Transfer Learning for Alzheimer's Disease Diagnosis from MRI Slices: A Comparative Study of Deep Learning Models

Abstract

The early detection of Alzheimer's disease is a significant research priority in the field, as it enables timely medical intervention and treatment that may potentially slow down the progression of the disease and improve the patient's quality of life. Research has shown that early indicators of this condition can be identified up to 20 years before symptoms onset through the use of Magnetic Resonance Imaging (MRI) scans. This study evaluates the potential of transfer learning and deep learning algorithms for the accurate diagnosis of Alzheimer's Disease (AD) using MRI scans. The evaluation focuses on two pre-trained models, ResNet-152 and AlexNet, using original vendor data from the ADNI dataset, as well as data subjected to skull stripping. The results showed that the ResNet-152 model achieved the highest accuracy of 99.96% on sagittal plane slices extracted from the original dataset. Furthermore, the study highlights that the models trained on original MRI images outperformed those trained on images that underwent skull stripping.