Implementation of computer-aided learning (CAL) into clinical cardiology is accompanied by a number of complex barriers that significantly slow down the translation of methodological, ethical and legal as well as infrastructural algorithms in practice.
OBJECTIVE
To analyze systematically and to structure data on key barriers of implementing computer-aided learning in clinical cardiology, as well as to propose recommendations for overcoming them considering international experience and Russian realities.
MATERIALS AND METHODS
A thematic content analysis of 40 publications (31 English-language, 2019—2024; 9 Russian-language, 2004—2024) selected in PubMed, Web of Science, Scopus and eLibrary.ru databases was performed by «machine learning», «cardiology», «implementation», «barriers» keywords. The analysis covered methodological, interpretative, ethical and legal as well as infrastructural aspects.
RESULTS
The main difficulties of implementing CAL methods in cardiological practice were identified. These include methodological limitations manifested in low reproducibility of models and necessity to recalibrate them for local data; «black box» problem when explaining the decisions of algorithms remains difficult for clinicians; insufficient development of ethical and legal base that creates a risk of responsibility in algorithms errors and means the need for standards unification; infrastructural barriers including the heterogeneity of medical images formats and the lack of centralized annotated registers that limits the volume of available data for learning.
CONCLUSION
Computer-aided learning algorithms in cardiology have the potential to improve prognosis, diagnosis and therapy personalization, nevertheless their implementation is limited by data quality, methodology, ethics and integration into clinical processes. Solving these issues by standardization, Explainable AI technology use and development of industrial protocols has the potential to improve integration process.
