On self-supervised multi-modal representation learning: An application to Alzheimer's disease

TL;DR

This paper explores contrastive self-supervised learning for multimodal brain imaging data to improve Alzheimer's disease classification and interpretability, addressing biases in supervised models.

Contribution

It introduces a comprehensive framework for multimodal contrastive self-supervised fusion of fMRI and MRI data for Alzheimer's disease analysis.

Findings

Multimodal fusion improves classification accuracy.

Fused features enhance interpretability in brain space.

Proposed method is numerically stable for brain projection.

Abstract

Introspection of deep supervised predictive models trained on functional and structural brain imaging may uncover novel markers of Alzheimer's disease (AD). However, supervised training is prone to learning from spurious features (shortcut learning) impairing its value in the discovery process. Deep unsupervised and, recently, contrastive self-supervised approaches, not biased to classification, are better candidates for the task. Their multimodal options specifically offer additional regularization via modality interactions. In this paper, we introduce a way to exhaustively consider multimodal architectures for contrastive self-supervised fusion of fMRI and MRI of AD patients and controls. We show that this multimodal fusion results in representations that improve the results of the downstream classification for both modalities. We investigate the fused self-supervised features projected into the brain space and introduce a numerically stable way to do so.