Transformer-based neural network models surpass diagnostic performance of examiners for identifying ovarian cancer in ultrasound images
By Elana Gotkine HealthDay Reporter
THURSDAY, Jan. 9, 2025 (HealthDay News) — Transformer-based neural network models exhibit strong generalization and surpass diagnostic performance of expert and nonexpert examiners for detecting ovarian cancer in ultrasound images, according to a study published online Jan. 2 in Nature Medicine.
Filip Christiansen, from the Karolinska Institutet in Stockholm, and colleagues developed and validated transformer-based neural network models using a comprehensive dataset of 17,119 ultrasound images from 3,652 patients across 20 centers in eight countries. For each center in turn, a model was trained using data from the remaining centers, using a leave-one-out cross-validation scheme.
The researchers found that across centers, ultrasound systems, histological diagnoses, and patient age groups, the models demonstrated robust performance, significantly outperforming expert and nonexpert examiners on all evaluated metrics (F1 score, sensitivity, specificity, accuracy, Cohen’s kappa, Matthew’s correlation coefficient, diagnostic odds ratio, and Youden’s J statistic). Artificial intelligence (AI)-driven diagnostic support reduced referrals to experts by 63 percent in a retrospective triage simulation, while significantly surpassing the diagnostic performance of current practice.
“Our study demonstrates the potential of AI models in improving the accuracy and efficiency of ovarian cancer diagnosis,” the authors write. “Our models demonstrated robust generalization and significantly outperformed both expert and nonexpert examiners on all evaluated metrics.”
Several authors disclosed ties to medical technology companies, including Intelligyn; several authors have applied for a patent that is pending to Intelligyn.
Copyright © 2025 HealthDay. All rights reserved.
