Diagnosis of Alzheimer’s disease via an attention-based multi-scale convolutional neural network

Alzheimer’s disease (AD) is one of the most common neurodegenerative diseases. Accurate diagnosis of mild cognitive impairment (MCI) in the prodromal stage of AD can delay onset. Therefore, the early diagnosis of AD is particularly essential. The convolutional neural network (CNN) extracts feature of image layer-by-layer, and the observed features are obtained by setting different receptive fields. However, the brain structure is very complicated, and the etiology of AD is unknown, in addition, most of the existing methods do not consider the details and overall structure of the image. To address this issue, we propose a novel multi-scale convolutional neural network (MSCNet) to enhance the model’s feature representation ability. A channel attention mechanism is introduced to improve the interdependence between channels and adaptively recalibrate the channel direction’s characteristic response. To verify the effectiveness of our method, we segment the original MRI data to obtain white matter (WM) and gray matter (GM) datasets and train the model. Extensive experiments show that our method obtains the state-of-the-art performance with fewer parameters and lower computational complexity.