MRI-based Alzheimer's disease classification using deep learning: A novel small-data approach.

Abstract

Alzheimer's disease (AD) is a progressive neurodegenerative disease that causes cognitive impairment among elder people. It is the most common cause of dementia. Early and timely diagnosis of AD can help with the overall treatment process of the patients. Classifying AD patients from mild cognitive impairment (MCI) and cognitive normal (CN) patients is very important for AD diagnosis. Recently use of machine learning techniques and deep learning-based algorithms have become a popular choice for the AD classification task. However, most of the existing approaches depend upon large data sets or suffer from data leakage. In this work, we present a novel framework for AD classification in a small data regime. Our approach is based on three main steps: 1) PGGAN based medical image generation to deal with data scarcity

2) Model pretraining using SimCLR framework

3) Train the pretrained model for the final classification task. We used ResNet-18 in combination with CBAM (Convolutional Block Attention Module). The CBAM module enhances the useful features in the images, which helps in better training of the model. We performed a clinical evaluation of our model on novel test data set. We achieved an accuracy of 83% for the AD vs. CN classification task using only a few slices for the training process. We also compared our model with the previous approaches and found our results to be comparable even with such small data. Results indicate the effectiveness of our proposed framework and can help in the early diagnosis of AD patients.