It is a study which, at a time of fascination with artificial intelligence (AI) and that of its implementation in the national health strategy planned by 2030 in France, rekindles the debate on its effects on our brain. Just published in the journal The Lancet Gastroenterology & Hepatology, It demonstrates for the first time that the use of AI makes doctors less efficient, with the ability of the experimenters, who are nevertheless experienced here, declining very quickly, in just six months.
The authors of this study wanted to know if AI modified the behavior of users, here digestive endoscopists performing colonoscopies,The examination is used to detect precancerous polyps, known as adenomas. As a reminder, if these lesions are discovered at an early stage, the examination allows them to be removed, which prevents the development of colorectal cancer.
For several years now, digestive endoscopy technical platforms have been gradually equipped with endoscopes including AI algorithms to both better detect and better characterize lesions discovered during the examination. This translates into practical for this aid to radiological diagnosis, by an audible or visual alert occurring during the colonoscopy, in order to draw the radiologist's attention to a potentially critical area.
Here, the Polish researchers' objective was to analyze the impact of the introduction of AI in the practice of a group of 19 experienced endoscopists, each with more than 2,000 examinations to their credit. The originality of this work, the first of its kind, was therefore to compare their diagnostic performance, i.e. their rate of detection of adenomas ateye naked without AI, before his arrival but also three months after.
Read alsoMedicine: the era of automated diagnosis
“ When technology assists you, the brain starts to function differently.
Conducted in four centers in Poland, this observational and so-called retrospective study was conducted from the end of 2021 to the beginning of 2022, the period when AI was being implemented in these structures. The observations lasted six months during which the researchers compared the quality of 1,443 colonoscopies, all performed without AI: 795 were performed before its deployment and 648 others after. And the results are as follows: in six months, the adenoma detection rate went from 28.4 % to 22.4 %, which corresponds to an absolute difference of -6 %, i.e. a drop in human expertise and therefore in the performance of radiologists in identifying lesions.
A result which suggests that it is not the tool itself which is at fault, but rather its use, via the loss of vigilance which it induces, and this through the automation which it generates. “ I'm not really surprised by these numbers., reacts Professor Geoffroy Vanbiervliet, gastroenterologist at the Nice University Hospital and secretary general of the French Society of Digestive Endoscopy (SFED). When technology assists you, the brain starts to function differently, a bit like how, since the arrival of mobile phones, we no longer make an effort to learn numbers by heart. But what matters most is whether the use of AI is beneficial or not for the patient, and we don't know that yet.".
And the specialist continued: “ While various studies have demonstrated that AI in endoscopy can increase the polyp detection rate by approximately 10% in real time, this actually mainly concerns the identification of small benign polyps, those that would not have degenerated anyway or that other examinations would have ultimately found later. On the other hand, the impact of AI on the reduction in the incidence of colorectal cancer remains to be demonstrated..
A subtle balance to find
Today in France, it is also difficult to know what practices look like, the AI tools offered by manufacturers varying considerably depending on the models and brands of endoscopes, without any comparative study conducted between them, as with medications.
In its editorial published in the same issue as the study, the British journal states that " While caution is needed, a blanket moratorium risks stifling beneficial innovation, especially when lives are at stake. Instead, a rigorous approach is needed that balances innovation with safety. Rigorous oversight, along with ongoing training and evaluation by clinicians, patients, and developers, can help make AI a valuable tool for healthcare..
“ For us at the SFED, the challenge today is to combine teaching and quality in our continuing education and training programs for radiologists, with the use of AI always remaining under human control,” insists Professor Vanbervielt. In short, a subtle balance between technological assistance, know-how, safety and benefit for the patient.
In the meantime, don't forget colon cancer screeningTwo-thirds of those affected, those over 50, do not do so and each year, 45,000 new cases are identified and 17,000 die.
English 
French