Key points are not available for this paper at this time.
Artificial intelligence (AI) is thought by some to be the most fundamental transformation in our lives since the industrial revolution With the rapid increase in patient-specific information and computing power, there has been tremendous interest in the medical physics community to deploy machine/deep learning (ML/DL) algorithms in a wide range of diagnostic and therapeutic radiological applications to automate laborious processes, improve workflow, and aid physicians in their pursuit to realize precision medicine. This includes but is not limited to applications in computer-aided detection, classification, and diagnosis in radiology and auto-contouring, treatment planning, response modelling (radiomics, radiogenomics), image-guidance, motion tracking, and quality assurance in radiation oncology. Despite this interest by medical physicists, ML/DL algorithms have been surrounded by misunderstandings about their strengths, weaknesses and best practices for training, validation, and testing that have limited their practical clinical implementation in day-today clinical and medical physics operations.
Naqa et al. (Fri,) studied this question.