The Diagnosis of Alzheimer's Disease Based on Enhanced Residual Neutral Network

Alzheimer's disease (AD) is a neurodegenerative sickness. Recently, the end-to-end process of using neural networks to classify patterns has gradually replaced the tedious process of manually extracting features. The residual neural network (ResNet) can effectively avoid gradient explosion, gradient disappearance and network degradation. For AD, this paper attempts to apply ResNet to the diagnosis. Although the ResNet works better compared with the machine learning methods. However, the performance of Mild Cognitive Impairment (MCI) and Normal Control (NC) recognition is poor due to the insignificant difference in magnetic resonance imaging (MRI) brain image features. In this paper, the selective kernel network (SKNet) and channel shuffle are introduced in the Resnet network, and an enhanced Resnet (EResNet) is proposed to accurately diagnose the AD symptoms. The method solves the influence of the multiple scales of input information on the network through SKNet, and evenly disturbs the channel feature through the channel shuffle, so that the network can fully utilize the feature information of the input image to improve the network identification capability. To evaluate the performance of the EResNet, MRI images of the brain were used to diagnosis of AD. The classification accuracy in AD / MCI and MCI / NC control groups reached 96.47% and 90.70%. The experimental results show that the proposed model presents a good performance on AD classification.