Multi-parametric Magnetic Resonance Imaging-based Radiomics Analysis Of Cervical Cancer For Preoperative Prediction Of Lymphovascular Space Invasion And Prognosis

Objective Lymphovascular space invasion(LVSI)impairs the surgical outcome of cervical cancer(CC)patients.Due to the lack of a single highly reliable factor to predict LVSI before surgery,we explore the predictive performances of radiomics features extracted from pretreatment multi-parameter magnetic resonance imaging(mp MRI)to predict LVSI and clinical outcome in patients with CC.Materials and Methods One hundred twenty-five surgically resected CC patients were retrospectively analyzed in this study.We carried out a radiomic-based characterization of each lesion on the pretreatment mp MRI to develop and validate a noninvasive imaging biomarker capable of distinguishing between LVSI + and LVSI-.Image modalities included a small field of view(s FOV)High-resolution(HR)T2-weighted imaging(T2WI),apparent diffusion coefficient(ADC),T2 WI,and contrast-enhanced T1-weighted(T1c).A total of 107 extracted features in the volume of interest(VOI)of 6 different sequence images were further selected by univariate analysis(least absolute shrinkage and selection operator)LASSO and stepwise logistic regression analysis.A Rad-score and 14 clinical factors were integrated into a predictive model,namely combined(COMB)model,which was built with stepwise logistic regression.Twenty times 3-fold cross-validation was used to evaluate the accuracy of the trained classifiers and the stability of the selected features.Prognostic follow-up was performed.The progression-free survival(PFS)survival curve was divided according to the predicted LVSI group,and the difference in the model for the PFS grouping was observed.Results Radiomics related to intratumoral heterogeneity was the primary feature for predicting LVSI.The related Rad-score was significantly different depending on the LVSI status(p <0.001).Multivariate logistics identified 3 LVSI risk factors,including two clinical factors and an R-score,and the Rad-score(odds ratio [OR] 2.626)was more important than squamous cell carcinoma antigen(OR 1.061)and hemoglobin(OR 0.982).The area under the curve(AUC)of the radiomic model using these predictors was 0.823 in the cohort.Progression-free survival(PFS)was significantly different between LVSI + group and the LVSI-group predicted by the COMB model(median PFS: 64.8 vs.58.3 months).Conclusion The radiomics features combing with mp MRI radiomics and clinical features can predict the LVSI status and clinical outcome of CC patients.It may show utility for improved patient stratification strategies in both surgery and neoadjuvant settings.