Patch-Based Classification for Alzheimer Disease using sMRI

Abstract

Alzheimer's disease, the most severe form of dementia, is a neuronal destructive brain ailment that worsens over time with no cure thereby realizing its importance in its early detection. Nowadays, convolutional neural network, especially 3-Dimensional networks are becoming popular for detecting medical illness due to its inherent nature of capturing spatial dimensions as well. In our study, we have worked on 3D patch-based feature extraction technique where these patches are generated using torch library and passed into 19 layered ConvNet for classification. The MRI images (Magnetic Resonance Imaging) are obtained from MIRIAD database (Minimal Interval Resonance Imaging in Alzheimer's disease) are pre-processed for bias correction, skull stripping and registration and further augmented by rotation algorithm to increase dataset size and finally classified into Normal Control (NC) and Alzheimer Disease (AD) with 99.79 percent accuracy. This classification will provide great assistance to all especially in lack of clinicians' availability during the time of pandemic and remote areas where experts are not in reach.