Validation of artificial intelligence (AI) based software as medical device (SaMD) for retinopathy of prematurity (ROP)

验证基于人工智能 (AI) 的软件作为治疗早产儿视网膜病变 (ROP) 的医疗设备 (SaMD)

• 批准号: 10760401
• 负责人: John Peter Campbell
• 金额: $ 190.71万
• 依托单位: SILOAM VISION, INC.
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-30 至 2025-08-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10760401
• 关键词: Algorithms; Artificial Intelligence; Award; Blindness; Blood Vessels; Caring; Case Study; Child; Childhood; Classification; Clinical; Clinical Research; Code; Collaborations; Computer software; Country; Custom; Data; Data Management Resources; Devices; Diagnosis; Diagnostic; Disease; Dropout; Early Diagnosis; Early treatment; Ensure; Eye; Funding; Fundus photography; Goals; Health; Image; India; Infrastructure; International; Investments; Manufacturer; Marketing; Medical; Medical Device; Medical Imaging; Mission; Oregon; Output; Paper; Performance; Phase; Privatization; Process; Provider; Reader; Reference Standards; Research; Resources; Retinopathy of Prematurity; Science; Services; Severities; Small Business Innovation Research Grant; Specificity; Speed; Standardization; System; Technology; Telemedicine; Testing; Training; United States Food and Drug Administration; Universities; Update; Validation; Vision; Work; accurate diagnosis; artificial intelligence algorithm; blind; clinical diagnosis; cloud based; design; detection sensitivity; diagnosis standard; diagnostic tool; digital; disease classification; disease diagnosis; expedited review; fundus imaging; image processing; improved; improved outcome; low and middle-income countries; medically underserved; meetings; neonatal care; prevent; primary outcome; retinal imaging; screening; secondary outcome; segmentation algorithm; software development; transfer learning; underserved area

The purpose of this application is to perform the necessary clinical studies to seek regulatory approval for an artificial intelligence (AI) software as medical device (SaMD) for retinopathy of prematurity (ROP) diagnosis. ROP is a leading cause of childhood blindness worldwide, with approximately 50,000 babies going blind annually, most of which is preventable with accurate and timely diagnosis. The i-ROP DL algorithm was developed by the i-ROP research consortium and has been shown to provide expert-level diagnosis of plus disease, a component of severe ROP, based on images from the Retcam (Natus, Middleton, WI) digital fundus camera. The output is a vascular severity score (VSS) that corresponds to spectrum of plus disease, as defined by the International Classification of ROP, and has been endorsed by the Food & Drug Administration (FDA) as an appropriate output for an ROP SaMD. If incorporated into a clinical workflow, this technology could provide automated, immediate, expert-level diagnosis of ROP to the bedside, solving one of the key gaps in care that results in preventable blindness worldwide. The first aim of this project is to update and retrain the i-ROP DL algorithm to improve speed and repeatability for clinical use, finalize the image quality and pre-processing pipeline, and integrate it into the iTeleGEN data management system, an ROP telemedicine software platform. The second aim is to perform the necessary clinical studies for the two proposed indications for use (IFU): The first IFU will be as an assistive diagnostic study to improve the clinical diagnosis of plus disease with regulatory approval based on a multi-reader multi-case study with a primary outcome of improved diagnosis of plus disease, based on a five expert reference standard diagnosis, with the use of the VSS. The second IFU will be for autonomous ROP screening for more than mild ROP (MTMROP, defined as type 2 or worse according to the Early Treatment for ROP study definition). The pivotal study will have a primary outcome of 85% sensitivity and 85% specificity for the diagnosis of MTMROP, with a secondary outcome of greater than 95% sensitivity for detection of treatment-requiring ROP. The third aim of the proposal is to validate the i-ROP DL algorithm on a digital fundus camera made by Forus Health (Bengaluru, India), a digital eye care company, with ROP camera distribution in more than 20 countries. If successful, then once FDA approval is obtained on the Retcam it may be extended through a 510K process to a camera that is more affordable than the Retcam and widely available in low- and middle-income countries. This work will be done by Siloam Vision, a company started by two of the inventors of the i-ROP DL algorithm, in conjunction with Oregon Health & Science University. At the end of the study period, the goal will be to have the necessary data to support FDA approval of the i-ROP DL algorithm for two IFUs on two digital fundus cameras and being one step closer to bringing this technology to the bedside to reduce the number of babies going blind from ROP worldwide.

本申请的目的是进行必要的临床研究，以寻求监管部门对 人工智能（AI）软件作为医疗器械（SaMD）用于早产儿视网膜病变（ROP）诊断。 ROP是全球儿童失明的主要原因，约有50，000名婴儿失明 每年，其中大多数是可以预防的准确和及时的诊断。i-ROP DL算法是 由i-ROP研究联盟开发，并已被证明可提供专家级的诊断， 根据Retcam（Natus，Middleton，WI）数字眼底图像， 相机输出是血管严重程度评分（VSS），对应于定义的正疾病谱 国际ROP分类，并已被美国食品和药物管理局（FDA）认可为 ROP SaMD的适当输出。如果纳入临床工作流程，这项技术可以提供 将ROP的自动化、即时、专家级诊断带到床边，解决了护理中的一个关键空白， 导致全世界可预防的失明。本项目的第一个目标是更新和再培训i-ROP DL 算法，以提高临床使用的速度和可重复性，最终确定图像质量和预处理 管道，并将其集成到iTeleGEN数据管理系统，ROP远程医疗软件平台。 第二个目的是对两种拟定适应症（IFU）进行必要的临床研究： 第一份IFU将作为辅助诊断研究，以改善监管疾病的临床诊断 批准基于一项主要结局为改善plus疾病诊断的多位阅片人多病例研究， 基于五个专家参考标准诊断，使用VSS。第二份IFU将用于 超过轻度ROP的自主ROP筛查（MTMROP，定义为2型或更差，根据 ROP研究定义的早期治疗）。这项关键研究的主要结局将具有85%的敏感性， MTMROP诊断的特异性为85%，次要结局的敏感性大于95%， 检测需要治疗的ROP。该提案的第三个目的是在一个虚拟机上验证i-ROP DL算法。 数字眼底照相机，由Forus Health（印度Bengalu），一家数字眼保健公司制造，带有ROP照相机 分布在20多个国家。如果成功，那么一旦Retcam获得FDA批准， 通过510 K工艺扩展到比Retcam更实惠且广泛可用的相机 在低收入和中等收入国家。这项工作将由Siloam Vision公司完成，该公司由两名 i-ROP DL算法的发明者，与俄勒冈州健康与科学大学合作。结束时 在研究期间，目标是获得必要的数据来支持FDA批准i-ROP DL算法， 两个数字眼底相机上的两个IFU，并且更接近于将这项技术带到床边， 减少全世界因ROP而失明的婴儿数量。