Early Diagnostic Classification of Alzheimer's Disease Using Structural and Resting-State Functional MRI.
- 1Jilin Provincial Key Laboratory for Numerical Simulation, Jilin Normal University, Siping 136000, China.
Abstract
Background: Alzheimer's disease (AD) is an irreversible neurodegenerative disorder where early diagnosis at the mild cognitive impairment (MCI) stage is critical for timely intervention. While structural MRI (sMRI) and resting-state functional MRI (rs-fMRI) provide complementary pathological information, clinical translation is limited by high feature heterogeneity and the heavy computational burden of 3D deep learning networks. This study aims to propose a lightweight, hybrid multimodal fusion model to overcome these limitations and improve diagnostic efficiency. Methods: A lightweight MobileViT-XS network was utilized to extract global and local pathological features from gray-matter sMRI images to reduce computational costs. Concurrently, graph theory was applied to rs-fMRI data to construct brain functional networks and calculate global topological properties. Following pretraining, feature-level fusion was performed to integrate sMRI features, rs-fMRI properties, demographic information, and two clinical neuropsychological examinations. Results: The validation results show that the proposed hybrid model achieved high diagnostic performance in the three-class classification task, yielding an accuracy of 0.928 ± 0.010, sensitivity of 0.901 ± 0.014, precision of 0.933 ± 0.014, and an F1 score of 0.911 ± 0.012. On the held-out testing subset of the independently collected dataset after dataset-specific fine-tuning, the proposed model achieved an accuracy of 0.857, with a sensitivity of 0.867, a precision of 0.867, and an F1 score of 0.859. Conclusions: The proposed framework successfully minimizes computational overhead while maintaining high classification accuracy. These findings suggest that the hybrid multimodal fusion model provides a promising computer-aided approach for three-class classification of AD, MCI, and NC.
Related
- Plasma Proteomic Classification of Alzheimer Disease, Parkinson Disease, and Amyotrophic Lateral Sclerosis Using a Single Neurodegeneration Panel: A Diagnostic Cohort Study
- Multimodal patch-based deep learning for early Alzheimer’s disease classification using MRI and PET
- Efficient Alzheimer’s disease classification with robust rank pooling
- An uncertainty-aware multimodal deep learning framework for Alzheimer’s disease classification using MRI and electronic health records
- EEG Biomarkers for ATN Classification in Early Alzheimer's disease
BibTeX
@article{liu2026early,
title = {{Early Diagnostic Classification of Alzheimer's Disease Using Structural and Resting-State Functional MRI.}},
author = {Junhu Liu and Bing Zhu and Miao Yu and Shichao Du},
journal = {Neurology international},
year = {2026},
month = oct,
doi = {10.3390/neurolint18100191},
url = {https://europepmc.org/article/MED/42931145},
note = {CC BY},
}
RIS
TY - JOUR
TI - Early Diagnostic Classification of Alzheimer's Disease Using Structural and Resting-State Functional MRI.
AU - Junhu Liu
AU - Bing Zhu
AU - Miao Yu
AU - Shichao Du
PY - 2026
DA - 2026/10/13
JO - Neurology international
DO - 10.3390/neurolint18100191
UR - https://europepmc.org/article/MED/42931145
AB - Background: Alzheimer's disease (AD) is an irreversible neurodegenerative disorder where early diagnosis at the mild cognitive impairment (MCI) stage is critical for timely intervention. While structural MRI (sMRI) and resting-state functional MRI (rs-fMRI) provide complementary pathological information, clinical translation is limited by high feature heterogeneity and the heavy computational burden of 3D deep learning networks. This study aims to propose a lightweight, hybrid multimodal fusion model to overcome these limitations and improve diagnostic efficiency. Methods: A lightweight MobileViT-XS network was utilized to extract global and local pathological features from gray-matter sMRI images to reduce computational costs. Concurrently, graph theory was applied to rs-fMRI data to construct brain functional networks and calculate global topological properties. Following pretraining, feature-level fusion was performed to integrate sMRI features, rs-fMRI properties, demographic information, and two clinical neuropsychological examinations. Results: The validation results show that the proposed hybrid model achieved high diagnostic performance in the three-class classification task, yielding an accuracy of 0.928 ± 0.010, sensitivity of 0.901 ± 0.014, precision of 0.933 ± 0.014, and an F1 score of 0.911 ± 0.012. On the held-out testing subset of the independently collected dataset after dataset-specific fine-tuning, the proposed model achieved an accuracy of 0.857, with a sensitivity of 0.867, a precision of 0.867, and an F1 score of 0.859. Conclusions: The proposed framework successfully minimizes computational overhead while maintaining high classification accuracy. These findings suggest that the hybrid multimodal fusion model provides a promising computer-aided approach for three-class classification of AD, MCI, and NC.
SN - 2035-8385
SN - 2035-8377
ER -