Deep Relative Learning for Prediction of Preclinical Alzheimer's Disease

Alzheimer’s disease (AD) is the most common form of dementia that primarily affects elderly individuals, gradually impairing their memory and cognitive abilities. As the disease progresses, Alzheimer’s patients often experience a decline in their ability to perform mundane tasks. Numerous studies have shown intervening during the preclinical stage of AD holds significant promise for mitigating its advancement. However, identifying preclinical AD cases presents a major challenge as such patients exhibit minimal to no detectable neurodegeneration, and therefore generally classified as ‘cognitively normal’. Given the difficulty of detecting these subtle alterations through unaided human eye, we propose a novel approach called Relative Scoring Network (RSNet), which integrates relative learning and contrastive learning to detect preclinical AD patients based solely on their MRI scans. RSNet also produces a score that can be used to aid clinicians in gauging the severity of preclinical AD. We demonstrate the feasibility of the idea and its potential, especially the differentiation between healthy individuals and those in the preclinical stage of AD, utilizing the OASIS-3 dataset.