Applying deep learning models to structural MRI for stage prediction of Alzheimer’s disease

Abstract

Alzheimer’s disease is a brain disease that causes impaired cognitive abilities in memory, concentration, planning, and speaking. Alzheimer’s disease is defined as the most common cause of dementia and changes different parts of the brain. Neuroimaging, cerebrospinal fluid, and some protein abnormalities are commonly used as clinical diagnostic biomarkers. In this study, neuroimaging biomarkers were applied for the diagnosis of Alzheimer’s disease and dementia as a noninvasive method. Structural magnetic resonance (MR) brain images were used as input of the predictive model. T1 weighted volumetric MR images were reduced to two-dimensional space by several preprocessing methods for three different projections. Convolutional neural network (CNN) models took preprocessed brain images, and the training and testing of the CNN models were carried out with two different data sets. The CNN models achieved accuracy values around 0.8 for diagnosis of both Alzheimer’s disease and mild cognitive impairment. The experimental results revealed that the diagnosis of patients with mild cognitive impairment was more difficult than that of patients with Alzheimer’s disease. The proposed deep learning-based model might serve as an efficient and practical diagnostic tool when MRI data are integrated with other clinical tests.