Anomaly Detection for Alzheimer’s Disease in Brain MRIs via Unsupervised Generative Adversarial Learning
Jean Nathan Cabreza, Geoffrey A. Solano, Sun Arthur A. Ojeda, Vincent Munar
Abstract
Alzheimer’s disease (AD) is a neurodegenerative disease that results in cognitive decline, and even dementia, in patients. To diagnose AD, a combination of tools is typically used, with structural magnetic resonance imaging (sMRI) being one of them. sMRI images have mostly been used in supervised deep learning approaches, which requires large amounts of labeled data. To alleviate the need for labels, unsupervised deep learning could be used as an alternative. This study proposes an unsupervised model based on the deep convolutional generative adversarial network that performs anomaly detection on brain MRIs to diagnose AD. The model is able to yield an AUROC of 0.7951, a precision of 0.8228, a recall of 0.7386, and an accuracy of 74.44%.