Erin T. Welsh, MA , 2025-05-17 00:43:00
Key takeaways:
- Four adnexal mass classification algorithms successfully distinguished lesions when used by a novice operator.
- A two-step strategy performed better than O-RADS algorithms.
MINNEAPOLIS — Four commonly used adnexal mass classification algorithms successfully identified lesions as benign or malignant when used by a novice operator, according to data presented at the ACOG Annual Clinical & Scientific Meeting.
“Adnexal masses are frequent incidental findings of pelvic ultrasound. Most of them are benign, but approximately 20% of them are malignant cancers,” Luigi A. De Vitis, MD, a gynecologic oncologist fellow at the College of Medicine and Science at Mayo Clinic, said during the presentation. “Every additional enhancement in detection and treatment of suspicious lesions in gynecologic oncology centers significantly improves prognosis and, on the other hand, identifying lesions that do not require surgery reduces patient exposure to unnecessary risks and expenses.”
Adnexal mass classification algorithms successfully identified lesions as benign or malignant, even when used by a novice operator. Image: Adobe Stock
De Vitis and colleagues identified 556 women with adnexal masses who were treated at Mayo Clinic in Rochester, Minnesota in 2019. A nonexpert operator classified each lesion using the Assessment of Different Neoplasias in the Adnexa (ADNEX), a two-step strategy with benign descriptors followed by ADNEX, Ovarian-Adnexal Reporting and Data System (O-RADS) 2019 and O-RADS 2022.
The primary outcome was the area under the receiver operating characteristic curve (AUC) among the classification algorithms.
Overall, 452 women had benign and 104 had malignant adnexal masses. The AUC was:
- 0.9 (95% CI, 0.87-0.94) for ADNEX;
- 0.91 (95% CI, 0.88-0.94) for the two-step strategy;
- 0.88 (95% CI, 0.84-0.91) for O-RADS 2019; and
- 0.88 (95% CI, 0.84-0.91) for O-RADS 2022.
The two-step strategy outperformed O-RADS 2019 and O-RADS 2022, but the “difference was not clinically meaningful,” De Vitis said.
The observed malignant lesion rate for all algorithms ranged from 1.9% to 2.2% among lesions that were characterized as “almost certainly benign,” which was almost double the expected rate of less than 1%, according to De Vitis and colleagues. In addition, of the lesions that were wrongly classified as almost certainly benign, four were borderline tumors and three were metastases.
“Although these models can be used as a triage system in referral centers, we strongly suggest use in peripheral centers where experts are not available to use this model to ultimately increase ovarian cancer detection and hopefully improve patient outcomes,” De Vitis said.
