Early prediction of Alzheimer's disease based on attention mechanism of WCNN and multi-modal feature fusion

Alzheimer’s disease (AD) is a neurodegenerative disorder that is one of the common causes of dementia in the elderly. Because its early symptoms are nonspecific and easily confused with normal age-related memory decline, it is particularly important for early prediction and classification of AD. Herein, a multi-modal feature fusion AD diagnosis model based on the width convolution neural network (WCNN) and attention mechanism is presented. Specifically, we first use WCNN to extract features from magnetic resonance imaging (MRI) and enhance the attention on important parts of the image through attention mechanism. For clinical text information, we use the decision tree machine learning method and apply Clinical Dementia Rating Sum of Boxes(CDR), Alzheimer's Disease Assessment Scale-11(ADAS11) and Mini-Mental State Examination(MMSE) separately to diagnose patients. Then, we fuse the extracted image features with three types of clinical text information through multi-modal fusion to obtain the final classification results. The overall experimental results obtained show that the proposed model has well performance in different multi-mode scenarios and different classification problems, and improves the accuracy of AD prediction and classification.