Detecting Early Risk of Alzheimer’s Disease Using Self-Supervised Multimodal Representation Learning

Alzheimer’s disease is an irreversible neurogenerative disorder whose treatment is mainly focused on slowing its progression. Most studies aimed at automating the detection of this disease generally focus on a single type of data (or modality) to train AI models, which limits its massive application. This work explores the impact of integrating multiple modalities (structural and functional MRI) for classifying features of patients with Alzheimer’s disease (AD) versus normal cognition. In addition, the trained model was tested with data from individuals with Subjective Cognitive Complaint (SCC) to estimate the probability of belonging to the AD group. This could be useful to follow up SCC patients with a high risk of developing AD according to the model. Our experimental results reveal that our proposed multimodal approach outperforms the unimodal approach by approximately 2.45% AUC score.