Improving the Accuracy of Deep Learning Models for Alzheimer’s Disease Classification Trained on Very Small Training Sets Using Topological Data Analysis

Alzheimer's disease (AD) is a pernicious brain degeneration which has a very steady growth rate. There are many deep learning models that have done AD classification using neuroimaging data, but their performance is related to large datasets, which are very costly and rare to obtain. This paper proposes a novel approach to raise the accuracy of the deep learning models for classification when trained using small training sets, leveraging the power of Topological Data Analysis (TDA). However, applications of TDA to image classification is very unpopular. This study applies TDA to Alzheimer's disease classification and shows that a deep learning model with TDA notably refines the model's accuracy if the training set at hand is very small.