Background: In recent years, there has been a significant focus on the development of artificial intelligence (AI) applications in healthcare. However, current scientific evidence is still not convincing enough for the general public and the medical community to widely adopt AI in clinical practice. Objective: We conducted this study to investigate the correlation between left ventricular function indices assessed by AI and those evaluated by physicians. Methods: This cross-sectional descriptive study was conducted on 136 patients who attended and received treatment at Hue University of Medicine and Pharmacy Hospital from April 2022 to June 2023. Using QLAB version 15, Philips Healthcare. Results: The AI software accurately identified 98.5% of the echocardiographic cine-loops. However, there were about 1.5% of cine-loops that the software failed to recognize. The sensitivity of Ejection Fraction (EF) calculated by AI was 73.3%, specificity was 81.3%, and accuracy stood at 78.6%. A strong positive correlation was observed between EF measured by AI and that assessed by physicians, r = 0.701, p < 0.01. The sensitivity of Global Longitudinal Strain (GLS) calculated by AI was 42.1%, specificity was 84.8%, and accuracy was 67.6%. A moderate positive correlation was found between GLS measured by AI and physician’s assessment, r = 0.460, p < 0.01. Conclusion: The use of AI software for evaluating left ventricular function through ejection fraction and global longitudinal strain is rapid and yields results comparable to cardiologists’ echocardiographic assessments. The AI-powered software holds a promising and feasible future in clinical practice.
Key words: Automated software, artificial intelligence, ejection fraction, longitudinal strain.
|
