Deep Learning for Alzheimer's Disease Detection using Multimodal MRI-PET Fusion

Abstract

Alzheimer's disease (AD) is an irremediable brain disorder that is progressive and causes irreparable damage to brain cells, neurotransmitters, and nerves. This in turn severely affects brain functionalities and ultimately leads to dementia. Although there is currently no cure for AD, there are treatments that can slow down the disease's development. Hence, early diagnosis of AD is the need of the hour and researchers across the world have shifted their focus on developing robust and intelligent systems that can aid in early and accurate diagnosis of AD this has been the main motivation behind this study. The main objective of this paper is to present a comparative study of 2D and 3D Convolutional Neural Network (CNN) architectures for AD classification and to choose the most robust model for AD classification. The models are trained on MRI and PET individually and with the fusion of MRI and PET. 2D feature fusion is performed using pre-trained neural networks and 3D fusion involves a series of operations such as skull-stripping, image segmentation, and co-registration. 2D CNN provided the highest accuracy of 91.29% on MRI images followed by 3D CNN with an accuracy of 91.07%. Comparing the performance on multimodal fusion, 3D MRI -PET fusion exhibited a significantly good accuracy of 86.90%. This paper briefly describes the GUI developed for easy visualization of AD classification and the possibilities of integrating the trained machine learning models with various mobile and web applications and with instruments that facilitate real-time diagnosis and classification of AD.