Suppression of artifact-generating echoes in cine DENSE using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular MRI

Background:

• Cardiac MRI using displacement encoding with stimulated echoes (DENSE) visualizes myocardial motion.
• T1-relaxation echo artifacts can contaminate DENSE images, necessitating longer scan times for artifact reduction.
• Accurate myocardial strain quantification is crucial for diagnosing cardiac conditions.

Purpose of the Study:

• To develop and validate a deep learning method for suppressing T1-relaxation echo artifacts in DENSE MRI.
• To enable reduced scan times without compromising image quality or strain quantification.
• To compare the deep learning approach against traditional k-space zero-filling methods.

Main Methods:

• A U-Net architecture (DAS-Net) was trained using complementary phase-cycled DENSE data as ground truth.
• A data-augmentation technique generated synthetic DENSE images with varying encoding frequencies.
• DAS-Net processed non-phase-cycled DENSE images acquired during shorter breath-holds, comparing results to phase-cycled DENSE and zero-filling.

Main Results:

• DAS-Net effectively suppressed T1-relaxation echo artifacts, achieving high image similarity (SSI = 0.85 ± 0.02) and low error (RMSE = 5.5 ± 0.8).
• DAS-Net significantly outperformed k-space zero-filling in artifact suppression (P < .01).
• Myocardial strain quantification using DAS-Net on accelerated scans closely matched results from conventional phase-cycled DENSE.

Conclusions:

• The DAS-Net method offers an effective deep learning solution for T1-relaxation echo artifact suppression in DENSE MRI.
• This approach enables a significant 42% reduction in scan time compared to standard phase-cycling techniques.
• DAS-Net facilitates faster and robust myocardial strain analysis, improving clinical applicability of DENSE MRI.