A Res_shift Module Based Architecture for The Diagnosis of Alzheimer's Disease

Objective: Alzheimer's disease (AD) is an insidious neurodegenerative disease. Early diagnosis is crucial to preventing the development of AD. As a result, doctors can start preventive treatments as early as possible, and patients can slow or prolong the onset of the disease with medications designed to improve cognitive function. The purpose of this work is to develop a convolutional neural network architecture based on Positron Emission Tomography (PET) images for early diagnosis of Alzheimer's disease. Materials and Methods: The data sets used in this study are all from the AD public data set Alzheimer's Disease Neuroimaging Initiative (ADNI). In this work, Inception module is used to increase the network width to extract richer image information. Res_shift module is used to solve the problem of missing inter-image information in 2D images. The Res_shift module not only enables the model to have the effect of the training speed of 2D model, but also the effect of the performance of 3D model. In addition, in order to solve the problem of class imbalance in the training process, the Focal loss function is used. Results: Comprehensive experimental results show that the proposed model outperforms several competing approaches in AD-related diagnostic tasks. The accuracy of the model for AD diagnosis can reach 91% in the CN vs. AD diagnosis tasks.