Diagnostic Performance of Computed Tomography-Based Machine Learning Models in the Classification of Adnexal Masses - A Systematic Review.

Introduction: Accurate characterization of adnexal masses is a key issue and a crucial step toward improving the outcome of managing a patient with a gynecologic oncology issue. Though ultrasound is a dominant tool for this process, it is subjected to operator variability and is less reliable from a diagnostic perspective. Advances in computed tomography-based radiomics and ML hold great promise as objective diagnostic solutions. Methods: This systematic review was performed according to the guidelines suggested by PRISMA. The literature research using PubMed, Embase, Scopus, and Web of Science databases included studies that examined CT-based radiomics and ML model performances for classification of adnexal masses and reported diagnostic performance metrics, including AUC, sensitivity, and specificity. Quality assessment of included studies was performed using the QUADAS 2 tool. Results: Eleven studies were included in the review. The performance of CT-based ML models was found to be moderate to excellent, with an AUC ranging from 0.72 to 0.99. Hybrid radiomics-DL algorithms were found to have a higher performance compared to other algorithms. The studies were found to have low risk of bias. Conclusion: CT-based radiomics and AI models also hold good prominence as adjunctive tools in differentiating between both benign and malignant adnexal masses and in predicting prognosis.