Ovarian-adnexal reporting and data system MRI scoring: diagnostic accuracy, interobserver agreement, and...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• The Ovarian-Adnexal Reporting and Data System Magnetic Resonance Imaging (O-RADS MRI) is crucial for standardizing the assessment of adnexal lesions.
• Evaluating interobserver agreement and diagnostic accuracy is essential for refining O-RADS MRI protocols.
• The potential of machine learning (ML) in enhancing diagnostic performance warrants investigation.

Purpose of the Study:

• To assess the interobserver agreement and diagnostic accuracy of O-RADS MRI.
• To evaluate the applicability of machine learning models to O-RADS MRI data.
• To compare the performance of ML models with radiologist assessments.

Main Methods:

• Retrospective analysis of 471 adnexal lesions using dynamic contrast-enhanced pelvic MRI.
• Assessment of lesions by 3 radiologists according to O-RADS MRI criteria.
• Extraction of radiomic features and construction of ML models (ANN, SVM, Random Forest, Naive Bayes).

Main Results:

• Interobserver agreement was lowest for O-RADS 4 (kappa: 0.669) and O-RADS 5 (kappa: 0.709) categories.
• O-RADS MRI predicted malignancy with AUCs of 74.3% for O-RADS 4 and 95.5% for O-RADS 5.
• The artificial neural network (ANN) model achieved the highest performance, with an AUC of 0.948 for distinguishing O-RADS groups.

Conclusions:

• The interobserver agreement and diagnostic sensitivity of O-RADS MRI for O-RADS 4-5 require improvement.
• Machine learning models, especially ANN, demonstrate high accuracy in classifying O-RADS MRI categories.
• Integrating AI into MRI protocols holds promise for enhancing diagnostic performance in adnexal lesion assessment.