AI Outperforms Human Experts in Ovarian Cancer Detection, New Study Finds

In a study from Sweden’s Karolinska Institutet, AI has demonstrated superior accuracy in detecting ovarian cancer compared to human doctors. Published in Nature Medicine, the AI cancer detection study offers hope for improved diagnostic tools for the nearly 20,000 women in the U.S. diagnosed with ovarian cancer annually.

Key Findings

The AI model was trained using over 17,000 ultrasound images from 3,652 patients across 20 hospitals in eight countries. According to the study, the AI achieved an 86% accuracy rate in identifying ovarian cancer, surpassing the 82% accuracy rate of human experts and 77% for less experienced examiners.

AI cancer detection study author Elisabeth Epstein, a professor at Karolinska Institutet, noted, “I was surprised that the AI models outperformed all 33 of the expert examiners.”. Epstein highlighted the benefits of AI in ovarian cancer diagnosis, emphasizing its potential to reduce diagnostic errors, address the shortage of expert examiners, and enhance access to high-quality diagnostics in underserved regions. “This will help reduce waiting times, avoid unnecessary interventions, and facilitate earlier cancer detection, ultimately improving patient outcomes,” she added.

Potential Impact

Ovarian tumors are often detected by chance, making early and accurate diagnosis crucial. Dr. Brian Slomovitz, director of gynecologic oncology at Mount Sinai Medical Center in Florida, called early detection the “holy grail” for reducing deaths from ovarian cancer.

“This large retrospective trial clearly demonstrates that there may be a role in incorporating AI-driven support to better interpret ultrasound findings in patients with a pelvic mass,” he said.

Dr. Harvey Castro, an emergency medicine physician and AI expert, acknowledged the promise of AI in improving cancer diagnostics but warned of its current limitations. “The AI relies on diverse, high-quality data, and bias could limit effectiveness,” Castro explained.

He also pointed to regulatory and transparency challenges that need addressing before AI becomes routine in clinical settings.

Challenges and Next Steps

While the AI cancer detection study results are promising, the researchers acknowledged limitations, noting that the data is retrospective. Epstein emphasized the need for prospective studies to evaluate AI performance in real-world clinical settings.

Clinical trials are set to begin at Stockholm South Hospital in Sweden to further validate the AI’s effectiveness. Epstein also clarified that AI should complement, not replace, human physicians. “It is still the physician who remains responsible for the patient’s diagnosis and treatment,” she stated.

Future Directions

Experts believe incorporating additional patient factors — such as menopausal status, symptoms, and blood test results — could enhance AI’s diagnostic capabilities. Moreover, demonstrating a survival benefit will be key to achieving widespread adoption.

The study was supported by the Swedish Research Council, the Swedish Cancer Society, and other organizations, highlighting the global interest in advancing AI in healthcare. As AI continues to evolve, its role in cancer diagnostics could transform patient care, offering more accurate diagnoses, reducing unnecessary interventions, and ultimately saving lives.