Enhanced CT-based Radiomics Nomogram In Predicting The Probability Of Postoperative Adjuvant Radiotherapy In Early-stage(FIGO ⅠB-ⅡA) Cervical Cancer

ObjectivesRadical surgery and radcial radiochemotherapy are both first-line recommendations for FIGO(Federation International of Gynecology and Obstetrics)stage ⅠB-ⅡA of cervical cancer due to the similar survival.For patients with risk factors,it’s necessary to accept adjuvant radiation or chemoradiation,which can reduce the risk of recurrence.However,it makes these patients suffer from double complications due to surgery and radiotherapy but no survival benefit.It is a better choice to accept radical radiation therapy for those with risk factors.But there is no non-invasive way to foresee whether you need adjuvant radiotherapy.Radiomics is an emerging approach to extract large amounts of high-dimensional quantitative features from multimodality medical images,which can reflect the holistic outlook of cancer.CT(Computed Tomography)-based radiomics has been used in outcome prediction in different kinds of cancers,which show a promising future.Thus,we aim to use enhanced CT-based radiomics features and clinical features to predict the probability of postoperative adjuvant radiotherapy.Methods200 patients with ⅠB-ⅡA cervical cancer who received radical hysterectomy in Cancer Hospital of Shantou University College were enrolled.Radiomic features were extracted from enhanced pelvic CT imaging of each patient and reduced by three methods.Clinical features were sifted by univariate and multiple regression analysis.Three models were developed with radiomic features,clinical features,and both respectively using logistic regression in training cohort(n=140)and tested in a validation cohort of 60 patients.Nomogram was drawn based on the best model which was chosen using Delong test,and its discrimination and calibration performances were estimated.Results1595 radiomic features were extracted from each patient and 8 features were finally involved to build the radiomic model,which yielded an AUC(Area Under Curve)of 0.761(95%CI 0.678-0.845)vs 0.748(95%CI 0.624-0.872)in the training and validation cohorts.FIGO stage,lymph node metastasis in CT,pathological type,tumor size by palpation and the number of RBC(Red Blood Cell)were proved to be meaningful,and model based on them showed a good performance with an AUC of 0.887(95%CI 0.893-0.979)vs 0.815(95%CI 0.772-0.963)in the training and validation cohorts.The combined model incorporating radiomics features with clinical factors had a best predictive performance with an AUC of 0.936(95%CI 0.835-0.940)in the training cohort and an AUC of 0.867(95%CI 0.703-0.926)in the validation cohort.Subgroup analysis proved the importance of radiomic features and the combined model had been recognized to be the best model according to Delong test.Nomogram based on the combined model showed excellent discrimination and the calibration curve showed a good agreement.ConclusionsNomogram basd on enhanced CT radiomics and clinical featurses demonstrated a good prediction ability in predicting the probability of postoperative adjuvant radiation of early stage(FIGO ⅠB-ⅡA)cervical cancer patients,which had the prospect of clinical application as a noninvasive biomarker.