Abstract: Objective  To construct a radiomics model based on ¹⁸F⁃FDG PET/CT images, and analyze its efficacy in predicting radiotherapy sensitivity of locally advanced cervical cancer (LACC). Methods The clinical data of cervical cancer patients who underwent radical radiotherapy in the Department of Gynecologic, Guangxi Medical University Cancer Hospital from January 2018 to December 2022 were collected and randomly divided into a training set (169 patients) and a test set (43 patients) in a ratio of 8∶2. Pre⁃treatment  ¹⁸F⁃FDG PET/CT images of patients were collected, the areas of interest were manually delineated and the radiomics features were extracted. The features were regularized and screened by the intra⁃group correlation coefficient, Pearson correlation analysis and the least absolute shrinkage and the selection operator regression. The radiomics features and their weighting efficients obtained and Radscore values were calculated for each patient. The differences in Radscore between radiosensitive and radioresistant groups were compared. The Logistic regression machine learning model for PET, CT and combined PE/CT were constructed, respectively, and the model efficacy was evaluated based on the receive operating characteristic curve (ROC), calibration curve, and decision curve analysis. Results A total of 212 LACC patients were included in this study, and the Radscores of the radioresistant group were significantly higher than those of the radiosensitive group, with a statistically significant difference (P<0.001). In both the training and test sets, the area under the ROC curve of the combined PET/CT model was 0.900 (95%CI: 0.832-0.968) and 0.754 (95%CI: 0.569-0.939), respectively, and the predictive efficiency of the combined PET/CT model was better than that of the individual PET or CT model. The DCA curve showed that the PET/CT⁃based radiomics models showed significant intervention benefits and had better clinical benefits in predicting LACC sensitivity before radiotherapy compared to those without the prediction model. The calibration curve showed that the predicted values of the three models were in good agreement with the observed values. Conclusions The PET/CT⁃based radiomics models has a good predictive value for the sensitivity of LACC to radiotherapy.

Key words: Cervical cancer, Locally advanced, Radiosensitivity, PET/CT radiomics, Machine learning model