Artificial intelligence for scoring prostate MRI: ready for prospective evaluation
Menée auprès de 62 radiologues de 20 pays et à partir de 10 207 IRM réalisées sur 9 129 patients atteints d'un cancer de la prostate, cette étude évalue la non infériorité de systèmes d'intelligence artificielle par rapport aux radiologues pour détecter un cancer de la prostate à l'aide d'images IRM
Artificial intelligence (AI) is likely to transform medicine, with radiology being one of the first medical fields in which AI can broadly impact clinical practice. Prebiopsy MRI of the prostate has been shown to improve prostate cancer diagnosis. 1 , 2 However, the use of MRI is impeded by its dependence on experienced radiologists and the high inter-radiologist variability in Prostate Imaging Reporting and Data System (PI-RADS) scoring of prostate MRI exams. 3 , 4 Accurate and robust AI algorithms for reading prostate MRI could alleviate these challenges. In The Lancet Oncology, Anindo Saha and colleagues 5 present results from the Prostate Imaging—Cancer Artificial Intelligence (PI-CAI) challenge for developing AI algorithms for detecting clinically significant prostate cancers (defined as Gleason grade group ≥2; hereafter referred to as International Society of Urological Pathology [ISUP] grade) in prostate MRI scans and compare the diagnostic performance of the algorithms with that of radiologists. Saha and colleagues 5 concluded that ensemble results from top-performing AI systems were superior at detecting clinically significant prostate cancer when compared with the average results of 62 radiologists (area under the receiver operating characteristic curve 0·91 [95% CI 0·87–0·94] vs 0·86 [0·83–0·89]; p<0·0001), and had similar performance to radiologists' reads made during clinical routine.
The Lancet Oncology , commentaire, 2023