AI for Doctors—A Course to Educate Medical Professionals in Artificial Intelligence for Medical Imaging

Dennis M. Hedderich, Matthias Keicher, Benedikt Wiestler, Martin Gruber, Hendrik Burwinkel, Florian Hinterwimmer, Tobias Czempiel, Judith Spiro, Daniel Pinto dos Santos, Dominik Heim, Claus Zimmer, Daniel Rückert, Jan S. Kirschke, Nassir NavabHealthcare 2021 · DOI

TL;DR

A 12-week online course on AI in medical imaging was developed to address the scarcity of tailored educational materials for medical professionals. The course successfully improved participants' self-assessed AI skills and led to a more realistic, less overoptimistic view of AI's patient benefits, though high dropout rates highlight the challenge of time constraints for busy clinicians.

What Problem Does This Paper Address?

The successful integration of Artificial Intelligence (AI) into medical imaging hinges on medical professionals understanding its fundamental principles and practical applications. However, there's a significant gap in educational resources specifically designed for this audience. Existing AI training often targets computer scientists or engineers, leaving clinicians without the necessary knowledge to critically evaluate, validate, and ethically deploy AI tools in their practice. This lack of understanding can lead to overoptimism about AI's capabilities or hesitance to adopt it, hindering its clinical translation. This paper addresses this by developing and evaluating a novel educational program aimed at bridging this knowledge gap for medical professionals.

Methods

The authors designed and implemented a 12-week, online-only course titled "AI for Doctors: Medical Imaging." The curriculum was structured into two blocks:

  • Block 1 (Weeks 1-6): Focused on foundational AI concepts, including machine learning basics, artificial neural networks, linear algebra, and special considerations for medical imaging. This block utilized a blended learning approach with live online lectures, pre-recorded screencasts, readings, and multiple-choice quizzes.
  • Block 2 (Weeks 7-12): Delved into special-focus topics of applied AI in medical imaging, such as explainable AI, computational pathology, AI in dermatology, and ethical considerations. This block employed a flipped-classroom model with pre-recorded lectures, readings, and live Q&A sessions.

Participants' opinions on AI and their self-assessed skills were measured using pre- and post-course questionnaires employing a five-point Likert scale. Statistical analysis was performed using Wilcoxon's signed-rank test to compare pre- and post-course evaluations.

Datasets

Not discussed in the paper. The study focuses on the educational course and participant feedback, not on training or evaluating AI models on specific datasets.

Key Insights and Ideas

  • Demand for AI Education is High: The overwhelming interest from medical professionals, particularly doctors and medical students, underscores a significant unmet need for AI education in healthcare.
  • Education Moderates Optimism: While participants were initially very optimistic about AI's potential, deeper knowledge gained through the course led to a more nuanced and less overoptimistic view, particularly regarding patient benefits. This suggests education can foster a more realistic appreciation of AI's capabilities and limitations.
  • Skill Improvement is Tangible: Participants reported significant improvements in their self-assessed AI-related skills, including understanding Python code, linear algebra concepts for ML, and critically appraising AI research papers.
  • Time is the Major Hurdle: The substantial dropout rate, attributed primarily to the demanding schedules of clinicians and medical students, highlights a critical challenge in delivering effective AI education to this demographic.
  • Bridging the Gap is Feasible: Despite challenges, the study demonstrates that educating medical professionals on AI is achievable and can positively influence their attitudes and competencies, paving the way for better AI adoption.

Key Results

  • Participant Demographics: 93 participants enrolled, comprising medical doctors (43%), medical students (37.6%), PhD students (7.5%), and non-MD researchers (11.8%).
  • Opinion Shift: A statistically significant decrease in overoptimism was observed regarding the statement "Using AI in medical imaging will benefit patients in the foreseeable future" (p = 0.020), shifting from "complete agreement" to "rather agree."
  • Skill Enhancement: Self-assessed skills showed significant improvement across all measured domains after the course. For example, the median score for "Understanding Python code when reading it" increased from 1 to 2.5 (p = 0.001), and for "Creating Python code for statistical analysis," it rose from 1 to 2 (p = 0.002).
  • Course Appreciation: Participants rated the course highly, with the "well organized" and "study material well prepared" questions receiving median scores of 5.
  • Dropout Rate: A substantial dropout rate was observed, with only 28 participants completing the course.

Limitations and Future Work

  • Subjectivity of Self-Assessment: The authors acknowledge that self-assessed skill ratings are subjective and suggest that future educational programs should define clear, testable learning objectives and incorporate objective skill assessments.
  • Time Constraints: The high dropout rate due to time limitations is a significant challenge. Future efforts should explore strategies to mitigate this, such as integrating the course into mandatory curricula, offering it for credit, or potentially introducing a participation fee to increase adherence.
  • Online-Only Format: The online-only nature of the course, necessitated by the COVID-19 pandemic, led to participants missing in-person interactions. The authors hypothesize that in-person meetings could improve engagement and reduce dropout rates.
  • Future Directions: Future research should focus on harmonizing AI curricula for medical schools and residency programs, defining clear learning objectives, and investigating optimal methods for teaching computer science and AI concepts to medical professionals. Exploring ways to make AI education more accessible and less time-intensive for busy clinicians remains a key area for future work.

Summary updated

Abstract

Successful adoption of artificial intelligence (AI) in medical imaging requires medical professionals to understand underlying principles and techniques. However, educational offerings tailored to the need of medical professionals are scarce. To fill this gap, we created the course "AI for Doctors: Medical Imaging". An analysis of participants' opinions on AI and self-perceived skills rated on a five-point Likert scale was conducted before and after the course. The participants' attitude towards AI in medical imaging was very optimistic before and after the course. However, deeper knowledge of AI and the process for validating and deploying it resulted in significantly less overoptimism with respect to perceivable patient benefits through AI (<i>p</i> = 0.020). Self-assessed skill ratings significantly improved after the course, and the appreciation of the course content was very positive. However, we observed a substantial drop-out rate, mostly attributed to the lack of time of medical professionals. There is a high demand for educational offerings regarding AI in medical imaging among medical professionals, and better education may lead to a more realistic appreciation of clinical adoption. However, time constraints imposed by a busy clinical schedule need to be taken into account for successful education of medical professionals.
BibTeX
@inproceedings{hedderich2021ai,
  author = {Dennis M. Hedderich and Matthias Keicher and Benedikt Wiestler and Martin Gruber and Hendrik Burwinkel and Florian Hinterwimmer and Tobias Czempiel and Judith Spiro and Daniel Pinto dos Santos and Dominik Heim and Claus Zimmer and Daniel Rückert and Jan S. Kirschke and Nassir Navab},
  title = {{AI for Doctors—A Course to Educate Medical Professionals in Artificial Intelligence for Medical Imaging}},
  booktitle = {Healthcare},
  year = {2021},
  doi = {10.3390/healthcare9101278},
}
Figure