Mandating Limits on Workload, Duty, and Speed in Radiology

The Regulation of Artificial Intelligence in Digital Radiology in the Scientific Literature: A Narrative Review of Reviews

“…The search on Pubmed returned 16 papers [ 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 ] (Scopus returned 4 further works, which, being works at congresses in accordance with the qualification procedure, were excluded), all studies from 2018 to today.…”

“…The analysis showed that the scientific literature has produced important and far-reaching reviews. Twelve reviews [ 21 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 33 , 34 , 35 , 36 ] out of fourteen were focused on radiology in general, two reviews were more specific, one review was focused on nuclear medicine [ 24 ], and one on imaging in gastrointestinal pathology [ 22 ]. As far as the regulatory aspects were concerned, some interesting issues emerge.…”

“…Two other studies offered new and important points of view on the regulatory aspects in this area; one focused on AI in DR, stressing the concept of software as medical device (SaMD) [ 27 ]; the other one dealt with the importance of providing certified datasets in training AI systems [ 31 ]. The three most recent reviews focused on some heterogeneous problems of AI regulation [ 34 , 35 , 36 ]. The review proposed in [ 34 ] faced the problems of regulating the workloads and the relevant implications.…”

“…The review proposed in [ 34 ] faced the problems of regulating the workloads and the relevant implications. In addition, the hurdles of the legal issues to AI autonomy have been faced in [ 35 ], where the ambiguity of the legal management and the impact on all the actors were stressed. The imaging-based algorithms validated by the FDA were analyzed in [ 36 ] with the focus also on the emerging problems.…”

“…The last three reviews published on these issues have focused specifically on bottlenecks in regulatory aspects [ 34 , 35 , 36 ] and represent mutually complementary and overall exhaustive contributions. Alexander et al [ 34 ] focused on the impact of workload on the decision correctness.…”