Study On The Curative Effect Of Radical Radiotherapy For Cervical Cancer Based On Contrast-enhanced CT Radiomics

Objective:To investigate the twoclassification predictionof the efficacy of radical radiotherapy in patients with cervical cancer based on enhanced CT radiomics,and to further predict its fourclassification,so as to provide basic support for the clinical precision treatment of cervical cancer.Materials and Methods:A retrospective analysis of 316 patients with cervical cancer admitted to the Department of Radiation Oncology,Affiliated Hospital from July 2016 to July 2021.The results were as follows:77 cases of complete remission(CR),161 cases of partial remission(PR),71 cases of stable disease(SD)and 7 cases of progressive disease(PD).Five general clinical data were collected:age,pathological type,federation of international of gynecologists(FIGO)stage,age of first pregnancy and number of pregnancies,and then they were subjected to factorial variance analysis,and those with statistical significance were screened out.Clinical information of scientific significance for further analysis.The pre-treatment enhanced CT images of the included patients were used to delineate the region of interest(ROI),and the relevant radiomics features were extracted.Then,the data were balanced to 644 cases using the adaptive synthetic sampling method.When performing two classification prediction:it is divided into two categories about valid(CR+PR)and invalid(SD+PD).According to the two classification labels,the most relevant radiomics features are obtained after feature screening through mutual information,LASSO regression,and recursive feature elimination based on random forests.These radiomics features were introduced into logistic regression(LR)algorithm and support vector machine(SVM)algorithm for model training,validation and testing,and finally the optimal model was selected according to model evaluation indicators.When performing four classification prediction:it is divided into four categories about CR,PR,SD,and PD.Similarly,according to the four classification labels,the most relevant radiomic features are obtained through mutual information and random forest recursive feature elimination method.And these radiomics features are brought into LR and SVM algorithms for model training,validation and testing,and finally the optimal model was selected according to model evaluation indicators.Results:Statistical analysis of general clinical information:No matter in the two or four classification,all 5 items of clinical information were not statistically significant(P>0.05).For twoclassification prediction:select the 5 most relevant radiomics features after feature screening.For the LR model test set accuracy(0.709)and AUC value(0.741);for the SVM model test set accuracy(0.736),AUC value(0.772).The two models were tested by the Delong test P=0.03 7<0.05,and the AUC difference was statistically significant.For fourclassificationprediction:After feature screening,the 5 most relevant radiomics features were selected.For the LR model test set,the accuracy rate(0.531),the micor AUC value(0.755),and the macro AUC value(0.737);the SVM model test set was accurate rate(0.636),micor AUC value(0.827),macor AUC value(0.803).The two models were tested by Delong’s test,P=0.046<0.05,and the difference in AUC was statistically significant.Conclusion:1.After a rigorous process,radiomics based on enhanced CT can not only predict the efficacy of radical radiotherapy for cervical cancer in two categories,but also predict the four categories to a certain extent.2.In this study,the SVM model has better accuracy and stability than the LR model for both binary and quaternary classification.3.Support vector machine model can well predict the four categories of curative effect of radical radiotherapy for cervical cancer(Micor AUC=0.827).