E1505. An Artificial Intelligence Training Workshop for Diagnostic Radiology Residents
  1. Ricky Hu; Queen's Univeristy; School of Medicine, Queen's University
  2. Arsalan Rizwan; Queen's Univeristy
  3. Zoe Hu; Queen's Univeristy
  4. Anthony Li; School of Medicine, Queen's University
  5. Andrew Chung; Queen's Univeristy
  6. Benjamin Kwan; Queen's Univeristy
To develop, implement, and evaluate an introductory artificial intelligence (AI) training workshop for radiology residents.

Materials and Methods:
A 3-week AI training curriculum was developed and integrated within curricular time during academic half days of a Competency-Based Medical Education (CBME) radiology training program. The curriculum was developed by radiology faculty, residents, and AI engineers and consists of content aimed to develop foundational literacy of AI concepts. Concepts included defining an AI task, data preprocessing, popular machine learning models, validation methods, assessing bias, and modern techniques such as convolutional neural networks. Content was delivered by lectures, review of literature for case studies, and demonstration of AI processing with programming examples provided to participants to execute on the Google CoLaboratory environment. To evaluate the efficacy of the workshop, identical pre- and post-workshop surveys were provided for participants. The survey consisted of 18 questions gauging their confidence in AI concepts and 3 questions gauging perception of AI training, using a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Survey results were analyzed with first order statistics and the pre- and post-workshop distributions were compared with two-tailed t-tests: p < 0.05 was considered significant.

Twelve radiology residents participated in the workshop and n = 11 participants completed the survey. When asked if the workshop improved AI knowledge, an average score of 4.0 (standard deviation: 0.7) was recorded. There was a significant increase (p < 0.05) in 16/18 questions gauging confidence in AI concepts. There were no (p > 0.05) significant differences in perception of AI training. There was high baseline enthusiasm on the pre-workshop survey, with an average score of 3.8 regarding interest in continuing AI education and an average score of 4.1 regarding perceived importance of AI.

As AI applications are increasingly introduced into radiology, training to understand AI is required to assess the applicability of an AI tool. However, AI training for a medical audience presents unique challenges related to selecting scope of content and delivery methods. To address this, we developed an AI workshop to provide concise foundational AI concepts to radiology residents. The workshop resulted in positive feedback and improved comprehension for participants. This work supports future institutional integration of AI training to radiology trainees, particularly suitable for flexible programs such as a CBME curriculum.