Quantitative Analysis Of Liver Cancer Based On Dynamic Contrast-enhanced Medical Image

Since entering the era of artificial intelligence and big data,the development of medical field is more rapid.Cancer is one of the most challenging disease.Liver cancer is a cancer disease with high incidence rate and high mortality.Therefore,there are great demands for the diagnosis,treatment and prevention of liver cancer.It is of great significance to research and develop more efficient and precise liver cancer treatment technology.Based on the dynamic contrast-enhanced CT(DCE-CT)images,in this paper,a quantitative analysis method of liver cancer is proposed,which provides a powerful technical support for the analysis of medical big data of liver cancer.In this paper,the multi-phase DCE-CT images of liver are segmented and registered,and a quantitative analysis method of liver cancer based on multi-phase DCE-CT image is designed.The specific research contents are as follows:According to the different characteristics of the same area in different periods of DCE-CT image,the U-net model is applied to the scene of multi-phase DCE-CT simultaneous segmentation,the liver region of multi-phase image is segmented at the same time.6617 DCE-CT images from 16 patients with liver cancer were used to train the U-net model,1754 DCE-CT images from 6 patients were used as testing set,and the average of the Dice value between results and labels was 0.91 ± 0.02.The results show that the model can effectively segment the liver region in multi-phase DCE-CT images at the same time.In order to solve the problem of liver region mismatch in CT images of different periods,a liver registration scheme based on multi-phase DCE-CT image segmentation is proposed.The 3-D reconstruction is carried out by superposition by liver region of multi-stage DCE-CT images,and the centroid of each 3-D liver model is calculated,so the problem of 2-D image registration is transformed into the alignment of Centroid of the 3-D liver model.After experimental demonstration,the registration scheme can register the liver region in multi-stage CT images.It can provide strong technical support for the registration and information fusion DCE-CT images of liver.For the quantitative analysis of liver cancer,a scheme of liver cancer classification based on DCE-CT images is proposed.After the registration,extracting the information of each pixel in the image of liver region,training the LightGBM model,and evaluateing by cross validation,the results of SVM?Random Forest(RF)?K-Nearest-Neighbor(KNN)were compared with LightGBM.Through experiments,the accuracy of LightGBM is 91.14%,and its performance is obviously better than other models.The 3-D Dice values between the results of active liver cancer area and labels in CT images of different subjects by different classifiers are compared.The results show that the accuracy of LightGBM for active liver cancer area is higher than SVM?RF and KNN,which can correctly distinguish the area of liver cancer,and the whole project provided a reliable basis for quantitative analysis of liver cancer.