Impact of Training Data Composition on Deep Learning-Based Cardiac Motion Estimation
Open Access
- 5 December 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Ultrasonics, Ferroelectrics, and Frequency Control
- Vol. 5, 286-291
Abstract
Cardiac ultrasound deformation imaging is a non-invasive and widely used modality for assessing left ventricular function. However, due to the variable image quality observed in clinical routine, deformation estimation remains challenging. Recently, deep learning (DL) approaches have been proposed to estimate motion from ultrasound images. In this study, we investigated whether combining in vivo data with a synthetic dataset during training improves motion estimation accuracy while also benchmarking the DL-based estimates against a state-of-the-art traditional optical flow method. Results demonstrate that the deep learning-based method excels in tracking the dense motion fields on a frame-by-frame basis. However, when performing contour tracking, it results in lower end-systolic peak strain values compared to the traditional optical flow method.Keywords
Funding Information
- Research Foundation Flanders (G0C7120N)
- Foundation for Science and Technology (FCT), through Portuguese National Funds (PTDC/EMD-EMD/1140/2020)
- Foundation for Science and Technology (FCT), through Portuguese National Funds (CEECIND/03064/2018)
This publication has 14 references indexed in Scilit:
- Large-Scale Simulation of Realistic Cardiac Ultrasound Data With Clinical Appearance: Methodology and Open-Access DatabaseIEEE Access, 2024
- GMFlow: Learning Optical Flow via Global Matching2022
- Motion Estimation by Deep Learning in 2D Echocardiography: Synthetic Dataset and ValidationIEEE Transactions on Medical Imaging, 2022
- An unsupervised learning approach to ultrasound strain elastography with spatio-temporal consistencyPhysics in Medicine & Biology, 2021
- Myocardial Function Imaging in Echocardiography Using Deep LearningIEEE Transactions on Medical Imaging, 2021
- Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow EstimationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2019
- MITT: Medical Image Tracking ToolboxIEEE Transactions on Medical Imaging, 2018
- Intervendor Differences in the Accuracy of Detecting Regional Functional AbnormalitiesJACC: Cardiovascular Imaging, 2018
- Definitions for a Common Standard for 2D Speckle Tracking Echocardiography: Consensus Document of the EACVI/ASE/Industry Task Force to Standardize Deformation ImagingJournal of the American Society of Echocardiography, 2015
- A fast convolution-based methodology to simulate 2-Dd/3-D cardiac ultrasound imagesIEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 2009