Alzheimer's Disease Imaging Recognition Based on Improved 3D-ShufflenetV2 Network

To solve the problems of low accuracy, data leakage, and low efficiency in current deep learning based two dimensional Alzheimer's disease (AD) imaging recognition, an improved 3D-ShufflenetV2 network has been proposed. Firstly, the network utilizes 3D convolution to extract spatial continuity information from three dimensional images, enabling the extraction of more brain lesion information. Secondly, a Squeeze-and-Excitation attention mechanism is introduced into the backbone network, recalibrating channel-level feature responses based on global information to enhance the model's focus on important features, thereby enhancing the model's classification performance. Finally, to increase the model's ability to extract global features, a global feature fusion branch has been added to the network structure. Experiments have shown that the improved network is capable of recognizing AD, Mild Cognitive Impairment (MCI), and Cognitively Normal (CN). The classification performance of the modified network model shows significant improvement over the original network, demonstrating its practical value.