基于残差混合注意力机制的脑部CT图像分类卷积神经网络模型  

A Convolutional Neural Network for Brain CT Image Classification Based on Residual Hybrid Attention Mechanism

作　　者：乔思波 庞善臣[1] 王敏[2] 翟雪 于世行 丁桐 QIAO Si-bo;PANG Shan-chen;WANG Min;ZHAI Xue;YU Shi-hang;DING Tong(College of Computer Science and Technology,China University of Petroleum,Qingdao,Shandong 266580,China;College of Control Science and Engineering,China University of Petroleum,Qingdao,Shandong 266580,China;College of Mechanical Engineering,Tiangong University,Tianjin 300387,China;College of Software,Shandong University,Jinan,Shandong 250101,China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580 [2]中国石油大学(华东)控制科学与工程学院,山东青岛266580 [3]天津工业大学机械工程学院,天津300387 [4]山东大学软件学院,山东济南250101

出　　处：《电子学报》2021年第5期984-991,共8页Acta Electronica Sinica

基　　金：山东省重大科技创新工程(No.2019TSLH0214);泰山产业领军人才(No.tscy20180416);国家自然科学基金(No.61873281,No.61972416)。

摘　　要：针对阿尔兹海默症、病变(如脑肿瘤)和健康老化的3类脑部CT图像分类问题,本文提出了一种改进的ResNet-10卷积神经网络模型.该模型在网络的残差映射结构中加入残差混合注意力模块,解决了原模型提取的特征分辨性弱的问题,精确捕捉了脑部组织在CT图像中的位置和内容信息;此外,本文设计了全局平均池化层,简化了模型的复杂度,并在其后引入Dropout机制,缓解了过拟合.在训练阶段,该模型建立了标签平滑交叉熵损失函数,使模型在样本数量有限的情况下仍有较强的泛化能力.系列实验证明了改进后的ResNet-10网络模型在分类脑部CT图像时达到97.47%的分类精度.To classify three types of brain CT(computerized tomography)images in Alzheimer's disease,lesion(e.g.,brain tumor)and healthy aging,an improved ResNet-10 convolutional neural network is proposed in this papers.A residual hybrid attention module is embedded in the residual identity mapping to capture the location and content information of brain tissue in brain CT images,solving the original model to extract weak distinguish features problems.In addition,to simplify the improved model and alleviate the overfitting,several techniques such as global average pooling and Dropout are used in the model.Moreover,to have strong generalization ability in the case of limited sample quantity,tag smoothing cross-entropy loss function is adopted to train the model.Experimental results show that the improved ResNet-10 achieves 97.47% accuracy in classifying brain CT images.

关 键 词：残差混合注意力模块 标签平滑 脑部CT 卷积神经网络 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]