Seeing through the leak: a global perspective on aortic regurgitation assessment.

Abstract

Aims: Despite established guidelines, the echocardiographic quantification of aortic regurgitation (AR) remains challenging in clinical practice. While artificial intelligence (AI) solutions are being developed to support diagnostic assessment using echocardiography, their successful implementation will depend on understanding both current diagnostic challenges and clinician attitudes towards AI adoption. This study aimed to evaluate current practices in AR assessment, identify key challenges, and assess educational needs in AR diagnostics, while also investigating how healthcare professionals perceive AI assistance compared with human expert assessment.

Methods and results: We conducted a global online survey among sonographers and physicians. Participants answered questions about their current AR quantification practices, perceived limitations, and willingness to seek assistance from experienced colleagues or AI tools. Additionally, they were asked to grade AR severity in three sample echocardiographic cases. Among 1032 participants from 104 countries, 42% considered AR the most challenging valve lesion to assess. While guidelines recommend a multi-parameter approach, most practitioners relied primarily on visual colour jet assessment (51.5%) and basic measurements, with advanced quantitative parameters being notably underutilized (21.7%). Main limitations included eccentric jets (61.3%) and poor image quality (49.8%). Case-based assessments revealed significant variability in AR grading across experience levels (P < 0.001). Participants showed high confidence in both experienced colleagues and validated AI models (median confidence score of 7/10 for both) but less trust in newly developed AI tools (median confidence score 5/10).

Conclusion: This study demonstrates a substantial gap between guideline recommendations and clinical practice in AR quantification, with significant grading variability across and within expertise levels. While practitioners remain sceptical of newly developed AI tools, their openness to validated AI models suggests a potential pathway for improving the consistency of AR assessment.