Radiology, AI and Education: Keeping up with the Status Quo
Why technological advancements should position the users into the center of the change process
Artificial intelligence (AI) is about to fundamentally disrupt the way radiology will be practiced in the future. It has already found multiple applications in radiology, from image acquisition and processing to augmented reporting, as well as follow-up planning, data mining, and many others. AI algorithms can help spot details that may be too subtle to be seen by the human eye. They can develop entirely new pathways of interpreting medical images, sometimes in a nature humans are not able to explain. In addition, they are not — in principle — susceptible to perceptual biases, do not get tired, and can process millions of data points in parallel, unburdening the physicians of repetitive and time-consuming tasks.
‘Will AI replace the radiologist?’ or ‘The most frequently asked and discussed question in Radiology AI’
When human intelligence hits its limit, AI can take over, enhancing the radiologist’s performance. Some people even go as far as to speculate that radiologists will be among the first to be replaced by AI.
While experts see such a scenario to be very unlikely, this pessimistic outlook on job security has daunted medical students; a recent study published in the European Journal of Radiology found that more than 25% of the students who are not considering radiology as a specialty cite ‘AI’ as the main reason. Furthermore, it suggests that “students might be afraid to specialize in radiology because, among other reasons, they seem to fear the unknown future of AI.” Another study comes to the same conclusion, stating that almost half of the students who took the survey found that the prospect of AI caused them anxiety when considering the specialty. A lack of education on the fundamental principles of AI and its impact on radiology during medical training might be the root cause of the problem.
Being informed vs. being afraid
Daniel Pinto Dos Santos, who surveyed medical students in Germany, also came to the conclusion that they are not well-informed on the potential consequences of AI in radiology. Only 50% of the medical students were even aware that AI is a hot topic in radiology, and less than one-third of the respondents stated that they had a basic understanding of the AI techniques in question. Students pointed out that their information came more from the media than from academia. Interestingly, the students who were the most informed were also the least afraid of the new technology. At the moment, many educational programs do not have the competencies or the resources to include AI into their curricula. The goal, then, should be to educate students about AI early on in medical school, such that they can make informed decisions when choosing their specialty later on.
Education is key
It is not only students who can benefit from learning about AI. Hospitals and many other institutions are already heavily investing in AI for diagnostic imaging. Every year, more AI algorithms are certified, able to be deployed into the radiological workflow for increased efficiency and efficacy. Not surprisingly, many radiologists believe that AI should be included in their workflow as a clinical decision support system. As the use of some first algorithms is being reimbursed by insurance companies, radiologists are starting to find themselves at the heart of the technological transformation of their workplaces. Curtis Langlotz, a Stanford radiologist, encourages his colleagues to not fear AI, stating “AI won’t replace radiologists, but radiologists who use AI will replace radiologists who don’t”.
If radiologists want to be in charge of this transformation process, they must have a basic understanding of AI, including its opportunities, as well as its limitations and challenges. In the future, both students and practicing radiologists should familiarize themselves with terms like machine learning, supervised and unsupervised learning, deep learning, model generalizability, and other concepts, such as how training data affects model generalizability or how bias could creep into AI.
Michael Recht from the Department of Radiology at NYU says that “it will be necessary to incorporate a standard of continuing competence in AI for practicing radiologists into current imaging CME requirements.” Additionally, resources need to be allocated that allow for the education of already practicing radiologists. Furthermore, a standard curriculum should be developed for radiology trainees. While this might take years, there are fortunately a couple of free resources already available to help shed some light on the dark.
Here is our deepc shortlist:
The Data Science Institute of the American College of Radiology The DSI provides several services, working together with a variety of stakeholders. Check out their resources section’s links to articles and whitepapers on topics such as “Getting Started With AI” or “Integrating AI in Medical Imaging.” Their blog also provides a great way to keep oneself current on the latest topics and to read opinion pieces by different experts.
The AI blog of the European Society of Radiology The ESR’s blog is probably one of the best resources to be kept up to date. The posts have a very academic nature and are updated regularly, showcasing the newest research on AI in radiology.
AI-RADS The AI-RADS is an artificial intelligence curriculum for residents, where radiologists learn various foundational algorithms in AI without the requirement for a computational background.
The NHS AI Lab The NHSX leads the largest digital health and social care transformation program in the world. The AI Lab was created to address the challenge of fully harnessing the benefits of AI technology, using it safely and ethically at scale. They don’t just cover AI use in radiology, they also provide many great and insightful articles, reports, and white papers divided into the sections “Understand AI,” “Develop AI,” “Adopt AI.”
AIMed AIMed is a clinician-led movement dedicated to AI in medicine and healthcare. They provide many resources, regularly hosting webinars and other events. This is a place to discuss, educate, innovate, and drive change. Here, clinicians and experts from all over the world come together to talk about AI in healthcare, discussing the many resources focused on AI in radiology.
AI in Healthcare Specialization course from Stanford on Coursera This course discusses the current and future applications of AI in healthcare, with the goal of learning to bring AI technologies into the clinic safely and ethically. The course is designed for both healthcare providers and computer science professionals, offering insights that facilitate collaboration between the disciplines.
AI in Medicine course from deeplearning.ai on Coursera This course is a bit more advanced and technical, as participants should be able to program in Python, as well as being comfortable with statistics and probability. This course is for people who want to get a more in-depth understanding of the topic, with an interest in building AI models.
Technologically driven change processes need to strongly focus on the human
Radiology has always been a discipline driven by technological change, and at the same defining the forefront of technological change in medicine. While AI offers to augment the abilities of the human, there is little need for fear. Enabled by AI, radiologists might have the opportunity to become even better at doing their jobs. In the foreseeable future, radiologists will find themselves orchestrating and interpreting different AI models in a clinical context. In order to stay in the driver’s seat during this transformation process, radiologists must be well-informed to best assess the value of different algorithms, understand AI’s strengths and its limitations, and collaborate with AI researchers and other stakeholders.