Researchers beat clinical STE with AI-driven physics simulation for heart strain
New AI method reduces global strain variability to 1.42%, outperforming clinical gold standard.
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
A large international research team led by Thierry Judge (Université de Sherbrooke) and collaborators from Norwegian and French institutions have published a paper demonstrating that deep learning, when trained on physics-based simulations, can outperform the clinical gold standard for myocardial strain estimation. Speckle tracking echocardiography (STE) is currently the standard for measuring heart muscle deformation, but its accuracy for regional strain—critical for early detection of subtle abnormalities—remains limited. Deep learning offers a promising alternative, but development has been hampered by a lack of reliable ground-truth motion references.
The researchers tackled this by proposing a novel simulation strategy that incorporates speckle decorrelation measures from real echocardiographic videos and uses an iterative refinement process to improve motion realism. They created and open-sourced a photorealistic dataset of 1,478 synthetic videos with known reference motion. Using this data, they trained a motion estimation algorithm that achieved unmatched performance: a global longitudinal strain (GLS) variability of 1.42% in an inter-expert setting, significantly better than the 1.78% of the clinical reference. The method also showed improved regional strain accuracy, addressing a long-standing limitation of STE. The paper, available on arXiv (2605.28697), represents a significant step toward AI-powered cardiac diagnostics that could enable earlier and more precise detection of heart disease.
- New physics-based simulation uses real speckle decorrelation and iterative refinement to generate 1,478 photorealistic videos for training
- AI model achieves GLS variability of 1.42% vs 1.78% for clinical STE, a 20% improvement
- Regional strain accuracy—often poor with STE—is significantly enhanced, enabling earlier detection of subtle cardiac abnormalities
Why It Matters
This AI method could replace clinical STE for more accurate, earlier heart disease diagnosis using existing ultrasound hardware.