Analysis of artificial intelligence errors in practical medicine and rehabilitation

Aim: To structure the types of errors that occur when using artificial intelligence in healthcare, as well as assess their impact on the accuracy of diagnostics and therapeutic decisions. Identify ways to minimize errors and increase the effectiveness of the use of AI in practical healthcare and rehabilitation.

Materials and Methods: Publications published from January 2021 until December 2025 were processed and analyzed according to the keywords of the topic of work "Pathology", "research on the results of diagnostics and treatment", "artificial intelligence", "machine learning", "deep learning", "federated learning", "use of AI in rehabilitation", "structuring of AI errors" in the databases of PubMed, MEDLINE, Web of Science. Articles were selected based on the presence of: quantifiable results and usage of AI as the main or secondary evaluation method. A total 57 articles were reviewed, out of which 38 were excluded based on eligibility criteria and 2 were excluded as duplicates.

Conclusions: Artificial intelligence is becoming an integral part of modern medical diagnostics and therapeutic solutions. Its implementation significantly increases the accuracy of diagnostic processes and allows for personalized treatment, but today there is no universal solution for the practical use of AI. A lot of errors are still recorded when using AI in diagnostic and prognostic processes, ethical issues have not been resolved, integration of all molecular information available to the patient is not always ensured, there are no uniform standards for collecting and processing medical data, a unified medical language, etc. To ensure professional machine learning, widespread implementation of an open healthcare system, balanced and unified guiding principles is necessary. The implementation of AI technologies depends on the training of doctors and the availability of technologies.