Research On Classification Of Liver CT Images And Texture Information Based On Convolutional Neural Network

Convolutional neural networks have been widely used in medical imaging.However,there are still many further problems in liver disease classification based on CT imaging datasets and CT radiomics features.This paper mainly focused on the classification modeling of liver CT images and liver texture features,which are classified into four disease categories: liver cancer,hepatic cyst,hepatic hemangioma,and hepatic normal.Due to the large difference in the number of samples from different disease categories with liver cancer data accounting for nearly 50%,two-classification and four-classification experiments are designed in this work.This thesis firstly employed vgg-16,Inception-V3,ResNet network to classify the liver imaging datasets.In the two-classification experiment,parameter random initialization and transfer learning are adopted for network pretraining,and the results show that the accuracy of transfer learning is higher than random initialization strategy,while in the four-classification experiment,transfer learning cannot solve the problem of sample proportion imbalance.Therefore,a new IRNet model based on CNN is designed,which can better adapt to sample imbalance than the previous models.Further,the classical machine learning such as SVM and XGBoost were compared with our CNN model in the classification based on liver disease radiomics features.The experimental results show that the classification performance of our CNN is better than the machine learning models,especially our CNN model can adapt to the sample imbalance in the four-classification.