An Intelligent Diagnostic Approach for Early Alzheimer's Diagnosis Based on Brain Area Screening and Self-Attention

Aiming at the current Alzheimer's early diagnosis method has the problems of information redundancy and low precision. A brain region screening and self-attention based Alzheimer early intelligent diagnosis method is proposed. Firstly, a residual block-based fully convolutional neural network (Res-FCNN) generative model is built for MRI image pre-training. Subsequently, the first layer feature maps of the Res-FCNN model are read for brain region ranking to screen out some brain regions with lower weights. Then, a neural network (SPANN) classification model based on spatial position encoding and self-attention mechanism was constructed, which was built based on the results of brain region screening, and at the same time, coding and decoding the brain regions based on the self-attention mechanism was performed. The experimental results show that the proposed brain region screening and self-attention based Alzheimer's early intelligent diagnosis method effectively eliminates the redundant information of MRI images and achieves a classification accuracy of 98.7%. The proposed method provides a new idea for the early diagnosis of Alzheimer.