From ACR O-RADS 2022 to Explainable Deep Learning: Comparative Performance of Expert Radiologists, Convolutional Neural Networks, Vision Transformers, and Fusion Models for Ultrasound-Based Risk Stratification of Ovarian Masses

This study aims to evaluate the performance of expert radiologists using the updated Ovarian-Adnexal Reporting and Data System (O-RADS) v2022 for risk stratification of ovarian masses and to compare it with various deep learning (DL) models, including convolutional neural networks (CNNs) and vision transformers (ViTs). The findings indicate that while DL models, particularly ViT16-384, outperform radiologist assessments, the integration of radiologist scores with AI significantly enhances diagnostic accuracy, suggesting that hybrid human-AI frameworks could standardize ultrasound interpretation and improve the identification of high-risk ovarian lesions.