Clinical trials (CT) of artificial intelligence (AI) powered software are the youngest and methodologically immature field. Thus, there is a critical need to systematize global experience and to create a unified regulatory framework for stages of CT of AI powered software in radiology. Considering available regulatory requirements, national standards, principles of Good Clinical Practice, literature data, global and own experience, we have developed the rules for such trials. It is a structured description of full-cycle CT from request for CT, trial design, dataset creation, analytical and clinical validation to preparation and archiving of final documentation. These rules were validated using routine imaging data. The paper raises the issue of scaling to the regions of the Russian Federation and considers the possibility of potential application of this methodologically «transparent» and comprehensible framework at the industry level. Implementing the unified rules will ensure high quality of trials and facilitate market adoption of safe and effective AI products for routine clinical practice.
Tretyakova et al. (Tue,) studied this question.