AI-based models can outperform human experts in detecting ovarian cancer

AI-based models can outperform human experts in detecting ovarian cancer

A new international study led by researchers at Karolinska Institutet in Sweden shows that AI-based models can outperform human experts in identifying ovarian cancer in ultrasound images. The study is published inNatural medicine.

“Ovarian tumors are common and are often discovered by chance,” says Professor Elisabeth Epstein from the Department of Clinical Science and Training at Södersjukhuset (Stockholm South General Hospital) at Karolinska Institutet and senior consultant in the hospital’s Department of Obstetrics and Gynecology. "There is a severe shortage of ultrasound experts in many parts of the world, which has led to concerns about unnecessary procedures and delayed cancer diagnoses. We wanted to find out whether AI could complement human experts."

AI outperforms experts

Researchers developed and validated neural network models capable of distinguishing between benign and malignant ovarian lesions by training and testing the AI ​​on over 17,000 ultrasound images from 3,652 patients in 20 hospitals in eight countries. They then compared the diagnostic capacity of the models with a large group of expert and less experienced ultrasound examiners.

The results showed that the AI ​​models outperformed both experts and non-experts in detecting ovarian cancer, achieving an accuracy rate of 86.3 percent, compared to 82.6 percent and 77.7 percent for the experts and non-experts, respectively.

This suggests that neural network models can be a valuable aid in the diagnosis of ovarian cancer, especially in difficult-to-diagnose cases and in situations where there is a shortage of ultrasound experts.”

Professor Elisabeth Epstein, Department of Clinical Science and Training, Södersjukhuset (Stockholm South General Hospital), Karolinska Institutet

Reducing the need for expert recommendations

The AI ​​models can also reduce the need for expert recommendations. In a simulated triage situation, AI support reduced the number of referrals by 63 percent and the misdiagnosis rate by 18 percent. This can lead to faster and more cost-effective care for patients with ovarian lesions.

Despite the promising results, researchers emphasize that further studies are needed before the full potential of neural network models and their clinical limitations are fully understood.

"Through continuous research and development, AI-based tools can be an integral part of tomorrow's healthcare, freeing up experts and optimizing hospital resources. However, we need to ensure that they can be adapted to different clinical environments and patient groups," says Filip Christiansen, doctoral student in Professor Epstein's research group at Karolinska Institutet and joint first author with Emir Konuk at KTH Royal Institute of Technology.

Assessing the security of AI support

The researchers are currently conducting prospective clinical trials at Södersjukhuset to assess the everyday clinical safety and utility of the AI ​​tool. Future research will also include a randomized multicenter trial to examine the impact on patient management and healthcare costs.

The study was carried out in close collaboration with researchers at the KTH Royal Institute of Technology and was funded by grants from the Swedish Research Council, the Swedish Cancer Society, the Stockholm Regional Council, the Radiumhemmet Cancer Research Funds and the Wallenberg AI, Autonomous Systems and Software Program (WASP).

Elisabeth Epstein, Filip Christiansen and three co-authors have applied for a patent for methods of computer-aided diagnostics through the company Intelligyn. Elisabeth Epstein, Filip Christiansen and Kevin Smith, researchers at the KTH Royal Institute of Technology, also own shares in Intelligyn, for which Professor Epstein is a volunteer manager. A full list of conflicts of interest can be found in the paper.

Sources: