Researchers at Karolinska Institutet in Sweden have developed AI models they say outperform human experts in identifying ovarian cancer through ultrasound images.
“Ovarian tumors are common and are often detected by chance,” said Elisabeth Epstein, senior consultant at Stockholm South General Hospital and professor at Karolinska Institutet. “There is a serious shortage of ultrasound experts globally, raising concerns of unnecessary interventions and delayed cancer diagnoses. We wanted to explore if AI could complement human expertise.”
The AI models were trained on over 17,000 ultrasound images from 3,652 patients across 20 hospitals in eight countries. When tested, the neural networks achieved an accuracy rate of 86.3%, outperforming experts (82.6%) and less-experienced examiners (77.7%).
“This suggests that neural network models can offer valuable support in diagnosing ovarian cancer, especially in complex cases and areas lacking ultrasound specialists,” according to Epstein.
In a simulated triage setting, the AI models reduced the need for expert referrals by 63% and lowered the misdiagnosis rate by 18%. This could translate into faster and more cost-effective care for patients with ovarian lesions.
While the results are promising, the researchers emphasized the need for further studies to explore the AI tools’ limitations and adaptability to different clinical environments and patient populations. “With continued research, AI-based tools could relieve experts and optimize hospital resources, but they must be tailored to diverse clinical contexts,” said Filip Christiansen, a doctoral student at Karolinska Institutet and co-first author.
The study authors also disclosed potential conflicts of interest, including patents and ownership stakes in the diagnostic technology company Intelligyn.
The team is currently conducting clinical trials at Södersjukhuset to evaluate the tool's safety and effectiveness in real-world settings. Plans for a multicenter randomized study are also underway to assess its impact on patient care and healthcare costs.
This research was conducted in collaboration with the KTH Royal Institute of Technology and funded by several Swedish research organizations.
