Agreement between automated echocardiography and expert cardiologist for aortic valve hemodynamic parameters in severe aortic stenosis patients

Abstract

Background

Aortic stenosis is one of the leading valve diseases. Transthoracic echocardiography (TTE) is the primary imaging technique for diagnosing aortic stenosis, providing comprehensive evaluations of valve hemodynamics and cardiac remodeling. However, it is time-consuming and requires significant expertise. While automated echocardiographic tools hold promise in potentially streamlining diagnostic processes and reducing measurement variability, their accuracy and reliability relative to expert cardiologist measurements require validation.

Purpose

To assess the agreement of aortic valve (AV) hemodynamic parameters derived from automated echocardiography analysis software compared to measurements performed by experienced cardiologists.

Methods

Fifty-eight patients with severe symptomatic aortic stenosis underwent comprehensive TTE evaluation. Each patient underwent a comprehensive TTE evaluation, where several key parameters were measured. For AV assessment, maximum velocity (Vmax), maximum gradient (Gmax), mean gradient (Gmean). TTE measurements were performed using standard protocols and guidelines. AV measurements were obtained from multiple acoustic windows, ensuring optimal visualization and alignment for accurate velocity and gradient assessments. Automated analysis algorithms were applied to the same echocardiographic studies to perform analogous measurements of AV parameters.

Results

The analysis demonstrated a high yield of 94.83% for aortic valve (AV) parameters. Pearson correlation analysis revealed strong agreement between human and AI measurements for AV Vmax, AV Gmax, and AV Gmean, with correlation coefficients of 0.78, 0.80, and 0.77, respectively. There was minimal bias for AV Vmax (0.1, ± 0.81 limits of agreement) and AV Gmax (bias of 3.34, ±30.02 limits of agreement). However, AV Gmean exhibited a bias of -3.63 (±22.14 limits of agreement).

Conclusion

The study shows strong agreement between automated and human measurements for aortic valve parameters.

Author notes

Funding Acknowledgements: None.