Turkbey Baris, Huisman Henkjan, Fedorov Andriy, Macura Katarzyna J, Margolis Daniel J, Panebianco Valeria, Oto Aytekin, Schoots Ivo G, Siddiqui M Minhaj, Moore Caroline M, Rouvière Olivier, Bittencourt Leonardo K, Padhani Anwar R, Tempany Clare M, Haider Masoom A
From the Molecular Imaging Branch, National Cancer Institute, National Institutes of Health, 10 Center Dr, MSC 1182, Bldg 10, Room B3B85, Bethesda, MD 20892 (B.T.); Department of Radiology and Nuclear Medicine, Radboud University Medical Center, Nijmegen, the Netherlands (H.H.); Department of Radiology, Brigham and Women's Hospital, Harvard Medical School, Boston, Mass (A.F., C.M.T.); The Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, Md (K.J.M.); Department of Radiology, Weill Cornell Medicine/New York Presbyterian, New York, NY (D.J.M.); Department of Radiological Sciences, Oncology and Pathology, Sapienza University, Rome, Italy (V.P.); Department of Radiology, University of Chicago, Chicago, Ill (A.O.); Department of Radiology and Nuclear Medicine, Erasmus University Medical Center, Rotterdam, the Netherlands (I.G.S.); Department of Surgery, Division of Urology, University of Maryland School of Medicine, Baltimore, Md (M.M.S.); Division of Surgery Interventional Science, University College London, London, UK (C.M.M.); Department of Urology, University College London Hospitals NHS Foundation Trust, London, UK (C.M.M.); Department of Urinary and Vascular Imaging, Hospices Civils de Lyon, Hôpital Edouard Herriot, Lyon, France (O.R.); Faculté de Médecine Lyon Est, Université de Lyon, Université Lyon 1, Lyon, France (O.R.); Department of Radiology, University Hospitals Cleveland Medical Center/Case Western Reserve University School of Medicine, Cleveland, Ohio (L.K.B.); Paul Strickland Scanner Centre, Mount Vernon Hospital, Middlesex, UK (A.R.P.); Joint Department of Medical Imaging, Mount Sinai Hospital, Princess Margaret Hospital, University of Toronto, Toronto, Canada (M.A.H.); and Lunenfeld-Tanenbaum Research Institute, Toronto, Canada (M.A.H.).
Radiology. 2025 Apr;315(1):e240140. doi: 10.1148/radiol.240140.
This document defines the key considerations for developing and reporting an artificial intelligence (AI) interpretation model for the detection of clinically significant prostate cancer (PCa) at MRI in biopsy-naive men with a positive clinical screening status. Specific data and performance metric requirements and a checklist are provided for this use case. Data requirements emphasize the need for sufficient information to provide transparency and characterization of training and test data. The definition of a true-negative examination (which includes a minimum of 2-year follow-up), the need for image quality assessments, and nonimaging metadata requirements are provided. Performance metrics ranges are included, such as a cancer detection rate of 40%-70% for Prostate Imaging Reporting and Data System, or PI-RADS, 4 or higher lesions and demonstration of equivalent or better than human performance using receiver operating characteristic and precision-recall curves. The use of open datasets such as those used in the AI challenge model is encouraged. The study design should include conformity with the Checklist for Artificial Intelligence in Medical Imaging requirements. This article should be taken in the context of the current and evolving regulatory landscape. This review provides guidance based on subspeciality expertise in prostate MRI and will hopefully accelerate the clinical translation of AI in PCa detection.
本文档定义了为活检初诊且临床筛查呈阳性的男性在MRI检查中检测具有临床意义的前列腺癌（PCa）而开发和报告人工智能（AI）解释模型的关键考量因素。针对此用例提供了特定的数据和性能指标要求以及一份清单。数据要求强调需要足够的信息以确保训练和测试数据的透明度及特征描述。给出了真阴性检查的定义（包括至少2年的随访）、图像质量评估的必要性以及非成像元数据要求。还包括性能指标范围，例如前列腺影像报告和数据系统（PI-RADS）4级或更高等级病变的癌症检测率为40%-70%，以及使用接收者操作特征曲线和精确召回率曲线证明性能等同于或优于人类。鼓励使用如AI挑战模型中所使用的开放数据集。研究设计应符合医学影像人工智能清单要求。本文应结合当前不断演变的监管环境来理解。本综述基于前列腺MRI的亚专业知识提供指导，有望加速AI在PCa检测中的临床转化。
