基于结合SimAM模块的三维Densenet的阿尔茨海默症分类算法研究

Alzheimer's Disease Classification Algorithm Based on Three-dimensional Densenet Combined with SimAM

阿尔茨海默症是一种不可逆的大脑神经退行性疾病。利用深度学习技术辅助医生对提升识别阿尔茨海默症患者的效率有重要意义。论文将三维卷积技术,密集连接卷积神经网络(Densenet)和一种简单、无参数的卷积神经网络注意模块,简称SimAM的注意力模块相结合,设计了一个3D-SAMDensenet算法,用于对大脑的磁共振图像(MRI)进行分类。用此模型分别对阿尔茨海默症(AD)和正常人(NC),轻度认知障碍(MCI)和正常人(NC)进行两种二分类时,与对比的传统和深度学习的分类方法相比,都取得更好的分类结果,并且在图像差异较小的MCI/NC分类中,得到了比AD/NC分类更大的优势。

Alzheimer's disease is an irreversible neurodegenerative disease of the brain.Deep learning technology is used to assist doctors is of great significance to improve the efficiency of identifying patients with Alzheimer's disease.Combining the three-dimensional convolution technology,densely connected convolutional networks(Densenet)and a simple,parameter-free attention module for convolutional neural networks,referred to as the attention module of SimAM,this paper designs a 3D-SAMDensenet algorithm to classify the magnetic resonance images(MRI)of the brain.When using this model to classify Alzheimer's disease(AD)and normal people(NC),mild cognitive impairment(MCI)and normal people(NC),better classification results are obtained compared with the traditional and deep learning classification methods,and greater advantages are obtained than AD/NC classification in MCI/NC classification with small image differences.