Research On Hepatic Cystic Echinococcosis CT Image Classification Method Based On Deep Learning

Objective: To explore the application of deep learning convolutional neural network method and technology in CT image classification of hepatic cystic echinococcosis,in order to provide auxiliary information with practical reference value for radiologists' diagnostic decision,and improve the accuracy and efficiency of the diagnosis of hepatic echinococcosis..Methods:(1)The CT images of hepatic cystic echinococcosis were preprocessed by Python image processing technology,which includes normalization and data enhancement.The images were uploaded into improved ResNet,LeNet and InceptionV3 models for training in batches.The improved training model uses the Relu and Softmax activation functions,the ResNet model uses the SGD optimization function,and the LeNet and InceptionV3 models use the Adadelta optimization function.The data set was divided into training group and test group according to 8:2.The test group adopted the cross-validation method,and the learning rate was 0.005.The training model was imported into the classification model,and the three-dimensional matrix was output after labeling monocystic type,polycystic type and solitary cystic with 0,1 and 2sequences,and the type of the image was determined according to the position and ratio of the matrix.(2)The traditional MATLAB image processing software was used to manually intercept the lesions and preprocessed by normalization,desiccation and enhancement.The images were classified by the ten-fold cross validation method using the C4.5 decision tree and the support vector machine classifier after the 13-dimensional eigenvector values of the gray co-occurrence matrix and gray histogram were extracted;(3)the accuracy,sensitivity and specificity of the two methods were used to evaluate the performance of each classification model.Results:(1)Deep learning convolutional neural network method: the best average training accuracy of the three kinds of hepatic echinococcosis CT images in the ResNet,LeNet and InceptionV3 training models were93.22%,76.55% and 90.75%,respectively.In the classification model,the accuracy was0.93,0.68 and 0.81,the sensitivity was 0.92,0.58 and 0.76,and the specificity was 0.94,0.78 and 0.86.(2)Traditional methods: the classification accuracy of two feature vectorvalues in the C4.5 decision tree classifier was 87.6% and 89.31%,respectively.The classification accuracy of SVM classifier is 68.34% and 81.93%,respectively.The accuracy,sensitivity and specificity of C4.5 decision tree classifier were 0.92,0.89 and0.94,respectively.and those of support vector machine classifier are 0.87,0.83 and 0.91 respectively.Conclusion:(1)Deep Learning convolutional neural network method: the ResNet model is superior to LeNet and InceptionV3 models in training,image feature learning efficiency and evaluation indexes.Three classification models had the best classification effect on CT images of poly cystic hepatic echinococcosis,followed by monocystic hepatic echinococcosis.Considering the classification of CT images of hepatic cystic echinococcosis,ResNet model is most suitable for the accurate classification of three kinds of CT images.(2)Traditional methods: the classification effect of C4.5 decision tree classifier is better than that of SVM classifier,and the efficiency of gray histogram feature vector value is the best.The gray histogram eigenvector values used in C4.5 decision tree classifier are more suitable for the classification of CT images of hepatic cystic echinococcosis.(3)The deep learning convolutional neural network model is more feasible and reasonable for CT image classification of hepatic cystic echinococcosis than the traditional method.The results of this study can provide valuable references for radiologists in the diagnosis of hepatic cystic echinococcosis and lay a foundation for the development of a Computer Aided Diagnosis system for clinical hepatic cystic echinococcosis.