Generative Aging Of Brain MRI For Early Prediction Of MCI-AD Conversion

Abstract

Automatic diagnosis of Alzheimer’s disease (AD) from MR images of the brain promises to yield important information of a patient’s disease status or even early prediction of disease onset. This work investigates deep learning based methods to predict conversion of Mild Cognitive Impairment (MCI) to AD based on widely available T1-weighted MR brain images. We present a novel approach breaking up the conversion prediction into a generative and a discriminative step. Using the recently proposed Wasserstein-GAN model, we generate a synthetically aged brain image given a baseline image. The aged image is passed to an MCI/AD discriminator deciding the future disease status. Using only one coronal slice of a patient’s baseline T1image, our approach achieves 73% accuracy, 68% precision, and 75% recall on MCI-to-AD conversion prediction at a 48months follow-up.