基于改进ResNet的阿尔兹海默症分类网络  

Alzheimer′s disease classification network based on improved Residual Network

作　　者：王斌[1] 吴晓红[1] 辜蕊 卿粼波[1] 何小海[1] WANG Bin;WU Xiaohong;GU Rui;QING Linbo;HE Xiaohai(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China;Department of Neurology,Affiliated Hospital of Southwest Jiaotong University&The Third People's Hospital of Chengdu,Chengdu 610031,China)

机构地区：[1]四川大学电子信息学院,成都610065 [2]西南交通大学附属医院&成都市第三人民医院神经内科,成都610031

出　　处：《智能计算机与应用》2023年第3期69-76,82,共9页Intelligent Computer and Applications

基　　金：成都市重大科技应用示范项目(2019-YF09-00120-SN)。

摘　　要：针对阿尔兹海默症(AD)、轻度认知障碍(MCI)和正常人(CN)三阶段人群脑部核磁成像(MRI)难以识别分类的问题,提出了一种基于改进的ResNet的阿尔兹海默症的分类方法。该方法首先将预处理后的数据送入到根据AD的MRI特点设计的通道分离残差模块中,提取并组合低维特征与高维特征。然后,将提取的特征送入通道注意力模块,调整通道之间的权重,得到更加精确的分类特征。最后,将特征矩阵送入线性分类层输出分类结果。在随机划分的数据集上,AD/MCI/CN分类准确率达到了83.54个百分点,AD/CN准确率达到了95.27个百分点,MCI/CN准确率达到了85.07个百分点。证明本文方法对区分AD各个阶段的有效性;在以个体ID为主要划分依据的数据集上,实验结果与基础网络ResNet对比,AD/MCI/CN分类准确率提高了0.7个百分点,AD/CN准确率提高了10个百分点,MCI/CN准确率提高了4个百分点,证明本文对基础网络的改进能有效提高AD分类准确率。同时通过对比有无数据泄露的实验结果,证明正确划分数据集的必要性。Aiming at the difficulty in brain MRI classification of patients with Alzheimer's disease(AD),mild cognitive impairment(MCI)and normal subjects(CN),the paper proposes a classification method of Alzheimer's disease based on improved ResNet.Firstly,the pre-processed data are sent into the channel separation residual module designed according to the MRI characteristics of AD,and low-dimensional features and high-dimensional features are extracted and combined.Then,the extracted features are sent into the channel attention module,and the weight between channels is adjusted to obtain more accurate classification features.Finally,the eigenmatrix is sent to the linear classification layer to output the classification results.The AD/MCI/CN classification accuracy is 83.54 percentage points,AD/CN accuracy is 95.27 percentage points,and MCI/CN accuracy is 85.07 percentage points.The effectiveness of the proposed method in distinguishing the different stages of AD is proved.In the data set based mainly on individual IDS,the experimental results are compared with the basic network ResNet,thus the conclusion is that the classification accuracy of AD/MCI/CN is increased by 0.7 percentage points,the accuracy of AD/CN is increased by 10 percentage points,and the accuracy of MCI/CN is increased 4 percentage points,respectively.It is proved that the improvement of basic network in this paper can effectively improve the accuracy of AD classification.At the same time,the necessity of dividing data sets correctly is proved by comparing the experimental results of data leakage.

关 键 词：改进ResNet 通道分离残差模块 数据泄露 通道注意力机制 阿尔兹海默症 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]