Related Experiment Video
Updated: May 17, 2026
07:21
Magnetic Resonance Derived Myocardial Strain Assessment Using Feature Tracking
Published on: February 12, 2011
DSHARP: Deep Incompressible Motion Estimation With Sinusoidal-Transformed Harmonic Phase for Tagged MRI
IEEE Transactions on Medical Imaging
|May 15, 2026
Summary
Deep sinusoidally transformed HARP (DSHARP) enhances tagged MRI analysis by integrating deep learning with the harmonic phase approach. This method accurately quantifies incompressible motion fields, overcoming limitations of previous techniques.
Area of Science:
- Medical Imaging
- Biomedical Engineering
- Machine Learning
Background:
- Tagged magnetic resonance imaging (tMRI) visualizes and quantifies in vivo tissue deformation.
- Limitations include tag fading, long computation times, and difficulty ensuring diffeomorphic, incompressible motion fields.
Purpose of the Study:
- To develop a novel method for estimating 2D and 3D motion fields from tMRI data.
- To achieve motion estimation that is both diffeomorphic and nearly incompressible.
Main Methods:
- Integration of the harmonic phase (HARP) approach with an unsupervised deep learning registration framework.
- Development of deep sinusoidally transformed HARP (DSHARP) utilizing phase transformation for end-to-end network training.
- Estimation of stationary velocity fields and use of a Jacobian determinant loss term to ensure diffeomorphic and incompressible motion.
Main Results:
- DSHARP demonstrated superior tracking accuracy, computation speed, and incompressibility preservation compared to HARP, SinMod, SyN, PVIRA, VoxelMorph, and DeepTag.
- Successful evaluation on 2D/3D phantom data, human tongue motion during speech, and cardiac tagged MRI benchmark data.
Conclusions:
- DSHARP offers a significant advancement in tMRI motion analysis, overcoming key limitations of existing methods.
- The developed deep learning framework provides accurate, efficient, and robust estimation of diffeomorphic and incompressible motion fields.
