Diagnostic performance and interobserver agreement of O-RADS US v2022 and its combination with serum biomarkers in surgically treated epithelial ovarian-adnexal masses

Abstract

Objective To evaluate the diagnostic performance of the American College of Radiology (ACR) Ovarian-Adnexal Reporting and Data System for Ultrasound (O-RADS US v2022) and a combined O-RADS US v2022–CA125–HE4 model in differentiating benign from borderline or malignant epithelial ovarian-adnexal masses, and to assess interobserver agreement for O-RADS US v2022 classification. Methods This retrospective study included 282 histopathologically confirmed epithelial ovarian-adnexal lesions from 270 surgically treated patients between January 2022 and June 2025. Of the 282 lesions, 233 were benign and 49 were borderline/malignant, with borderline tumors included in the disease-positive group for the primary binary diagnostic analysis. Two sonographers with different levels of experience independently classified stored ultrasound images according to O-RADS US v2022. Categories 4-5 were prespecified as test-positive. Analyses of the combined O-RADS US v2022–CA125–HE4 model were restricted to 238 lesions with complete CA125 and HE4 data. Results Among the 282 lesions, 233 (82.6%) were benign and 49 (17.4%) were borderline/malignant. In a sensitivity analysis excluding borderline tumors, all 20 invasive malignant lesions were classified as O-RADS US v2022 categories 4–5 by both sonographers. Using O-RADS US v2022 categories 4–5 as test-positive, Sonographer A achieved an AUC of 0.924 (95% CI, 0.892–0.955), with a sensitivity of 0.980, specificity of 0.670, PPV of 0.384, and NPV of 0.994. Sonographer B achieved an AUC of 0.904 (95% CI, 0.863–0.945), with a sensitivity of 0.980, specificity of 0.674, PPV of 0.387, and NPV of 0.994. The AUC was 0.769 (95% CI, 0.684-0.853) for CA125 alone and 0.689 (95% CI, 0.593–0.784) for HE4 alone. Among the 238 lesions with complete biomarker data, the combined O-RADS US v2022–CA125–HE4 model yielded an AUC of 0.937 (95% CI, 0.901–0.972), with a sensitivity of 0.721, specificity of 0.954, PPV of 0.775, and NPV of 0.939. Interobserver agreement for O-RADS US v2022 classification was excellent (linearly weighted κ = 0.963; 95% CI, 0.940–0.982). Conclusion In this selected retrospective surgical cohort, O-RADS US v2022 demonstrated good diagnostic discrimination and excellent interobserver agreement. The combined O-RADS US v2022–CA125–HE4 model showed higher specificity but lower sensitivity than O-RADS US v2022 alone, without a statistically significant difference in AUC. Prospective external validation is required before clinical implementation.