Artificial intelligence education in medical schools
Developments of artificial intelligence (AI) in healthcare accelerated in the past few years. Already today, a number of medical fields rely on the help of AI-programs in their work [1, 2], including radiology [3], cardiology [4], or drug design [5], albeit mostly in laboratory or experimental environments. While AI and its currently most important technology, machine learning, penetrates more and more medical subspecialties, it is most prevalent in medical imaging [6,7,8,9,10]. There is a clear positive trend in the funding of AI research in the healthcare sector. This applies both to the private and venture capital sector, which has seen a steady increase in funding over the last few quarters [11, 12], as well as to the public sector. The governments of several countries are promoting the research and implementation of AI in healthcare, such as the USA with the Department of Health & Human Services’ “Artificial Intelligence Strategy” [13], Germany as part of their “National Strategy for Artificial Intelligence” [14], and China through its National Natural Science Foundation of China [15]. It can be assumed that these research results of these initiatives will reach clinical practice in the not too distant future.
However, the current medical school curricula very rarely cover aspects of medical informatics, let alone AI, as part of their mandatory study programs [16,17,18]. While there are some attempts to teach AI to (future) health care providers [19, 20] and to foster “AI literacy”, a term coined by Duri Long and Brian Magerko [21], the educational landscape is very heterogeneous. In addition to this, the medical curriculum in Germany and in other countries is strongly regulated by the government, listing in detail which topics must be taught during the 6 years of studying medicine (National Competence Based Learning Objectives Catalog, NKLM, www.nklm.de/zend). These relatively strict guidelines lead to little room to integrate courses about AI in regular teaching. This means that only a small fraction of medical students gets the opportunity to further their education in this important topic, although past research has already demonstrated that medical students would like to see more elaborate courses on digital health and AI [22, 23].
Apart from the fact that AI in medicine is an exciting and fast-moving research field in which future physicians and physician-scientists can conduct interesting experiments, knowing how it works and being able to interpret its results is also a prerequisite for their future jobs. Furthermore, AI will permeate much of the daily clinical work as a supporting entity [24], which will have an enormous impact on skills necessary for future practice [25]. However, medical students today do not feel they understand the subject matter particularly well or receive sufficient teaching on the topic [26].
It is important to note that promoting AI skills in medical education and increasing medical students AI readiness does not necessarily mean that all students must learn to program or understand the mathematical models underlying AI. Rather, training in AI for students from diverse study backgrounds who are not majoring in technical subjects should be about understanding how different AI applications work and what opportunities or threats come with the usage of these new technologies. As Long et al. nicely put it, the goal is to acquire a “casual understanding of AI – i.e., understanding how a search engine works, not necessarily understanding how to program one” [27].
Creation of a flipped classroom AI course for medical students
To address the aforementioned issue, the Institute of Medical Education at Bonn Medical School, together with experts from the department of radiology, ophthalmology, and neuroradiology, created the course “KI-LAURA” (the German abbreviation stands for “Künstliche Intelligenz in der Lehre der Augenheilkunde und Radiologie” and translates to “Artificial intelligence in the teaching of ophthalmology and radiology”).
KI-LAURA was a seven-month project which was subdivided into two sections. The first part dealt with the production of online self-study content, which was subsequently uploaded to the publicly accessible MOOC platform “AI-Campus” (or “KI-Campus” in German, www.ki-campus.org). In the second section of the project, a flipped classroom course was created at Bonn campus, which consisted of the on-demand online content on the one hand and interactive classroom sessions (held online due to Covid-19 restrictions) on the other.
In addition to explanations on how AI is (and will be) supporting the diagnostic process using medical image data, possible opportunities and risks of AI application and the future of physicians in the context of AI were discussed.
The flipped classroom incorporated the online content as one component, supplemented it with more in-depth explanations, and tutored exercises (see the supplementary material for a detailed description of the course curriculum). A description of the radiology classroom session, which was held by two experts in the field of AI in radiology, can be used as an example for the interactive classroom sessions: At the beginning of the lesson, a short input on AI and medical imaging was given by a clinical expert. This was followed by a diagnostic exercise using a web-based DICOM viewer (see get.pacsbin.com), which allowed students to view medical imaging data (e.g., CT-data sets) of real, anonymized patient data on their mobile devices or PCs. Finally, an AI researcher explained how AI-algorithms are able to analyze images, and presented studies in which AI-applications are already on the level of a human radiologist in terms of diagnostic accuracy.
Advantages of using the flipped classroom method for AI education in medical schools
Using the flipped classroom method to develop students’ AI competencies has several advantages, which are even more important in the context of medical education. First, knowledge transfer components are separated in time and space from in-depth practical teaching [28], which increases perceived learning flexibility [29] and student satisfaction [30]. Second, due to the low implementation effort, other medical schools can easily adopt the course structure, which means that the course is scalable with little to no additional resource requirements. Third, the increased flexibility and reduced effort ensures that the course can be easily integrated into existing medical curricula without the need for major changes. Last but not least, this kind of teaching format seems to be particularly suitable in times of the Covid-19 pandemic [31], as it makes it possible to switch spontaneously between online and face-to-face teaching since the online content is already available in its entirety.
Measuring changes in AI readiness
Karaca et al. [32] recently created the so-called “Medical Artificial Intelligence Readiness Scale for Medical Students” (MAIRS-MS), which tries to assess the perceived preparedness to use AI-applications in healthcare, i.e., “AI readiness”. The scale was psychometrically tested for reliability and validity on a population of Turkish medical students and achieved good results on both criteria. However, that scale was not originally designed to detect changes in AI readiness, but rather to evaluate the status quo of the concept. Nevertheless, it makes sense to resort to this scale and adapt it as an instrument to measure changes in AI readiness, as it is the only available instrument that has been psychometrically tested and developed specifically for medical students.
Purpose of the research
The main purpose of this study was to design and evaluate a novel AI-course for medical students. In addition to students’ attitudes towards the course, our primary research objective was to assess if and in what ways the course changed the AI readiness of the participants. Thus, this research attempts to answer two questions:
-
1)
What are students’ attitudes towards the course on AI in medical imaging?
-
2)
Does the AI-course presented here have an effect on students’ perceived AI readiness?
link