Left ventricle motion estimation for cine MR images using sparse representation with shape constraint.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Cardiology

Background:

• Accurate estimation of left ventricle (LV) motion is crucial for diagnosing cardiac diseases.
• Spatial-varying intensity distortions due to tissue deformation pose challenges for traditional motion estimation methods.

Purpose of the Study:

• To propose a novel left ventricle (LV) motion estimation method utilizing sparse representation.
• To address and overcome spatial-varying intensity distortions caused by myocardial deformation.

Main Methods:

• An adaptive dictionary was generated for myocardial landmarks by learning transformations from a training dataset.
• Landmark tracking was performed using sparse representation, followed by applying a point distribution model.
• Dense displacement fields of the LV myocardium were estimated, and circumferential strain was calculated to assess myocardial function.

Main Results:

• The proposed method achieved superior performance compared to state-of-the-art techniques, demonstrated by the smallest average perpendicular distance (APD) and mean symmetric contour distance (SCD), and the highest Dice metric on public cardiac datasets.
• A mean strain difference of -0.01 and an intraclass correlation coefficient of 0.91 were observed when compared to the commercial software Medis Suite MR.

Conclusions:

• The developed method accurately estimates the dense displacement field of the LV, outperforming existing techniques.
• Circumferential strain derived from the proposed method shows excellent agreement with commercial software, indicating its clinical potential for detecting segmental strain abnormalities in heart disease patients.