Fitness for Purpose of Text-to-Image Generative Artificial Intelligence Image Creation in Medical Imaging.

Abstract

The recent emergence of text-to-image generative artificial intelligence (AI) diffusion models such as DALL-E, Firefly, Stable Diffusion, and Midjourney has been touted with popular hype about the transformative potential in health care. This hype-driven, rapid assimilation comes with few professional guidelines and without regulatory oversight. Despite documented limitations, text-to-image generative AI creations have permeated nuclear medicine and medical imaging. Given the representation of medical imaging professions and potential dangers in misrepresentation and errors from both a reputation and community harm perspective, critical quality assurance of text-to-image generative AI creations is required. Here, tools for evaluating the quality and fitness for purpose of generative AI images in nuclear medicine and imaging are discussed. Generative AI text-to-image creation suffers quality limitations that are generally prohibitive of mainstream use in nuclear medicine and medical imaging. Text-to-image generative AI diffusion models should be used within a framework of critical quality assurance for quality and accuracy.