Primary outcomes of the PI-CAI challenge published in The Lancet Oncology

On June 11th, the primary outcomes of the PI-CAI challenge were published in The Lancet Oncology in an article titled, "Artificial Intelligence and Radiologists in Prostate Cancer Detection on MRI (PI-CAI): An International, Paired, Non-inferiority, Confirmatory Study". An editorial has been presented by Dr. Martin Eklund.

๐—–๐—ผ๐—บ๐—บ๐˜‚๐—ป๐—ถ๐˜๐˜† + ๐—–๐—ผ๐—น๐—น๐—ฎ๐—ฏ๐—ผ๐—ฟ๐—ฎ๐˜๐—ถ๐—ผ๐—ป ๐Ÿ‘ฉโ€โš•๏ธ๐Ÿง‘โ€๐Ÿ’ป Over the past two years, PI-CAI served as a global community-wide initiative that was able to mobilize and engage 129 radiologists, AI developers and researchers in prostate oncology, towards the common goals of a single large-scale confirmatory study โ€”and one that was conducted under the oversight of 17 multidisciplinary scientific experts in this domain, with endorsements from the European Association of Urology, the European Society of Urogenital Radiology, the MICCAI Society, and the MIDL Foundation.

๐—ข๐—ฝ๐—ฒ๐—ป ๐—ฆ๐—ฐ๐—ถ๐—ฒ๐—ป๐—ฐ๐—ฒ โš›๏ธ Source code for statistical analysis, the trained AI system developed in this study, a subset of 1500 multi-center, multi-vendor cases from the training dataset (with the means to access the full training dataset of 9107 cases) and the means for independent researchers to benchmark their AI systems across the hidden tuning and testing cohorts of 1100 cases in a fully blinded, standardized manner have been publicly released โ€”to promote reproducibility, transparency and facilitate future research.

๐—–๐—น๐—ถ๐—ป๐—ถ๐—ฐ๐—ฎ๐—น ๐—™๐—ถ๐—ป๐—ฑ๐—ถ๐—ป๐—ด๐˜€ ๐Ÿฉป Level 2b evidence was provided, supporting that an AI system (developed in parts by five teams based in France, United States, China, Australia and Turkey), was statistically superior to a pool of 62 radiologists (45 centers, 20 countries) in an international reader study, and comparable to the standard of care in routine practice (where multidisciplinary patient history and peer consultation were available) at detecting MRI examinations with clinically significant prostate cancers. A clinical trial is now required to determine if such a system translates to improvements in workflow efficiency, healthcare equity and patient outcomes.

The study was sponsored by Amazon Web Services and the overall project has been funded, in parts, by the European Commission Horizon 2020: ProCAncer-I project and Health~Holland.

We congratulate Anindo Saha, Joeran Bosma, Jasper Twilt, Jurgen Fรผtterer, Maarten de Rooij, Henkjan Huisman, and the rest of the PI-CAI consortium for their work! For more, you can listen to the (Dutch) interview with Henkjan Huisman featured on the 'NOS Radio 1 Journaal'.

← Back to overview