SimCMC: A Simple Compact Multiview Contrastive Framework for Self-Supervised Early Alzheimer’s Disease Diagnosis

Automated diagnosis of early Alzheimer’s disease (AD) is a challenging task that requires a large amount of manually labeled data. Securing large-scale, precise expert annotations for medical images poses a significant difficulty. Contrastive learning (CL), as a self-supervised representation learning method, is an effective solution for early AD diagnosis without few labeled data. Nevertheless, most current CL-based methods usually learn representations from a single view and suffer from a lack of correlations across different perspectives. These methods are also deficient in minimizing the redundancy between the correlation, which limits the ability to produce compact representations. To address the above issues, we propose a simple compact multiview contrastive (SimCMC) framework for early AD diagnosis. The intrinsic feature representations are extracted by the compact CL framework from multiview brain MRI slices. Then, a dual-attention-based multiview fusion (DAMF) block is designed to effectively integrate compact multiview features and inner correlations, further enhancing the capability of learning hybrid representations. Additionally, we establish a novel composite loss function to constrain the process of learning view-invariant and more compact representation. The experimental results on three publicly available datasets (ADNI1, ADNI2, and OASIS) demonstrate that SimCMC achieved satisfactory performance for early AD diagnosis compared with current state-of-the-art approaches.