Artificial intelligence in breaking the learning curve for echocardiography: a secondary analysis of a multicentre trial.

Abstract

Aims: The integration of point-of-care ultrasound (POCUS) by non-specialists and the shortage of trained sonographers highlights the need for scalable training approaches. This study aimed to evaluate the learning curve of novice operators performing artificial intelligence (AI)-guided limited transthoracic echocardiography (TTE) and to assess whether acquired images were sufficient for diagnostic interpretation of structural cardiac disease.

Methods and results: In this multicentre, prospective secondary analysis, nine novice operators performed limited TTE scans on 159 patients using a handheld device with AI-based acquisition guidance. Following eight hours of standardized training, novices independently obtained six standard TTE views. Three blinded expert reviewers graded image quality on a 1-5 scale and assessed diagnostic adequacy. Image scores were used to generate learning curves, and subgroup analyses examined the influence of patient characteristics. Of 954 novice-acquired images, 97.7% met the diagnostic threshold (score ≥3). After training, all operators achieved mean scores ≥3 across patients. AI-guidance consistently enabled high-quality imaging across all views, with minimal impact from sex, age, or pathology. Body mass index (BMI) showed a significant effect (P = 0.0029), though all subgroups exceeded diagnostic thresholds: 4.44 ± 0.17 (BMI <18), 4.40 ± 0.04 (18-24), 4.12 ± 0.12 (25-29), and 4.07 ± 0.07 (>30). Experts reliably ruled out left ventricular dysfunction (99.4%) and hypertrophy (98.7%); agreement was lower for wall motion abnormalities (80.7%) and atrial dilation (86.6%).

Conclusion: Novices with no prior POCUS experience achieved diagnostic-quality TTE images after one day of AI-guided training. AI may supplement conventional echocardiography training, and future research should evaluate its integration into routine clinical workflows.