Examining the role of AI in cancer imaging through the lens of clinical studies

Abstract Background The use of artificial intelligence (AI) in radiological applications is continuously increasing, making AI tools entrenched deeper and deeper into the medical workflow. Given that clinical studies, and more specifically randomized clinical trials offer a more reliable picture on the value proposition that AI brings to radiologists, the current paper is focused on the evidence supplied by such studies. Methods Search for relevant clinical studies and trials was conducted both using Pubmed/Medline for completed trials with results published over the past 7 years, and the clinicaltrials.gov website for ongoing or newly registered clinical studies. Results Use of AI shows 21%-53% reduction in screen-reading time, 20% advancement in reading confidence and 9%-20% increased performance in lesion characterizations. Differential diagnosis shows notable improvement compared to junior medical professionals and modest gains when compared to senior experts. Conclusions AI is a great tool to assist the medical profession, reducing time needed to express expert opinions and helping junior professionals to gain experience and confidence. Currently AI does not provide improvements over standard of care when used as a single tool, without human involvement.