A Novel Hierarchical Template Matching Model for Cardiac Motion Estimation

Abstract

Cardiovascular disease diagnosis and prognosis can be improved by measuring patient-specific in-vivo local myocardial strain using Magnetic Resonance Imaging. Local myocardial strain can be determined by tracking the movement of sample muscles points during cardiac cycle using cardiac motion estimation model. The tracking accuracy of the benchmark Free Form Deformation (FFD) model is greatly affected due to its dependency on tunable parameters and regularisation function. Therefore, Hierarchical Template Matching (HTM) model, which is independent of tunable parameters, regularisation function, and image-specific features, is proposed in this article. HTM has dense and uniform points correspondence that provides HTM with the ability to estimate local muscular deformation with a promising accuracy of less than half a millimetre of cardiac wall muscle. As a result, the muscles tracking accuracy has been significantly (p < 0.001) improved (30%) compared to the benchmark model. Such merits of HTM provide reliably calculated clinical measures which can be incorporated into the decision-making process of cardiac disease diagnosis and prognosis.

Hierarchical Template Matching for 3D Myocardial Tracking and Cardiac Strain Estimation

“…This fact motivates the use of 3D MRI instead of 2D short- and long-axis images, making it crucial to perform accurate 3D myocardial tracking. Medical image analysis researchers have contributed algorithms using direct detection-based methods 15 , Fourier-based methods 13,14,16 , tracking-based methods 17–22 , and block-matching-based methods 17,23–25 . Direct detection-based methods detect and track tag intersection points which could avoid image artefacts but too sparse tag points could limit the accuracy of myocardial tracking 12,26 .…”

“…The block-matching approach has provided promising results using 3D ultrasound images 17 , and 2D MRI 25 . The standard myocardial tracking technique uses 3D speckle tracking through ultrasound images 2,17 .…”

“…However, the accuracy of such a method is highly dependent on the size of the kernel, as too large or too small kernels could lead to incorrect tracking 17 . Such technical limitation could be overcome by using a hierarchy of matchings, which can provide a promising correlation up to the smallest level of the matching 25,27 . The hierarchical matching using 2D MRI is reported with promising accuracy in the literature 25 , however, 3D myocardial tracking using 3D MRI has received less attention in the literature 12 .…”

“…Such technical limitation could be overcome by using a hierarchy of matchings, which can provide a promising correlation up to the smallest level of the matching 25,27 . The hierarchical matching using 2D MRI is reported with promising accuracy in the literature 25 , however, 3D myocardial tracking using 3D MRI has received less attention in the literature 12 . Therefore, in this article, we have extended the standard block-matching technique by incorporating a hierarchical matching for 3D myocardial tracking.…”

“…It is also reported that the Cartesian coordinate B-spline model can generate errors due to ill-conditioned and ill-posed polynomials 30 . To address this issue, other approaches are suggested such as cylindrical coordinate B-spline model 30 (as LV is roughly cylindrical), explicit regularization 29 and local weighted-mean transformation function 25,28,31,32 . The cylindrical coordinate system and regularisation function with B-spline function could lead to detailed and complicated design and planning of myocardial tracking, whereas, Local Weighted Mean (LWM) function is proposed in the literature to overcome the elaborated procedure of spline interpolation 33 .…”