Transfer Learning-Assisted Prognosis of Alzheimer's Disease and Mild Cognitive Impairment Using Structural-MRI

Abstract

Alzheimer's is a neurodegenerative disease that damages human brain cells and causes dementia. When the brain cells gradually deteriorate, it leads to the inability to carry out everyday activities. While conventional machine learning (ML) has been shown to be efficient in assisting with AD diagnosis, relatively few research have examined the effectiveness of deep learning and transfer learning in this difficult challenge. We assessed the possibility of early recognition and prediction of Alzheimer's disease (AD) using pre-trained transfer-learning algorithms on structural brain MRI. Advances in artificial intelligence are assisting in the enhancement of early detection of Alzheimer's disease. Using open-source neuroimaging data, researchers have been able to construct programs that help in Alzheimer's diagnosis and prognosis. The presented study is based on an effective technique of applying transfer learning to classify the structural MRI (s-MRI) Axial brain scans by fine-tuning a pre-trained convolutional neural network (CNN), ResNet50, and VGG-16. We have taken s-MRI Axial data from an online available data repository Alzheimer's Disease Neuroimaging Initiative (ADNI). We implemented pre-trained model namely CNN, VGG-16 and ResNet50 trained on brain s-MRI axial scans to classify them into three classes: Cognitive Normal (CN), Mild cognitive impairment (MCI), and Alzheimer's disease (AD). Experiments show that ResNet50 outperformed CNN and VGG-60 with an accuracy of 95.30% on brain MRI axial scan for accurate and early prediction of AD and the onset of MCI.