Research And Application Of The Classification Of Hepatocellular Carcinoma And Intrahepatic Cholangiocarcinoma Based On The Medical-Cross-Contrast Neural Networks

Hepatocellular carcinoma(HCC)and intrahepatic cholangiocarcinoma(ICC)are the two most common primary liver cancers.The main clinical features of liver cancers are low predictability,rapid deterioration,and easy death.In addition,the choice of surgical plan and the prognosis assessment of these two liver cancers are very different.Therefore,it is important to create a non-invasive method to classify hepatocellular carcinoma and intrahepatic cholangiocarcinoma,which is good for the early diagnosis and treatment of l iver cancers.Compared with the images of MRI and ultrasound in the liver,the liver CT images have the advantages of good imaging quality and high resolution.However,using the patients’ CT images to study the classification of these two cancers has two problems.On one hand,the traditional feature extraction method is time-consuming and laborious,moreover,the features are fixed and single.On the other hand,compared with the larger data set used for deep learning,this medical data set for the study is small and contains 234 patients.In view of the above situation,we propose the Medical-Cross-Contrast Neural Networks(MCCNN)to classify the hepatocellular carcinoma and the intrahepatic cholangiocarcinoma on CT images,which combines the deep correlation convolution network VGG and the Information Based Similarity(IBS)method of similarity measurein statistical analysis.The main contents of this article are as follows.Firstly,the background,significance and pathogenesis of hepatocellular carcinoma and intrahepatic cholangiocarcinoma are introduced.Secondly,the Siamese network,deep convolutional network and the IBS method of statistical analysis are briefly introduced,which combine the Medical-Cross-Contrast Neural Networks(MCCNN).In addition,the training and testing process of this network are briefly explained.Finally,we apply this method on a 234-person(82 HCC,73 ICC,79 Normal)dataset and achieve fine results on this relatively small dataset than other deep convolutional networks.The classification slice/patient accuracy of two liver cancers on test set is 82.5%/87.2%in the two liver cancers and 81.8%/86.6%in the three categories.In comparison,the better accuracy rate in the classification of hepatocellular carcinoma and intrahepatic cholangiocarcinoma is 69.7%and the better accuracy in predicting microvascular invasion of hepatocellular carcinoma is 80-82.8%from the team of Professor Zhang of Jiangsu Provincial People’s Hospital.The MCCNN performs well,indicating the applicability and dominance of the method in hepatocellular carcinoma and intrahepatic cholangiocarcinoma.