作　　者： (吴聪）; (黄中勇）; (殷浩）; (刘罡）; (李江浩）;

机构地区： 湖北工业大学计算机学院,湖北武汉430068

出　　处： 《湖北工业大学学报》 2017年第4期60-64,共5页

摘　　要： 针对脑胶质瘤人工分级难度大、费时费力的情况,提出一种改进的卷积神经网络方法来对脑胶质瘤进行分级。在传统的卷积神经网络结构上增加一层卷积层和采样层,同时使用支持向量机作为分类器;依据大脑结构自动定位肿瘤区域并输入网络进行分类。根据实验结果得出,网络的训练准确率为85.27%,测试准确率为83.79%,均优于传统的网络结构。 Putting forward an improved convolution neural network method for classification of glioma since the glioma artificial classification is difficult and time consuming.In the traditional convolution,neural network structure adds a layer of convolution and sampling,using support vector machine as classifier at the same time.Classification is based on the automatically locating tumor area of the brain structure and the entering network.According to the experimental results it is concluded that,the network training accuracy is 85.27%,the test accuracy is 83.79%,which both were superior to the traditional network structure.

关 键 词： 卷积神经网络 特征提取 脑胶质瘤分级 磁共振成像