Research on early Intelligent Diagnosis Method of Alzheimer ’s Disease based on Residual Coordination CMT-Multiscale CNN

Alzheimer’s disease (AD) is an irreversible degenerative disease of the nervous system, and early diagnosis of AD is crucial. A single modality cannot obtain comprehensive pathological information, Sand multimodal data fusion provides richer information expression through the complementarity between different modalities. Aiming at the problems of low accuracy and insufficient modal fusion of traditional multi-modal classification methods, an AD classification method based on residual coordination CMT-Multiscale CNN is proposed. This method uses structural magnetic resonance imaging (sMRI) and electronic health record (EHR) to realize AD diagnosis. In the network design, a dual-branch feature extraction network of image data and non-image data is built. Branch 1 sends the image data sMRI to the improved network for feature extraction. The improved network takes CMT (Convolutional Neural Networks Meet Vision Transformers) as the main body, and introduces residual structure and coordinated attention mechanism in CMT block. The network model was focused on the key location area of AD lesions in sMRI; branch 2 sends EHR into multi-layer perceptron to extract feature information. Finally, the extracted sMRI features and EHR features are classified by Multi-scale dilated convolutional neural network(MDC). The experimental results show that the proposed model has excellent performance in the binary classification task and multi-classification task of ADNI open source data set, and can be used as an effective and convenient AD diagnosis method.