CNN and transfer learning-based deep learning architectures for Alzheimer’s disease detection from MRI scan: a comparative analysis

Abstract

Alzheimer’s disease, a neurodegenerative illness that gradually impairs cognitive function, affects a large number of people worldwide. It is crucial to enhance early detection methods so that treatment can be started at an earlier stage. This study adopts various tools and frameworks to investigate deep learning-based architectures for detecting Alzheimer’s disease. We have utilized the OASIS-1 dataset of brain MRI scans for Alzheimer’s patients. Our approach incorporates the use of SqueezeNet, SENet, MobileNetV2, and an independently developed CNN architecture to predict various stages of Alzheimer’s disease. It covers the processes of data preparation, model training, and evaluation of test data. Our proposed CNN architecture is more lightweight than other pretrained models, which can achieve 98.06% accuracy, and a 98.1% recall score, meaning missing out on a positive case of Alzheimer’s is distinctly low. Our study on detecting Alzheimer’s disease is based on a comprehensive setup that includes enhancing images to improve the model’s performance. We have explored the effectiveness of different deep learning algorithms for this purpose.