An Analysis on Structural MRI Images for Better Alzheimer Disease Detection using Deep Neural Networks

Alzheimer's disease is a severe neurological condition that damages memory and deteriorates brain tissue.Therefore, accurate classification of this disease is necessary for better and faster medical treatment.Recently, deep learning-based models have gained traction in this task due to the ability of them to learn hierarchical representations to examine complex patterns in medical imaging and clinical data related to Alzheimer's disease.However, medical images are usually noisy and contains unnecessary information, which is one of the main reasons for degrading model's performance.In this article, we have focused on introducing a series of pre-processing techniques for cleaning structural MRI images and incorporate the pre-processed data with the pre-trained models.In this research, the OASIS dataset images have been used, preprocessed with different pre-processing techniques and finally, fine tuned them with a set of pre-trained models like VGG16, VGG19, ResNet50, MobileNet, XceptionNet for better classification of different stages of Alzheimer's disease.The experiment provides strong evidence on the positive effect of using a series of pre-processing techniques on structural MRI images, that yields a lofty score in accuracy, sensitivity, precision and f1-score for MobileNet model.