ABSTRACT
Designing an artificial intelligence (AI) system presenting excellent performances, maybe even on par with human experts, may appear as a daunting task, and indeed requires significant efforts. This chapter introduces the concept of medical devices constituted by standalone software. It describes relevant concepts for the validation of AI/machine learning (ML) applications in the medical field. The chapter provides emerging approaches to the validation of AI/ML software devices that evolve continuously in time, and thus are challenging for most regulatory frameworks where changes to a device require regulatory approval prior to public use. It addresses the important topic of how to define the ultimate clinical utility of a device, and discusses how randomized trials and other study designs can be used to evaluate it. One of the most important steps in the ML process is the definition of the datasets employed for the training and the testing of the system.
