Brain-Aware Replacements for Supervised Contrastive Learning in Detection of Alzheimer's Disease

We propose a novel framework for Alzheimer's disease (AD) detection using\nbrain MRIs. The framework starts with a data augmentation method called\nBrain-Aware Replacements (BAR), which leverages a standard brain parcellation\nto replace medically-relevant 3D brain regions in an anchor MRI from a randomly\npicked MRI to create synthetic samples. Ground truth "hard" labels are also\nlinearly mixed depending on the replacement ratio in order to create "soft"\nlabels. BAR produces a great variety of realistic-looking synthetic MRIs with\nhigher local variability compared to other mix-based methods, such as CutMix.\nOn top of BAR, we propose using a soft-label-capable supervised contrastive\nloss, aiming to learn the relative similarity of representations that reflect\nhow mixed are the synthetic MRIs using our soft labels. This way, we do not\nfully exhaust the entropic capacity of our hard labels, since we only use them\nto create soft labels and synthetic MRIs through BAR. We show that a model\npre-trained using our framework can be further fine-tuned with a cross-entropy\nloss using the hard labels that were used to create the synthetic samples. We\nvalidated the performance of our framework in a binary AD detection task\nagainst both from-scratch supervised training and state-of-the-art\nself-supervised training plus fine-tuning approaches. Then we evaluated BAR's\nindividual performance compared to another mix-based method CutMix by\nintegrating it within our framework. We show that our framework yields superior\nresults in both precision and recall for the AD detection task.\n