Application Of Multiparametric MRI Radiomics And Clinical Features To Predict Lymph-Vascular Space Invasion Of Cervical Squamous Cell Carcinoma

Objective:This study aims to investigate the application value of a combined model based on the multiparameter magnetic resonance imaging（MRI）for preoperative prediction of lymph-vascular space invasion（LVSI）in cervical cancer.Methods:A total of 168 patients with cervical squamous carcinoma who were pathologically diagnosed at the Shanxi Cancer Hospital from June 2016 to March 2019 were retrospectively enrolled in this study,with an average age of 22-76（52.0±10.1）years old,including 128 FIGOⅠB cases and 40 FIGOⅡA cases.All patients underwent a multiparametric pelvic MRI scan before the surgery and a radical hysterectomy combined with pelvic lymph node dissection was performed.Patients were divided into two groups,the training group（n=117）and the validation group（n=51）according to the random ratio of 7∶3.Volume regions of interest（VOIs）were manually delineated slice by slice on the T2 weighted images（T₂WI）,apparent diffusion coefficient（ADC）,and enhanced T1 weighted images（cT₁WI）images of each patient.Radiomics features were extracted from each patient.A three-step dimensionality reduction method was used for feature selection and radiomics signature building.The combined radiomics model,including the clinical risk factors and the abovementioned radiomics signature,was constructed via the multivariate logistic regression method,and the corresponding nomogram was constructed.The prediction performance was determined by the calibration,discrimination,and clinical usefulness.Results:Postoperative pathological examination confirmed LVSI positive in 42 patients and negative in 126 patients.No significant differences were found for the general clinical information between the training and validation groups（all P values>0.05）.Seven key radiomics features were obtained after feature selection based on the T₂WI,ADC,and cT₁WI,all of which were significantly associated with lymph-vascular space invasion（all P values<0.05）.The area under the receiver operating characteristic curve（AUC）values of the three radiomics signatures derived from the independent sequence in the training group were 0.630（95%confidence interval[CI]0.557-0.698）,0.686（95%CI 0.563-0.694）,0.761（95%CI 0.702-0.818）and the corresponding AUC of the combined radiomics signature was 0.887（95%CI 0.842-0.925）,which had the best diagnostic efficacy and was validated in the validation group.The AUCs of radiomics nomogram that incorporated radiomics signatures and tumor differentiation degrees were 0.893（95%CI 0.851-0.929）and 0.854（95%CI0.749-0.943）in the training and validation groups,respectively.The calibration curves showed good calibration performance,and when the risk threshold was0.50-0.96,the net benefit of using a radiomics nomogram to predict LVSI was better than treating all patients as LVSI positive or LVSI negative as indicated by the decision curves.Conclusion:Radiomics nomogram based on multiparameter MRI and clinical features has good predictive value for the LVSI status in patients with cervical cancer.