Abstract
<jats:p>One of the areas of medicine where AI is currently being most actively implemented is radiology. However, until 2020, AI technologies were used only in scientific research. This was due to the existence of numerous insurmountable barriers. Previously, other research groups attempted to systematize the barriers to AI implementation, but these were largely theoretical and did not take into account the real-world experience of implementing AI in radiology. Purpose of the study is to systematize the barriers that hindered the implementation of AI in radiology, taking into account the experience of the Moscow experiment. Materials and methods. A retrospective analytical study was conducted to systematize the barriers that hindered the implementation of AI in radiology. Results. Twenty-one high-level barriers to the implementation of AI in radiology were identified, divided into six groups. The largest number of barriers were found in the regulatory sphere (6), while the fewest were related to data-related issues (2) and physicianpatient issues (2). Practical experience from the Moscow experiment on implementing computer vision in radiology revealed additional barriers not previously mentioned in theoretical studies. Conclusion. A systematization of barriers based on practical experience in AI implementation demonstrated their multifaceted and interconnected nature, which indicates the need to develop measures for their systematic, rather than sequential, elimination.</jats:p>