摘要
针对临床医学工作量大、诊断效率较低的问题,基于深度学习理论对计算机辅助诊断分析方法进行了研究。基于传统神经网络的结构,引出更深层次的卷积神经网络(CNN),对该网络中的卷积、池化操作进行深入的讨论,引入方差代价函数实现网络误差的反向传播训练。在仿真实验的设计上,基于临床的脑部CT数据库,考虑到医学影像数据的特殊性,对CNN网络的结构进行了定制,设计包含一个输入层,7个卷积层,5个池化层与3个全连接层的网络结构。与Alex-Net网络的对比测试实验结果表明,提出的网络结构对脑部疾病CT的分类准确率可达69.28%,AUC为0.53,在性能上有一定的提升;在具体脑部疾病的识别上,对脑肿瘤的识别精度可达到86.5%。
In order to solve the problems of heavy workload and low diagnosis efficiency in clinical medicine,this paper studies the computer⁃aided diagnosis and analysis method based on the deep learning theory.Based on the structure of traditional neural network,a deeper Convolution Neural Network(CNN)is introduced.The convolution and pooling operations in the network are discussed in detail.Variance cost function is introduced to realize the back propagation training of network error.In the design of the simulation experiment,based on the clinical brain CT database,considering the particularity of medical image data,the structure of CNN network is customized.The network structure includes 1 input layer,7 convolution layers,5 pooling layers and 3 full connection layers.Compared with Alex⁃Net network,the experimental results show that the accuracy of the proposed network structure for brain disease CT classification is 69.28%,AUC is 0.53,and the performance is improved to a certain extent;in specific brain disease recognition,the accuracy of brain tumor recognition can reach 86.5%.
作者
张宏庆
贾利
ZHANG Hongqing;JIA Li(Equipment Department,Weifang Yidu Central Hospital,Qingzhou 262500,China)
出处
《电子设计工程》
2020年第23期48-52,共5页
Electronic Design Engineering
基金
山东省科技厅项目(2017CXGC0413)。
