MRI-derived radiomics analysis improves the noninvasive pretreatment identification of multimodality therapy candidates with early-stage cervical cancer

Objectives To develop and validate a clinical-radiomics model that incorporates radiomics signatures and pretreatment clinicopathological parameters to identify multimodality therapy candidates among patients with early-stage cervical cancer. Methods Between January 2017 and February 2021, 235 patients with IB1-IIA1 cervical cancer who underwent radical hysterectomy were enrolled and divided into training ( n = 194, training:validation = 8:2) and testing ( n = 41) sets according to surgical time. The radiomics features of each patient were extracted from preoperative sagittal T2-weighted images. Significance testing, Pearson correlation analysis, and Least Absolute Shrinkage and Selection Operator were used to select radiomic features associated with multimodality therapy administration. A clinical-radiomics model incorporating radiomics signature, age, 2018 Federation International of Gynecology and Obstetrics (FIGO) stage, menopausal status, and preoperative biopsy histological type was developed to identify multimodality therapy candidates. A clinical model and a clinical-conventional radiological model were also constructed. A nomogram and decision curve analysis were developed to facilitate clinical application. Results The clinical-radiomics model showed good predictive performance, with an area under the curve, sensitivity, and specificity in the testing set of 0.885 (95% confidence interval: 0.781–0.989), 78.9%, and 81.8%, respectively. The AUC, sensitivity, and specificity of the clinical model and clinical-conventional radiological model were 0.751 (0.603–0.900), 63.2%, and 63.6%, 0.801 (0.661–0.942), 73.7%, and 68.2%, respectively. A decision curve analysis demonstrated that when the threshold probability was > 20%, the clinical-radiomics model or nomogram may be more advantageous than the treat all or treat-none strategy. Conclusions The clinical-radiomics model and nomogram can potentially identify multimodality therapy candidates in patients with early-stage cervical cancer. Key Points • Pretreatment identification of multimodality therapy candidates among patients with early-stage cervical cancer helped to select the optimal primary treatment and reduce severe complication risk and costs. • The clinical-radiomics model achieved a better prediction performance compared with the clinical model and the clinical-conventional radiological model. • An easy-to-use nomogram exhibited good performance for individual preoperative prediction.