Artificial intelligence in medical studies

AI and algorithms already have the potential to revolutionise medical education and thus fundamentally change the medical profession. But how can this best succeed?

AI in medical studies: promising but underused

Translated from the French and German versions.

Artificial intelligence, robotic surgery, nursing robots, quantified self, networked objects, etc. - robotics and algorithms have already taken a firm place in medical practice. But are medical faculties ready for the digital transition in healthcare? Unfortunately, this is still in doubt.

In terms of form and content, medical studies have changed little for over a century: always the same subjects, always the same lectures. When people talk about AI, it tends to be in the corridors. Yet these new technologies have the potential to revolutionise initial medical training and fundamentally change the medical profession. It just needs to be harnessed.

Xiaoyi: A student (almost) like any other

Xiaoyi is a Chinese AI that passed the selection process for medical school with flying colours.1 The algorithm finished the exam in just one hour - its "human" fellow students, on the other hand, needed ten hours. With 456 points, Xiaoyi was far above the national average and even achieved 96 points more than the minimum score required for admission.

This may be frightening, but: Xiaoyi will never become a real physician. Although it knows the curricula by heart, it still can't diagnose prostate carcinoma or leukaemia. Because Xiaoyi - like any AI - is too monotonous.

However, the example from China shows that medicine can be learned by a machine. Could AIs even be able to teach medicine in the foreseeable future? Yes - and no. "Human" professors will probably continue to teach the basics of medical education, while AIs will do the assimilation through adaptive learning.

The promise of adaptive learning

Adaptive learning (AL) is a pedagogical concept based on the findings of cognitive neuroscience. The aim: to adapt the learning path to the competences, abilities, and goals of each individual learner. An "à la carte" teaching, in other words, but without the need for one teacher per student.2

The idea is simple: the machine tests the students to learn how their memory works and what their cognitive patterns are. If the student is good at internal medicine but bad at cardiology, the AI tries to understand why the student's brain has these difficulties. It then uses neuroscience to apply the most appropriate learning method.

Many faculties have started with MOOCs (Massive Online Open Course), i.e. digital learning platforms where learners can take online courses and then test their knowledge. The health crisis has greatly accelerated their development. Unfortunately, only a few faculties in Europe provide AL on their MOOC platforms.

Such teaching tools are much more developed in English-speaking countries. An Australian study of mathematics students found that incorporating AL into a MOOC increased exam pass rates by 18% and reduced course dropouts by 47%.3

So far, there has been little research on the use of AL in medical studies. However, this is likely to change quickly. Elsevier's English-language version has an interface called Cerego. Its algorithm suggests a personalised learning path to the student by suggesting the next module to work on.

Cerego is even able to examine the profile of the student's memory, e.g. to determine how long it takes for a piece of knowledge to be forgotten. The algorithm can thus suggest repeating a module again shortly before this period expires. On a dashboard, the student can see where they stand at any point in time and visualise their progress.

Kellmann et al. piloted a teaching pathway using AL for teaching in the dermatological histopathology course at UCLA, California. Significant improvements were observed in the grades achieved.4

AL within a simulation framework

It should be noted that AL can also be done without AI. This is the case with simulation sessions, but where a teacher must be made available to a limited group of students. The cost is inevitably limiting.

Also in the field of simulation, the so-called serious games can benefit greatly from the contributions of AL. These types of video games present a clinical case to assess the learner's reasoning. Currently, if a mistake is made, the scenario does not continue or continues only slightly and the learner has to correct him/herself immediately. AI would allow the scenarios to evolve constantly, depending on the learner's choices, providing a realistic immersion experience.

AI at the heart of medical education

AI can therefore be a pedagogical tool. But it should also become the subject of teaching. According to an Odoxa-UNESS survey, 79% of medical students feel helpless and inadequately trained to incorporate AI into their professional practice.5

The fact that university diplomas dealing with AI are being introduced at European medical schools should be good news. The bad news is that AI can only be learned through paid and voluntary training. The risk is that the gap between "AI-friendly" doctors and those who distrust it, often due to lack of knowledge, will widen.

In order to integrate AI into medical training at an early stage, there is a need for convergence with engineering informatics courses. Incidentally, this is also suggested by the WHO in its report on AI in health care: "Governments and companies should anticipate the upheavals that will occur at the working level, especially the training of health workers who need to become familiar with the use of AI systems."6

Three types of measures should therefore be implemented without delay:

  1. The establishment of a separate field of study "Medicine and AI", in which students should be trained on technological advances, their application in daily practice and the general impact on the medical profession.
  2. Adaptive learning should be largely democratised.
  3. Medical degree programmes should systematically follow degree programmes that specialise in specific areas of informatics.

References:

  1.  French only: Enrique Moreira – «Un robot chinois réussit son concours de médecine» (Les Echos, 2017)
  2. French only: Thomas Blanc – «L’adaptive learning, définition et idées reçues» (Tactileo, 2018)
  3. Sharma N, Doherty I, Dong C.: Adaptive Learning in Medical Education: The Final Piece of Technology Enhanced Learning? Ulster Med J. 2017 Sep;86(3):198–200.
  4. Kellman PJ. et al.: Adaptive and perceptual learning technologies in medical education and training. Mil Med. 2013 Oct;178(10 Suppl):98–106.
  5. Online survey conducted at the end of 2018 with 752 health professionals (doctors, pharmacists, nursing assistants, midwives), 978 students and 258 teachers (medicine, pharmacy, odontology and sports).
  6. Ethics and governance of artificial intelligence for health. WHO guidance (2021)