Diagnosis Method of Alzheimer's Disease in PET Image Based on CNN Multi-mode Network

摘要

Developing a correct diagnosis of Alzheimer's disease (AD) is a challenging task. Positron emission tomography (PET) is a good method to help doctors assist in the diagnosis of AD. In recent years, artificial intelligence methods such as machine learning have been widely used in image analysis and judgment and medical auxiliary diagnosis. The current methods are mainly to manually extract image features from medical images and then train classifiers to judge AD, or use deep learning, neural networks for end-to-end AD classification, most methods only use a single-mode method, and the classification effect is limited. This paper proposes a multi-mode network structure based on CNN to classify and diagnose AD. The network is mainly divided into three parts: CNN-based multi-scale deep-level feature extraction module, image texture feature extraction module, and SVM-based feature integration classification module. The network fully combines the advantages of the two modes of manual feature extraction and neural network. Compared with single mode feature extraction, this method has higher accuracy and has a good performance on the classification and diagnosis of AD.