Estimation of Segmental Longitudinal Strain in Transesophageal Echocardiography by Deep Learning

TL;DR

This work tackles the challenge of automated regional LV function assessment in transesophageal echocardiography by developing autoStrain, a pipeline that fuses deep learning motion estimation with segmental strain calculation. It compares two DL approaches—TeeFlow (RAFT-based dense optical flow) and TeeTracker (CoTracker-based point trajectories)—using a novel synTEE synthetic dataset (80 patients) generated via SIMUS, including decorrelation and simulated infarctions. TeeTracker consistently outperforms TeeFlow, achieving a mean motion-estimation error of $0.65$ mm on synthetic data, and, when validated on 16 real TEEs, demonstrates clinically meaningful agreement for SLS ($1.09 ext{%}$ bias) and GLS ($-2.36 ext{%}$ bias). The combination of synthetic-dataset–driven training and infarct-aware augmentation yields the best performance, suggesting that AI-driven motion estimation can enhance precision and efficiency of perioperative cardiac monitoring, with further validation planned. $\,$

Abstract

Segmental longitudinal strain (SLS) of the left ventricle (LV) is an important prognostic indicator for evaluating regional LV dysfunction, in particular for diagnosing and managing myocardial ischemia. Current techniques for strain estimation require significant manual intervention and expertise, limiting their efficiency and making them too resource-intensive for monitoring purposes. This study introduces the first automated pipeline, autoStrain, for SLS estimation in transesophageal echocardiography (TEE) using deep learning (DL) methods for motion estimation. We present a comparative analysis of two DL approaches: TeeFlow, based on the RAFT optical flow model for dense frame-to-frame predictions, and TeeTracker, based on the CoTracker point trajectory model for sparse long-sequence predictions. As ground truth motion data from real echocardiographic sequences are hardly accessible, we took advantage of a unique simulation pipeline (SIMUS) to generate a highly realistic synthetic TEE (synTEE) dataset of 80 patients with ground truth myocardial motion to train and evaluate both models. Our evaluation shows that TeeTracker outperforms TeeFlow in accuracy, achieving a mean distance error in motion estimation of 0.65 mm on a synTEE test dataset. Clinical validation on 16 patients further demonstrated that SLS estimation with our autoStrain pipeline aligned with clinical references, achieving a mean difference (95\% limits of agreement) of 1.09% (-8.90% to 11.09%). Incorporation of simulated ischemia in the synTEE data improved the accuracy of the models in quantifying abnormal deformation. Our findings indicate that integrating AI-driven motion estimation with TEE can significantly enhance the precision and efficiency of cardiac function assessment in clinical settings.

Figures (6)

• Figure 1: Simulation pipeline that was deployed to generate synthetic 2D TEE B-mode sequences with reference myocardial contraction fields, referred to as the synTEE dataset.
• Figure 2: Examples of SLS curves after simulation of myocardial infarction, obtained in the synTEE dataset. (a) original strain curves before manipulation; (b) and (c) reduced local myocardial deformation in a given segment; (d) visualization of the color map for myocardial segments.
• Figure 3: Visualization of myocardial tracking by point trajectory estimation with TeeTracker. TeeTracker had a sliding window length of 8 frames.
• Figure 4: Bland–Altman plot comparing manual and automatically estimated SLS and GLS measures in synTEE data, including all four datasets (Table \ref{['tab:datasets_ratio']}). TeeTracker was used for automatic measurements. The blue line indicates the mean difference, and the red lines indicate the 95% limits of agreement. SLS, segmental longitudinal strain.
• Figure 5: Examples of SLS curves obtained in synTEE test data by (a) ground truth references with a synthetic infarction in the mid septal segment (cyan); (b) TeeTracker trained in a combined scheme; (c) TeeTracker trained in a combined scheme with synthetic infarcted segments (d) visualization of the colormap for myocardial segments.