Integrating the O-RADS MRI Score with Machine Learning: Incremental Value for Predicting Malignancy in Ovarian Tumors

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Ovarian malignancies require accurate preoperative risk stratification for optimal treatment.
• Conventional methods may lack sufficient discriminatory power for differentiating benign from malignant ovarian lesions.
• Integrating standardized reporting systems with advanced analytics can enhance diagnostic performance.

Purpose of the Study:

• To evaluate the incremental diagnostic value of the Ovarian-Adnexal Reporting and Data System (O-RADS) MRI score within a machine learning (ML) framework.
• To compare the performance of a traditional ML model with a combined model incorporating the O-RADS score for discriminating ovarian malignancies.
• To assess the impact of O-RADS score integration on diagnostic accuracy and clinical utility.

Main Methods:

• A retrospective multicenter study involving 257 patients (296 lesions) was conducted.
• Machine learning models were developed using semantic MRI features, clinical data, and the O-RADS score.
• Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) in internal and external validation cohorts.

Main Results:

• The integrated model (Model 2) incorporating the O-RADS score achieved a significantly higher AUC (0.942) compared to the traditional model (0.799) in internal testing (p = 0.005).
• External validation showed improved AUC from 0.791 to 0.909 with the integrated model.
• Decision curve analysis indicated a greater net clinical advantage for the integrated model, with O-RADS identified as a key predictor.

Conclusions:

• Integrating the O-RADS MRI score into an ML framework provides significant incremental diagnostic value for ovarian malignancy discrimination.
• The combined model demonstrates excellent discriminative efficiency, enhancing preoperative risk stratification.
• This approach offers a precise, interpretable, and radiologically based tool for clinical decision-making without additional resource burden.