Artificial Intelligence Education in Radiology: How We Can Better Prepare the Workforce

Purpose: Artificial intelligence (AI) has attracted increasing attention. Graduate medical education teaching of AI is inconsistent. Successful algorithm integration relies on simplicity of integration and well-versed AI users. Challenges include poor performance, lack of user technical knowledge and skepticism regarding quality and safety. AI education in radiology residencies may include simple or complex concepts such as pixel math or neural networks. We explored how curricula can be modified to improve resident trust and confidence in AI.

Methods/Materials: We surveyed residents understanding of artificial intelligence including applications, terminology, and comfort with use and implementation. Factors assessed included residency size, and well-versed AI colleagues. The questionnaire was distributed to residency program administrators and leadership for forwarding to residents.

Results: We received 45 responses, most (75.6%) from residencies with 7 or greater residents per class. 75.6% recognized algorithms as devices by the FDA. Only 15.6% of responders recognized there are over 150 FDA approved algorithms. 86.7-88.9% of residents were unable to differentiate between high and low risk FDA approval pathways. Most (68.9%) residents were unaware of common AI terminology such as CADe, CADt, CADx, image phenotype; although they were aware of applications such as radiogenomics and work-list prioritization. Comfort with AI’s diagnostic applications was rated 4.3 out of 10 (SD 1.7), comfort explaining algorithm training and validation was rated 3.9 out of 10 (SD 2.6). If there was a well-versed attending or resident within the program, residents had higher confidence levels explaining algorithm training and validation, although it didn’t increase resident trust levels.

Conclusions: Residents were unaware of how many applications of AI algorithms in radiology have been approved, how algorithms are trained and validated, of nuances related to FDA approval, and of common AI terminology. When there was a well-versed colleague, residents were more confident explaining these concepts. Incorporating education regarding training, validation and the FDA approval process should increase resident confidence in AI algorithms.