Related Experiment Video
Updated: Dec 3, 2025
12:54
Simultaneous Brightfield, Fluorescence, and Optical Coherence Tomographic Imaging of Contracting Cardiac Trabeculae Ex Vivo
Published on: October 2, 2021
3.5K
Cardiac motion estimation from medical images: a regularisation framework applied on pairwise image registration
Hadi Wiputra1, Wei Xuan Chan1, Yoke Yin Foo1
1Department of Biomedical Engineering, National University of Singapore, Singapore, 117583, Singapore.
Scientific Reports
|October 29, 2020
Summary
This study introduces a new method for cardiac motion estimation using a regularisation layer after image registration. The technique improves accuracy and consistency in ultrasound images, outperforming existing methods.
Area of Science:
- Medical Imaging
- Biomedical Engineering
- Computational Biology
Background:
- Accurate cardiac motion estimation is crucial for clinical evaluation of heart function.
- Existing methods for cardiac motion estimation often face challenges with spatial-temporal consistency and adherence to physiological constraints.
- Image registration is a key step, but subsequent motion refinement is needed.
Purpose of the Study:
- To present a novel regularisation layer for enhancing cardiac motion estimation accuracy.
- To demonstrate the effectiveness of this regularisation layer when applied after image registration.
- To provide a flexible and accurate add-on solution for various cardiac motion estimation algorithms.
Main Methods:
- Developed a novel regularisation layer utilizing a spatio-temporal model (b-splines of Fourier) for displacement fields from pairwise image registration.
- Enforced spatial and temporal smoothness, consistency, cyclic cardiac motion, and stroke volume adherence.
- Applied the method as an add-on layer post-registration, allowing flexibility with different registration algorithms.
Main Results:
- Achieved high accuracy in cardiac motion estimation.
- Demonstrated a 10% lower tracking error compared to Cardiac Motion Analysis Challenge (CMAC) participants on adult human ultrasound data.
- Obtained average Dice coefficients of 0.82-0.87 on diverse datasets including fetal echocardiography, chick, and zebrafish embryonic images.
Conclusions:
- Regularisation as an add-on layer post-image registration is a viable and accurate approach for cardiac motion estimation.
- This modular approach simplifies complex motion estimation algorithms and offers flexibility.
- The method shows broad applicability across various medical imaging modalities and biological models.
