Artificial intelligence is better at detecting prostate cancer on MRI than radiologists

Artificial intelligence (AI) is more likely to detect prostate cancer than radiologists. In addition, AI is half as likely to cause false alarms. This is shown by an international study coordinated by Radboud University Medical Center and published in The Lancet Oncology. This is the first large-scale study where an international team transparently evaluates and compares AI with radiologists' assessments and clinical outcomes.

Radiologists are facing an increasing workload as men at increased risk of prostate cancer now routinely undergo prostate MRIs. Diagnosing prostate cancer using MRI requires significant expertise, and experienced radiologists are in short supply. AI can help solve these problems.

Artificial intelligence expert Henkjan Hausman and radiologist Maarten de Rooy, leaders of the PI-CAI project, organized a major competition between AI teams and radiologists with an international team participating. Together with other centers in the Netherlands and Norway, they provided more than 10,000 MRI scans. They transparently identified the presence of prostate cancer for each patient. Various groups around the world have been allowed to develop AI to analyze these images.

The top five applications were combined into a super-algorithm to analyze MRI scans for the presence of prostate cancer. Finally, the AI's assessments were compared with those of a group of radiologists on four hundred prostate MRI scans.

Accurate diagnosis The PI-CAI community brings together more than two hundred AI teams and 62 radiologists from twenty countries. They compared the results of the AI and radiologists not only with each other, but also with the gold standard, tracking the results of men who had scans done. On average, the men were followed for five years.

This first international study on the use of AI in diagnosing prostate cancer shows that AI detects almost 7% more significant prostate cancers than a group of radiologists. In addition, AI identifies suspicious areas that later turn out to be non-cancerous in 50% fewer cases. This means the number of biopsies can be halved using AI.

If these results are confirmed in subsequent studies, it could greatly help radiologists and patients in the future. This can reduce radiologists' workload, provide more accurate diagnoses, and minimize unnecessary prostate biopsies. The developed AI still needs to be validated and is currently not yet available to patients in a clinical setting.

Quality System Houseman notes that society does not trust AI. "This happens because manufacturers sometimes create AIs that are not good enough," he explains. He's working on two things. The first is a public and transparent test to fairly evaluate AI. The second is a quality management system similar to that in the aviation industry.

"If planes almost collide, the safety committee will research how to improve the system to prevent this from happening in the future. I want the same for AI. I want to research and develop a system that learns from every mistake so that the AI is controlled and could continue to improve. In this way, we can build trust in AI in healthcare. Optimal, controlled AI can help make healthcare better and more efficient."