The Value of Model Based on Radiomics Features of T2-weighted Imaging and Clinical Feature in Diagnosing the Depth of Stromal Invasion of Cervical Squamous Cell Carcinoma

  Objective  To investigate the prediction value of a clinical-radiomics model based on T2- weighted imaging (T2WI) and clinical features for diagnosing deep stromal invasion (DSI) in patients with early-stage cervical squamous cell carcinoma.

  Methods  Patients with early-stage cervical squamous cell carcinoma that underwent radical hysterectomy in Peking Union Medical College Hospital from January 2017 to February 2021 were retrospectively included and randomly divided into the training set and the validation set with the the ratio of 8∶2. The preoperative clinical features and the radiomics features of sagittal T2WI images were obtained. After selection of key features, a radiomics model, a clinical model, and a clinical-radiomics model for diagnosing DSI in early-stage cervical squamous cell carcinoma were developed by Logistic regression based on the training set. The performance of different models was compared by the receiver operating characteristic (ROC) curve in the validation set.

  Results  A total of 168 patients with early-stage cervical squamous cell carcinoma that met the inclusion and exclusion criteria were included in this study. They were randomly divided into the training set (n=135) and the validation set (n=33), in which 72 cases had histopathologically confirmed superficial stromal invasion and 96 cases had DSI. Four radiomics features and three clinical parameters (age, Federation International of Gynecology and Obstetrics stage, and preoperative squamous cell carcinoma antigen levels) were selected and used to develop models. In the validation set, the clinical-radiomics model showed better diagnostic performance with the area under the curve (AUC) of 0.820 (95% CI: 0.665-0.974) than the clinical modelAUC: 0.797(95% CI: 0.623-0.971) and the radiomics modelAUC: 0.793(95% CI: 0.633-0.954).The sensitivity, specificity, and accuracy of the clinical-radiomics model were 85.7%(95% CI: 49.8%-100%), 73.7%(95% CI: 57.9%-100%), and 78.8%(95% CI: 69.7%-93.9%), respectively.

  Conclusion  Radiomics features based on T2WI images combined with clinical features can be used as a noninvasive preoperative method to determine the depth of stromal invasion in early-stage cervical squamous cell carcinoma.