Development and validation of the integrative model and risk stratification based on deep learning and radiomics to predict survival of advanced cervical cancer patients

Abstract Purpose Advanced cervical cancer (aCC) is aggressive and has a low 5-year survival rate. A novel integrative model needs to be developed to predict survival and guide clinical practice. Procedures: Patients were retrospectively studied from July 2014 to August 2016. The dataset was randomly split into training and validation cohorts. A deep learning model based on a Vision Transformer (ViT) and Recurrent Neural Network (RNN) was built to output a prognostic score (Rad-score) based on CT images. After combining Rad-score with clinical and pathological characteristics, Cox regression analysis was performed to establish the clinical model, radiomics model and integrative model, respectively, and the C-index was calculated. The time-dependent C-index curve, receiver operating characteristic (ROC) curve, calibration curve and decision curve analysis (DCA) were drawn to evaluate the performance of the model. Finally, a Nomogram was drawn to visualize the integrative model and develop the risk stratification. Results 396 patients were eventually enrolled in the study. The Cox regression analysis indicated that the Rad-score was an independent prognostic factor (HR = 4.06, 95% CI: 2.46–6.70, p < 0.001). Kaplan-Meier analysis showed that Rad-score allows significant risk stratification of patients. The integrative model achieved better performance in training (C-index: 0.784[0.733–0.835]) and validation cohorts (C-index: 0.726[0.677–0.785]) showed good survival prediction performance, which was superior to clinical model (training cohort: 0.686[0.625–0.747], validation cohort: 0.632[0.569–0.695]) and radiomics model (training cohort: 0.730[0.675–0.785], validation cohort: 0.723[0.658–0.788]). In addition, the ROC curve, time-dependent C-index curve, calibration curve and DCA curve of the integrative model all showed significant predictive performance advantages. Finally, the Nomogram visualized the survival prediction model. Kaplan-Meier curve showed the risk stratification had significant clinical value. Conclusions The integrative model and risk stratification based on Rad-score, clinical and pathological characteristics can be widely used in clinical practice to provide reliable clinical information for medical decisions.