Early Detection Models for Persons with Probable Alzheimer's Disease with Deep Learning

Abstract

Unfortunately, Alzheimer's disease (AD) cannot be cured or slowed with today's medication. Scientific studies have revealed that 1) a cognition drop is a precursor of AD, 2) the progression of AD is highly correlated to cognition decline, and 3) AD early detection and intervention becomes increasingly clear to be the best choice of improving quality of life for persons with probable AD as of today. This project aims to improve the predictive model we developed earlier by focusing on AD early detection. We present how recurrent neuron network (RNN) models can be adopted to AD early detection modeling (AD-EDM). Compared to models built from traditional approaches such as neuron networks, Bayesian networks, and tree-based algorithms, we demonstrate the prediction accuracy of RNN AD-EDM increases substantially.