MRI), combined with artificial intelligence techniques, has improved our understanding of brain structural change and enabled the estimation of brain age. Neurodegenerative disorders, such as Alzheimer's disease (AD), have been linked to accelerated brain aging. In this study, we aimed to develop a deep-learning framework that processes and integrates MRI images to more accurately predict brain age, cognitive function, and amyloid pathology. METHODS: In this study, we aimed to develop a deep-learning framework that processes and integrates MRI images to more accurately predict brain age, cognitive function, and amyloid pathology.We collected over 10,000 T1-weighted MRI scans from more than 7,000 individuals across six cohorts. We designed a multi-modal deep-learning framework that employs 3D convolutional neural networks to analyze MRI and additional neural networks to evaluate demographic data. Our initial model focused on predicting brain age, serving as a foundational model from which we developed separate models for cognition function and amyloid plaque prediction through transfer learning. RESULTS: The brain age prediction model achieved the mean absolute error (MAE) for cognitive normal population in the ADNI (test) datasets of 3.302 years. The gap between predicted brain age and chronological age significantly increases while cognition declines. The cognition prediction model exhibited a root mean square error (RMSE) of 0.334 for the Clinical Dementia Rating (CDR) regression task, achieving an area under the curve (AUC) of approximately 0.95 in identifying ing dementia patients. Dementia related brain regions, such as the medial temporal lobe, were identified by our model. Finally, amyloid plaque prediction model was trained to predict amyloid plaque, and achieved an AUC about 0.8 for dementia patients. CONCLUSIONS: These findings indicate that the present predictive models can identify subtle changes in brain structure, enabling precise estimates of brain age, cognitive status, and amyloid pathology. Such models could facilitate the use of MRI as a non-invasive diagnostic tool for neurodegenerative diseases, including AD.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1186/s13195-025-01773-z

Cites

30 of 30 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Multi-modality model architecture. A The original MRI data is skull-stripped and registered to the MNI standard space. To enhance model training and prevent over-fitting, we employ data augmentation techniques to generate additional training data. For model evaluation, we utilize the original preprocessed data. B The figure illustrates the primary architecture of our deep learning model. By constructing corresponding input modules, the data is transformed into vectors of size 256 × 1 or matrices of 256 × n. The transformer model is capable of concatenating all vectors and matrices as input. The output module processes the output vector from the transformer module to produce the final results. C This represents the main architecture of the MRI input modules. Each box corresponds to a convolution layer:"Cov"indicates a convolution layer, while"Res Conv"denotes a convolution layer with shortcut connections. The notation"k"refers to kernel size,"s"to stride,"p"to padding, and"c"to channels. D We utilize transfer learning to develop distinct models for different prediction tasks. On the left, we input MRI and age data to predict brain age, with all parameters randomly initialized. In the middle, we use MRI, sex, and chronological age as inputs to predict cognitive function. On the right, we input MRI, sex, age, and APOE genotype to predict the presence of amyloid deposition. Parameters in modules with yellow background were randomly generated, while parameters in modules with green background were transferred from the last model
Fig. 1 Multi-modality model architecture. A The original MRI data is skull-stripped and registered to the MNI standard space. To enhance model training and prevent over-fitting, we employ data augmentation techniques to generate additional training data. For model evaluation, we utilize the original preprocessed data. B The figure illustrates the primary architecture of our deep learning model. By constructing corresponding input modules, the data is transformed into vectors of size 256 × 1 or matrices of 256 × n. The transformer model is capable of concatenating all vectors and matrices as input. The output module processes the output vector from the transformer module to produce the final results. C This represents the main architecture of the MRI input modules. Each box corresponds to a convolution layer:"Cov"indicates a convolution layer, while"Res Conv"denotes a convolution layer with shortcut connections. The notation"k"refers to kernel size,"s"to stride,"p"to padding, and"c"to channels. D We utilize transfer learning to develop distinct models for different prediction tasks. On the left, we input MRI and age data to predict brain age, with all parameters randomly initialized. In the middle, we use MRI, sex, and chronological age as inputs to predict cognitive function. On the right, we input MRI, sex, age, and APOE genotype to predict the presence of amyloid deposition. Parameters in modules with yellow background were randomly generated, while parameters in modules with green background were transferred from the last modelCC BY-NC-ND · PubMed Central
Fig. 2 Brain age prediction results in the ADNI and CANDI test sets. A-D Scatter plots illustrated brain age for the CN (a), MCI (b), and dementia (c) groups, as well as for the entire sample (d) in the ADNI (test) datasets. Similarly. E–H, Scatter plots showing brain age for the CN (e), MCI (f), and dementia (g) groups, along with the overall sample (h) in the CANDI datasets. I Violin plots show brain age gap distribution in ADNI (test) datasets. J-l Scatter plots in CN, MCI and dementia subjects in ADNI (test) datasets. M Violin plots show brain age gap distribution in ADNI (test) datasets after correction. N-P Scatter plots in CN, MCI and dementia subjects in CANDI datasets after correction. Q-T Figures show corresponding images after corrected in ADNI (test) datasets. U-X Figures show corresponding images after corrected in CANDI datasets.”r” means Pearson correlation coefficient,”NS” means no significant
Fig. 2 Brain age prediction results in the ADNI and CANDI test sets. A-D Scatter plots illustrated brain age for the CN (a), MCI (b), and dementia (c) groups, as well as for the entire sample (d) in the ADNI (test) datasets. Similarly. E–H, Scatter plots showing brain age for the CN (e), MCI (f), and dementia (g) groups, along with the overall sample (h) in the CANDI datasets. I Violin plots show brain age gap distribution in ADNI (test) datasets. J-l Scatter plots in CN, MCI and dementia subjects in ADNI (test) datasets. M Violin plots show brain age gap distribution in ADNI (test) datasets after correction. N-P Scatter plots in CN, MCI and dementia subjects in CANDI datasets after correction. Q-T Figures show corresponding images after corrected in ADNI (test) datasets. U-X Figures show corresponding images after corrected in CANDI datasets.”r” means Pearson correlation coefficient,”NS” means no significantCC BY-NC-ND · PubMed Central
Fig. 3 Cognitive function prediction was evaluated in the ADNI (test) and the CANDI datasets. A-D The CDR prediction in the ADNI (test) datasets, which includes a violin plot of the distribution (a) and the binary classification results, including CN vs. MCI (b), MCI vs. dementia (c) and CN vs. dementia (d). E–H Feature scatter plots illustrating the CDR prediction results in the CANDI set, including a distribution violin plot (e) and binary classification including CN vs. MCI (f), MCI vs. dementia (g) and CN vs. dementia (h). I A fused feature map created by aligning the feature maps of all dementia samples from the training set to the MNI standard brain using the Grad-CAM algorithm. This map emphasizes the brain regions critical for cognitive function prediction, with purple-red areas indicating these significant regions
Fig. 3 Cognitive function prediction was evaluated in the ADNI (test) and the CANDI datasets. A-D The CDR prediction in the ADNI (test) datasets, which includes a violin plot of the distribution (a) and the binary classification results, including CN vs. MCI (b), MCI vs. dementia (c) and CN vs. dementia (d). E–H Feature scatter plots illustrating the CDR prediction results in the CANDI set, including a distribution violin plot (e) and binary classification including CN vs. MCI (f), MCI vs. dementia (g) and CN vs. dementia (h). I A fused feature map created by aligning the feature maps of all dementia samples from the training set to the MNI standard brain using the Grad-CAM algorithm. This map emphasizes the brain regions critical for cognitive function prediction, with purple-red areas indicating these significant regionsCC BY-NC-ND · PubMed Central
Fig. 4 Correlation between brain age gap and AD related biomarkers. A-H Scatter plots show correlation between blood biomarkers and brain age gap. I-N Scatter plots show correlation between CSF biomarkers and brain age gap. O-P, Scatter plots show correlation between imaging biomarker and brain age gap. Multiple linear regression was used to evaluate the correlation between brain age gap and biomarker. Sex, age and APOE genotype were adjusted. We use the t-test to examined whether each parameter (β) is correlated with brain age gap after FDR correlation. FDR method was used for multiple comparison correction
Fig. 4 Correlation between brain age gap and AD related biomarkers. A-H Scatter plots show correlation between blood biomarkers and brain age gap. I-N Scatter plots show correlation between CSF biomarkers and brain age gap. O-P, Scatter plots show correlation between imaging biomarker and brain age gap. Multiple linear regression was used to evaluate the correlation between brain age gap and biomarker. Sex, age and APOE genotype were adjusted. We use the t-test to examined whether each parameter (β) is correlated with brain age gap after FDR correlation. FDR method was used for multiple comparison correctionCC BY-NC-ND · PubMed Central
Fig. 5 The Amyloid pathology prediction in the ADNI (test) and CANDI datasets. A Scatter plot shown predicted SUVR value in ADNI (test) set. B ROC curve shown the performance of amyloid plaque prediction in ADNI (test) set. C Scatter plot show predicted SUVR value in CANDI set. D ROC curve shown the performance of amyloid plaque prediction in CANDI set
Fig. 5 The Amyloid pathology prediction in the ADNI (test) and CANDI datasets. A Scatter plot shown predicted SUVR value in ADNI (test) set. B ROC curve shown the performance of amyloid plaque prediction in ADNI (test) set. C Scatter plot show predicted SUVR value in CANDI set. D ROC curve shown the performance of amyloid plaque prediction in CANDI setCC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 30 references

Research works printed but not linked: 2

  1. Lee J, Burkett BJ, Min H-K, Senjem ML, Lundt ES, Botha H, et al. Deep learning-based brain age prediction in normal aging and dementia. Nature Aging. 2022;2:412–24. · DOI · Paper page
  2. Bashyam VM, Erus G, Doshi J, Habes M, Nasrallah IM, Truelove-Hill M, et al. MRI signatures of brain age and disease over the lifespan based on a deep brain network and 14 468 individuals worldwide. Brain. 2020;143:2312–24. · DOI · Paper page

Cited by

See all 11

Metadata

Authors
Wang, C., Zhang, W., Ni, M., Wang, Q., Liu, C., Dai, L., Zhang, M., Shen, Y., & Gao, F.
Affiliations
University of Science and Technology of China (CN)
Corresponding author
  • Shen, Yong
  • Gao, Feng
Venue
Alzheimer's Research & Therapy
Published
2025-05-31
Volume
17
Issue
1
Pages
126-126
ISSN
1758-9193 (ISSN-L)
OpenAlex
W4410921897
Cited by
11
References
30
Citations by year
2026: 10 · 2025: 1
FWCI
7.40
Type
journal article
Language
en
Open access
Licence
Locations
  • Alzheimer s Research & Therapy (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Chinese Academy of Sciences (XDB 39000000, 32121002)
  • University of Science and Technology of China
Topics
  • Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Functional Brain Connectivity Studies
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced MRI Techniques and Applications
Keywords
brain age prediction, cognitive function prediction, Alzheimer's disease, accelerated brain aging, T1-weighted imaging, multimodal deep learning, 3D convolutional neural networks, transfer learning, dementia detection, Clinical Dementia Rating, medial temporal lobe, neurodegenerative disease diagnosis
MeSH terms
Deep Learning, Aged, Aged, 80 and over, Aging, Alzheimer Disease, Brain, Cognition, Female, Humans, Magnetic Resonance Imaging, Male, Middle Aged, Neural Networks, Computer, Plaque, Amyloid

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: Deep-learning based multi-modal models for brain age, cognition and amyloid pathology prediction. https://app.cochat.ai/papers/10.1186/s13195-025-01773-z
The paper
Wang, C., Zhang, W., Ni, M., Wang, Q., Liu, C., Dai, L., Zhang, M., Shen, Y., & Gao, F. (2025). Deep-learning based multi-modal models for brain age, cognition and amyloid pathology prediction. Alzheimer's Research & Therapy. https://doi.org/10.1186/s13195-025-01773-z

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.