An international study led by the Radboud University Medical Center has demonstrated that artificial intelligence (AI) can detect prostate cancer on MRI scans more effectively than radiologists, while also reducing false positives by half.
The findings, published in The Lancet Oncology, mark a significant advancement in the use of AI in medical diagnostics, particularly for prostate cancer.
The increasing routine use of MRI for men at higher risk of prostate cancer has placed a substantial burden on radiologists, who require sufficient expertise to accurately diagnose the disease. The shortage of experienced radiologists has further compounded this challenge. AI offers a promising solution to these issues by assisting in the diagnostic process.
The PI-CAI study, coordinated by AI expert Henkjan Huisman and radiologist Maarten de Rooij, involved a large-scale competition between AI teams and radiologists. The study brought together an international team that included contributions from centers in the Netherlands and Norway.
Over 10,000 MRI scans were used to develop and test AI algorithms. The top five AI submissions were combined into a super-algorithm, which was then compared to the assessments of radiologists using 400 prostate MRI scans.
This study is the first of its kind to transparently evaluate and compare the performance of AI against radiologist assessments and clinical outcomes on such a large scale. The research involved more than 200 AI teams and 62 radiologists from 20 countries. The accuracy of AI and radiologists was measured against a gold standard, with patient outcomes monitored over an average of five years.
The results were striking: AI detected nearly seven percent more significant prostate cancers than the radiologists. Moreover, AI triggered false alarms 50 percent less often, potentially reducing the number of unnecessary biopsies by half. These findings suggest that AI could significantly alleviate the workload of radiologists, improve diagnostic accuracy, and minimize unnecessary procedures.
Huisman emphasized the importance of building trust in AI within the medical community. “This is because manufacturers sometimes build AI that isn’t good enough,” he explained. To address this, Huisman is working on creating a public and transparent test for AI, as well as a quality management system similar to those 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. 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.”
The AI developed in this study is not yet available for clinical use and still requires further validation. However, the researchers believe that AI has the potential to greatly enhance prostate cancer diagnostics in the future, by reducing the workload on radiologists and providing more accurate diagnoses.
