Research On TACE For Liver Cancer Based On Intelligent Image

Hepatocellular carcinoma(HCC)is the sixth most common malignancy and the fourth leading cause of tumor-related deaths.In recent years,almost 80% of patients with HCC are initially diagnosed at the intermediate or advanced stage,hence making them unqualified for curative treatments such as resection and ablation.The current standard treatment for unresectable intermediate-stage HCC is transarterial chemoembolization(TACE),which has been proven to prolong survival in a study period of 2–3 years.However,it was reported that some patients do not respond to TACE therapy,which subsequently leads to incomplete necrosis of the tumor and liver function deterioration.The failure could be mainly due to the vast heterogeneity of HCCs that creates a wide array of individual responses.Therefore,a preoperative prediction model is crucial for identifying patients who may benefit from the TACE therapy before any treatment decision is made.This paper analyzed 111 cases of advanced liver cancer patients with preoperative contrast-enhanced CT(CECT)in the First Affiliated Hospital of Soochow University.According to the modified RECIST(m RECIST 1.1),patients were divided into the objectiveresponse(OR)group(n=38)and the non-response(NR)group(n=73).The training set and test set were divided by stratified sampling of 3:7.Based on the ROI region drawn by doctors,1597 tumor features of arterial phase(AP)and venous phase(VP)were extracted from CECT image,including shape feature,first-order gray information,texture feature and wavelet transform feature.Firstly,this paper uses different feature ranking methods(m RMR and WLCX)and classifiers(RF,SVM and LASSO)to model the tumor characteristics of arterial phase,and analyzes the performance differences between different feature ranking methods and classifiers.Then,multi-phase radiomic signatures were built upon integration of images from two different CECT phases by using two kinds of information fusion methods: the decision-level fusion(DLF)and feature-level fusion(FLF).Finally,multivariable logistic regression was used to develop a radiomic-clinical nomogram combining radiomic signatures and clinical characteristics.The prediction performance was evaluated by the area under the curve(AUC)of the receiver-operator characteristic(ROC)on the test dataset.Compared with WLCX,m RMR has better performance(average AUC: 0.771 > 0.700),but there is no significant difference between different classification algorithms,Among them,the AUC of model using m RMR and SVM reached 0.814.The multi-phase radiomic signatures,which were integrated through DLF and FLF,showed better prediction performance(AUC = 0.883 and 0.844).Finally,the radiomic-clinical nomogram that incorporates multi-phase radiomic signature based on DLF,tumor size and tumor number had improved the AUC to a value of 0.913.The results of this study show that the model based on CECT images and clinical data can effectively predict the curative effect of TACE,and provide clinicians with auxiliary decision-making conclusions in the formulation of treatment plan.