Multi-modal deep learning for Alzheimer’s disease classification using MRI and clinical information

Accurate identification of different stages of Alzheimer’s disease (AD) is important for improving clinical assessment. This study aimed to develop a multimodal deep learning model integrating MRI and clinical data to improve AD classification performance. Participants with complete MRI and clinical data from the National Alzheimer’s Coordinating Center (NACC) were grouped as AD, mild cognitive impairment (MCI), and normal control (NC). After standardized preprocessing, MRI data were fed into ResNet, DenseNet, and SENet to extract imaging features. Clinical features were standardized and fused with imaging features at the feature level to build multimodal models. The results demonstrated that multimodal fusion substantially improved classification performance compared with unimodal MRI-based models. Specifically, the inclusion of clinical features increased AD vs. NC classification accuracy to approximately 0.95 and elevated the area under the curve (AUC) for AD vs. MCI and MCI vs. NC tasks to 0.87–0.90. Among the evaluated architectures, SENet exhibited sensitivity in distinguishing AD from MCI, highlighting the advantage of attention mechanisms in capturing subtle disease-related patterns. Furthermore, the results showed correlations with established clinical indicators of disease severity, as well as meaningful associations with education, age, sex, and neuropsychiatric symptoms. The proposed multimodal deep learning model integrating MRI and clinical information substantially enhances AD classification accuracy and interpretability and provides valuable complementary information for early detection of AD. Not applicable (retrospective study).