Deep learning fixes heart motion tracking errors in ultrasound imaging
Myocardial Strain Drift Correction in Deep Learning Based Ultrasound Tracking
Artificial IntelligenceComputer Vision and Pattern Recognition
Summary
Tracking how the heart muscle moves in ultrasound videos helps doctors understand heart health, but these tracking methods can sometimes drift and give wrong results. The authors show a way to reduce this drifting by making the system remember information over a whole heart cycle and teaching it to respect realistic heart motion. This approach makes measurements of heart muscle strain more reliable and stable when repeated. It could help doctors get better data from heart ultrasounds without extra manual corrections.
What this means in practice
- •For cardiology imaging teams: Improve reliability of heart muscle strain measurements from ultrasound scans over multiple heartbeats using enhanced deep learning tracking.
- •For ultrasound device developers: Incorporate drift correction in AI-powered myocardial motion tracking to provide more consistent cardiac function analysis features in ultrasound machines.$Commercial implications: Enables ultrasound manufacturers to offer more accurate and clinically reliable cardiac function software, differentiating their products.
Authors
Thierry Judge, Nicolas Duchateau, Andreas Østvik, Havard Dalen, Bjørnar Grenne, Pierre-Yves Courand, Lasse Lovstakken, Pierre-Marc Jodoin, Olivier Bernard
Abstract
Myocardial strain from echocardiography is a key biomarker for cardiac function. Recent deep learning methods show strong performance for myocardial motion tracking but often lack physiological constraints, leading to temporal drift across the cardiac cycle. Consequently, tracked points may not return to their relative initial positions at the end of each cardiac cycle, producing inaccurate strain estimates and even divergence in some cases. We propose a deep learning framework that compensates for drift during myocardial tracking. We extend a state-of-the-art echocardiographic tracking method (TAS-Net) with persistent memory tokens that share information across sliding windows over full cardiac cycles. A teacher-student fine-tuning strategy on real echocardiographic data then enforces physiologically consistent cyclic motion while preserving tracking accuracy. Experiments show reduced global and regional strain drift, improved agreement with clinical references, and better test-retest reproducibility, supporting more reliable myocardial strain estimation in clinical practice.