Artificial intelligence (AI) could help expand access to prostate cancer care and support future population-based screening programmes, according to SCARLET-1, one of the largest international studies ever conducted on AI for prostate cancer detection using MRI.
Published in Radiology, the study analysed 22,445 prostate MRI examinations from 21,252 patients across 46 cities in 22 countries. Led by researchers at Radboud university medical center (Radboudumc), the project investigated whether an AI system that performs well in specialist research centres can also be trusted across diverse hospitals, patient populations and healthcare systems worldwide.
The findings are encouraging. Overall, the AI system achieved diagnostic performance comparable to experienced radiologists in both clinical and screening settings. However, its reliability varied substantially between hospitals and imaging environments, highlighting the importance of local validation and ongoing performance monitoring before large-scale implementation.
Why AI could transform prostate cancer care
Prostate cancer is one of the most commonly diagnosed cancers in men worldwide. According to the Lancet Commission on Prostate Cancer, the number of new cases is expected to nearly double to around 2.9 million annually over the next two decades, increasing pressure on healthcare systems.
MRI has become a cornerstone of modern prostate cancer diagnosis, helping clinicians identify cancers that require treatment while reducing unnecessary biopsies. Yet interpreting prostate MRI scans requires highly specialised expertise, which is concentrated in relatively few centres. At the same time, radiology workforce shortages and growing demand for imaging services are creating challenges globally.
AI systems that automatically analyse prostate MRI scans could help address these pressures by supporting radiologists, increasing efficiency and improving access to high-quality diagnostic expertise. Potential benefits for patients include more consistent assessments, fewer unnecessary invasive procedures and earlier detection of clinically significant cancers.
AI achieves performance on par with radiologists
The SCARLET-1 researchers evaluated an AI system designed to detect clinically significant prostate cancer on MRI. After development and internal testing on 18,471 MRI examinations, the system was externally evaluated in three major international studies:
- PRIME, an international diagnostic trial involving 476 patients across 18 cities in 11 countries.
- STHLM3-MRI, a Swedish population-based screening study involving 1,108 participants.
- IP1-PROSTAGRAM, a UK screening study involving 390 participants.
Across all three cohorts, the AI system demonstrated a strong ability to distinguish patients with clinically significant prostate cancer from those without it. Its performance was comparable to that of experienced radiologists, both among men undergoing diagnostic evaluation and those participating in organised screening programmes.
These results suggest that AI could become a valuable tool in prostate cancer care. However, the researchers stress that the technology is not ready to replace radiologists. Further prospective studies are needed to determine how AI can best be integrated into clinical pathways and whether it can improve patient outcomes while reducing unnecessary investigations.
AI performance depends heavily on clinical context
While the overall results were positive, SCARLET-1 also revealed important differences in performance between healthcare settings.
Additional evaluation across hospitals in 25 cities showed that the AI system performed exceptionally well in 32% of hospitals, moderately well in 40%, and less accurately in the remaining 28%. These variations were associated with factors such as MRI image quality and patient characteristics.
The findings demonstrate that strong performance in a limited number of centres does not guarantee similarly reliable results elsewhere. According to the researchers, local validation and ongoing performance monitoring are therefore essential whenever AI systems are introduced into clinical practice.
The study also highlights considerations for healthcare providers, regulators and policymakers. Decisions on AI implementation should take into account not only the accuracy reported in validation studies, but also local patient populations, imaging quality, intended clinical use and mechanisms for continuous performance assessment.
Towards more equitable AI
Encouragingly, the researchers found no meaningful differences in AI performance across age or ethnic groups within the available datasets, suggesting that such systems may be applicable across increasingly diverse populations.
However, representation from the Global South remained limited. The authors emphasise that further international collaboration and additional validation studies in historically underrepresented regions will be essential to ensure that the benefits of medical AI can be delivered equitably across global healthcare systems.
Building a global foundation for trustworthy medical AI
SCARLET-1 was conducted through a collaboration of the PI-CAI, PRIME, STHLM3-MRI, IP1-PROSTAGRAM, ProCAncer-I and COMFORT research consortia, bringing together expertise in radiology, urology, oncology, medical imaging and artificial intelligence from across multiple continents.
Beyond the study itself, the collaboration has established a large international resource for evaluating AI systems in diverse clinical environments. The researchers intend to expand these efforts further through additional participating institutions, improved representation of underserved populations and evaluation of newer generations of medical AI.
First author Anindo Saha said: 'The standard of care in prostate cancer management has continuously evolved alongside new guidelines, imaging technologies and workforce changes. Prostate AI systems will likely require the same approach, with structured evaluation, monitoring and periodic refinement to ensure they remain aligned with contemporary clinical practice.'
The researchers believe SCARLET-1 represents an important step towards the safe and responsible implementation of AI in healthcare. While the study demonstrates the potential of AI to expand access to prostate cancer diagnosis and screening, it also provides a clear reminder that rigorous evaluation remains essential to ensure these technologies perform reliably across the diverse healthcare settings in which they are ultimately deployed.
In this photo: first author Anindo Saha.
About this publication
This study was published in Radiology: Artificial Intelligence for Prostate Cancer Detection and Screening on MRI in Global Populations: The SCARLET-1 Study. Anindo Saha, Joeran S. Bosma, Jasper J. Twilt,[...] Veeru Kasivisvanathan, Maarten de Rooij and Henkjan Huisman, on behalf of the PI-CAI–PRIME–STHLM3-MRI–IP1-PROSTAGRAM–ProCAncer-I–COMFORT consortia. DOI: https://doi.org/10.1148/radiol.262081.
-
Want to know more about these subjects? Click on the buttons below for more news.
More information
Pauline Dekhuijzen
wetenschaps- en persvoorlichter
