Biceph-Net: A robust and lightweight framework for the diagnosis of Alzheimer's disease using 2D-MRI scans and deep similarity learning

Abstract

Alzheimer's Disease (AD) is a neurodegenerative disease that is one of the major causes of death in the elderly population. Many deep learning (DL) techniques have been proposed for the diagnosis of AD using magnetic resonance imaging (MRI) scans. The prediction of AD using 2D slices extracted from 3D MRI scans is a challenging task as the inter-slice information gets lost. To this end, we propose a novel and lightweight framework termed ‘Biceph-net’ for AD diagnosis using 2D MRI scans that models both the intra-slice and inter-slice information. ‘Biceph-net’ has been experimentally shown to perform equally or better than spatio-temporal neural networks while being computationally more efficient. Biceph-net also is also superior in performance when compared to vanilla 2D convolutional neural networks (CNN) for AD diagnosis using 2D MRI slices. Biceph-net also has an inbuilt neighbourhood based model interpretation feature which can be exploited to further understand the classification decision taken by the network. We also give theoretical guarantees regarding the generalization performance of Biceph-net. Biceph-net experimentally achieves a test accuracy of 100\% for cognitively normal (CN) vs AD task, 98.16% for mild cognitive impairment (MCI) vs AD task and 97.80% for CN vs MCI vs AD task. IEEE