Abstract
Myocardial tracking and strain estimation can non-invasively assess cardiac functioning using subject-specific MRI. As the left-ventricle does not have a uniform shape and functioning from base to apex, the development of 3D MRI has provided opportunities for simultaneous 3D tracking, and 3D strain estimation. We have extended a Local Weighted Mean (LWM) transformation function for 3D, and incorporated in a Hierarchical Template Matching model to solve 3D myocardial tracking and strain estimation problem. The LWM does not need to solve a large system of equations, provides smooth displacement of myocardial points, and adapt local geometric differences in images. Hence, 3D myocardial tracking can be performed with 1.49 mm median error, and without large error outliers. The maximum error of tracking is up to 24% reduced compared to benchmark methods. Moreover, the estimated strain can be insightful to improve 3D imaging protocols, and the computer code of LWM could also be useful for geo-spatial and manufacturing image analysis researchers.
Introduction
To diagnose cardiovascular diseases like cardiomyopathy or myocardial infarction, clinicians need to measure "strain"—how the heart muscle (myocardium) shrinks, expands, and rotates during a cardiac cycle. While 3D MRI provides a wealth of data, tracking the motion of the heart wall in three dimensions is notoriously difficult.
Existing methods often rely on 2D slices, which are prone to misregistration, or global "spline-based" mathematical models. These global models require solving massive systems of equations and can become "ill-conditioned," leading to massive error outliers if the data is noisy or the heart shape is irregular. The authors sought a method that is computationally efficient, "locally" aware, and stable enough to avoid these large tracking failures.
Methodology
The researchers developed a new 3D myocardial tracking framework that combines Hierarchical Template Matching (HTM) with a 3D extension of the Local Weighted Mean (LWM) transformation.
The pipeline consists of three primary steps:
- Segmentation: The initial geometry of the left ventricle (LV) is defined using a mesh registered to the MRI volume at end-diastole.
- Hierarchical Template Matching (HTM): To find corresponding points between different frames of the MRI, a pyramid-like matching strategy is employed. The image is divided into a hierarchy of blocks: Templates (16px) $\rightarrow$ Segments (8px) $\rightarrow$ Chunks (4px) $\rightarrow$ Windows (2px). 3D Normalized Cross-Correlation (NCC) matches these blocks across frames, ensuring consistency across the hierarchy.
- 3D Local Weighted Mean (LWM) Transformation: This core innovation extends the LWM function to 3D. LWM calculates a dense transformation using local second-order polynomials centered on control points. These polynomials are weighted by distance, meaning the transformation at any given point is only influenced by its nearest neighbors.
This local approach is computationally efficient (avoiding large matrix inversions) and inherently smoother, which is vital because strain is calculated from the derivatives of displacement.
Key Insights and Results
The study utilized a state-of-the-art dataset of 3D tagged MRI volumes from 15 healthy volunteers, comparing results against four established benchmarks from major research institutions (MEVIS, IUCL, UPF, and INRIA).
- Stability and Accuracy: The method achieved a median tracking error of 1.49 mm. More importantly, the maximum tracking error was reduced by 24% compared to the MEVIS benchmark, demonstrating superior numerical stability and a significant reduction in outlier errors.
- The "Local" Advantage: By using local polynomials (LWM), the algorithm adapts better to the non-uniform shape of the left ventricle, avoiding the numerical instability common in global models when fitting irregular or sparse data.
- Physiological Consistency: The algorithm successfully captured expected healthy heart patterns, including negative longitudinal and circumferential strain (contraction) and a peak in strain at end-systole.
Conclusion and Future Work
The primary limitation is the difficulty in capturing radial strain (wall thickening) due to fading tag lines and limited spatial resolution in 3D tagged MRI. Future work will explore multi-modality fusion, combining 3D tagged MRI with 3D steady-state free precision (SSFP) images to improve boundary definition. Additionally, clinical validation on diseased datasets is planned to assess the algorithm's ability to locate akinetic muscle segments. The 3D LWM code also holds potential for cross-disciplinary applications in geo-spatial analysis and industrial manufacturing.
Summary updated