AI Outperforms Radiologists in Detecting Prostate Cancer on MRI

Radiologists are experiencing an increased workload as more men at high risk of prostate cancer are now routinely undergoing prostate MRI scans. Diagnosing prostate cancer using MRI is complex and requires considerable expertise, which is challenging due to a shortage of experienced radiologists. Now, an international study has demonstrated that artificial intelligence (AI) can alleviate these challenges by identifying prostate cancer more frequently than radiologists and generating fewer false alarms.

In this first large-scale study coordinated by Radboud University Medical Center (Nijmegenm, the Netherlands), an international research team conducted a transparent evaluation comparing AI with radiologist assessments against clinical outcomes. The researchers organized a major competition between AI teams and radiologists, involving participants from various centers in the Netherlands and Norway, who provided over 10,000 MRI scans. Each patient's scans were reviewed to confirm the presence of prostate cancer, allowing groups worldwide to develop AI tools for image analysis. The top five AI submissions were integrated into a super-algorithm for assessing prostate MRI scans. These AI assessments were then compared with those made by a panel of radiologists on four hundred scans. The PI-CAI community study engaged over two hundred AI teams and 62 radiologists from 20 countries in this initiative. They compared AI findings not only with those of radiologists but also against a gold standard, tracking the outcomes of the men whose scans were analyzed over an average follow-up of five years.

This pioneering international PI-CAI study on AI in prostate diagnostics revealed that AI detected nearly 7% more significant prostate cancers than the radiologists did. Furthermore, AI identified half as many non-cancerous suspicious areas compared to radiologists, suggesting that the number of unnecessary biopsies could potentially be reduced by half with AI integration. If these findings are consistent in further studies, they could significantly aid radiologists and patients by reducing radiologists’ workload, improving diagnostic accuracy, and decreasing unnecessary prostate biopsies. The AI developed through this study still requires validation and is not yet available for clinical use. The project leaders note that societal trust in AI is limited, as some systems developed are subpar. They are now focused on creating a public and transparent testing platform to fairly evaluate AI systems and are developing a quality management system similar to that used in the aviation industry.

“If planes almost collide, a safety committee will look at how to improve the system so that it doesn't happen in the future,” said Henkjan Huisman, AI expert and project leader of the PI-CAI study. “I want the same for AI. I want to research and develop a system that learns from every mistake so that AI is monitored and can continue to improve. That way, we can build trust in AI for healthcare. Optimal, governed AI can help make healthcare better and more efficient.” The findings of the PI-CAI study were published in The Lancet Oncology on June 11, 2024.

Related Links:
Radboud University Medical Center