Establishment And Validation Of Survival Prediction Model Based On MRI Radiomics For Cervical Cancer Patients After Radiotherapy And Chemotherapy

Objective: To explore the construction of radiomic label(Rad＿Score)based on MRI-T2WI images related to the prognosis of cervical cancer after radiotherapy and chemotherapy,and to establish an individual prediction model combined with clinical characteristic variable data,in order to achieve the prediction of survival status of cervical cancer patients with stage Ⅱ～Ⅲcervical cancer after radiotherapy and chemotherapy.Methods: A total of 120 inpatients with cervical cancer diagnosed by histopathology in our hospital from June 2016 to February 2019 were collected.Randomly divided into training set(n=84)and verification set(n=36),MRI image data of patients before treatment were collected and sorted out,and radiomic features were extracted from GTV of tumor target region delineated.After “ t ” test,Least absolute shrinkage and selection operator(LASSO)analysis,Select different radiomic characteristic parameters.By using machine learning algorithms(including Logistic regression,Knearest neighbor,naive Bayes,decision tree,random forest,gradient lifting tree and support vector machine),the most stable Logistic regression function algorithm was obtained and Radiomic label(Rad＿Score)was generated.Using R language software package based on Rad＿Score and clinical variables,a nomogram was established to predict 3-year local recurrence-free survival.Decision curve analysis(DCA)was used to evaluate the clinical benefit of multifactorial nomogram in predicting local recurrence in patients with cervical cancer after completion of treatment.Results: A total of 1561 radiomic features were extracted,and nine stable features were selected after LASSO analysis for machine learning algorithm modeling to generate Rad＿Score.The higher Rad＿Score was associated with an increased risk of local recurrence.The C index,specificity and sensitivity in the training set were 0.908(95%CI: 0.8463-0.9699),92.5%and 74.2%,respectively.In the validation set,they were 0.809(95%CI:0.6482-0.9705),95.7% and 69.2%,respectively.Conclusion(s): The ability to predict local recurrence risk of cervical cancer patients after radiotherapy and chemotherapy based on the nomogram model constructed based on MRI-T2WI radiomic characteristics and clinical variables maybe provide evidence support for clinicians to conduct early intervention treatment...