PURPOSE: The International Federation of Gynecology and Obstetric (FIGO) stage is critical to guiding the treatments of ovarian cancer (OC). We tried to develop a model to predict the FIGO stage of OC through machine learning algorithms with patients' pretreatment clinical, positron emission tomography scan (PET/CT) metabolic, and radiomics features.

METHODS: We enrolled OC patients who underwent PET/CT scans and divided them into two cohorts according to their FIGO stage. Then we manually delineated the volume of interest (VOI) and calculated PET metabolic features. Other PET/CT radiomics features were extracted by Python. We developed 11 prediction models to predict stages based on four groups of features and conducted three experiments to verify the meaning of PET/CT features. We also redesigned experiments to demonstrate the stage prediction performance in ovarian clear cell carcinoma (OCCC) and mucinous ovarian cancer (MCOC).

RESULTS: 183 OC patients were enrolled in this study, and we obtained 137 features from four groups of data. The best model was an adaptive ensemble with an area under the curve (AUC) value of 0.819. Our proposed models presented the best result of 0.808 in terms of AUC in OCCC and MCOC patients' groups.

CONCLUSION: Through artificial intelligence (AI) algorithms, the PET/CT metabolic and radiomics features combined with clinical features could improve the accuracy of staging prediction.