Reviewed by Lexie CornerJan 3 2025
An international study led by researchers at Karolinska Institutet in Sweden has shown that AI-based models can outperform human experts in identifying ovarian cancer from ultrasound images. The findings were published in Nature Medicine.
Image Credit: Black Salmon/Shutterstock.com
Ovarian tumors are common and are often detected by chance. There is a serious shortage of ultrasound experts in many parts of the world, which has raised concerns of unnecessary interventions and delayed cancer diagnoses. We, therefore, wanted to find out if AI can complement human experts.
Elisabeth Epstein, Professor, Department of Clinical Science and Education, Karolinska Institutet
Elisabeth Epstein is also a Senior Consultant at Södersjukhuset’s (Stockholm South General Hospital) Department of Obstetrics and Gynecology.
AI Outperforms Experts
The AI models were trained and tested on over 17,000 ultrasound images from 3,652 patients across 20 hospitals in eight countries. Neural network models were developed and validated to distinguish between benign and malignant ovarian lesions.
The models were then compared with a panel of experienced and novice ultrasound examiners. The results showed that the AI models achieved an accuracy rate of 86.3 %, surpassing the 82.6 % accuracy of experts and 77.7 % of non-experts.
This suggests that neural network models can offer valuable support in the diagnosis of ovarian cancer, especially in difficult-to-diagnose cases and in settings where there is a shortage of ultrasound experts.
Elisabeth Epstein, Professor, Department of Clinical Science and Education, Karolinska Institutet
Reducing the Need for Expert Referrals
The AI models also reduced the need for expert referrals. In a simulated triage scenario, the AI decreased the misdiagnosis rate by 18 % and reduced referrals by 63 %, potentially allowing for quicker and more cost-effective treatment for patients with ovarian lesions.
Despite these encouraging results, the researchers caution that further studies are needed to fully understand the potential and limitations of neural network models in clinical practice.
With continued research and development, AI-based tools can be an integral part of tomorrow’s healthcare, relieving experts and optimizing hospital resources, but we need to make sure that they can be adapted to different clinical environments and patient groups.
Filip Christiansen, Study Joint First Author, KTH Royal Institute of Technology
Christiansen is also a Doctoral Student in Professor Epstein’s research group at Karolinska Institutet.
Evaluating the Safety of the AI Support
To ensure the clinical safety and effectiveness of the AI tool, prospective clinical studies are underway at Södersjukhuset. Future research will include a randomized multicenter study to evaluate its impact on patient care and healthcare costs.
The study was conducted in collaboration with researchers from KTH Royal Institute of Technology and funded by the Swedish Research Council, the Swedish Cancer Society, the Stockholm Regional Council, the Cancer Research Funds of Radiumhemmet, and the Wallenberg AI, Autonomous Systems and Software Program (WASP).
Epstein, Christiansen, and three co-authors have also submitted a patent application for computer-supported diagnostic techniques through the business Intelligyn.
Journal Reference:
Christiansen, F., et al. (2025) International multicenter validation of AI-driven ultrasound detection of ovarian cancer. Nature Medicine. doi.org/10.1038/s41591-024-03329-4.