Predictive K-PLSR Myocardial Contractility Modeling with Phase Contrast MR Velocity Mapping

Abstract

With the increasing versatility of CMR, further understanding of intrinsic contractility of the myocardium can be achieved by performing subject-specific modeling by integrating structural and functional information available. The recent introduction of the virtual tagging framework allows for visualization of the localized deformation of the myocardium based on phase contrast myocardial velocity mapping. The purpose of this study is to examine the use of a non-linear, Kernel-Partial Least Squares Regression (K-PLSR) predictive motion modeling scheme for the virtual tagging framework. The method allows for the derivation of a compact non-linear deformation model such that the entire deformation field can be predicted by a limited number of control points. When applied to virtual tagging, the technique can be used to predictively guide the mesh refinement based on the motion of the coarse grid, thus greatly reducing the search space and increasing the convergence speed of the algorithm. The effectiveness and numerical accuracy of the proposed technique are assessed with both numerically simulated data sets and in vivo phase contrast CMR velocity mapping from a group of 7 subjects. The technique presented has a distinct advantage over the conventional mesh refinement scheme and brings CMR myocardial contractility analysis closer to routine clinical practice.

Nonlinear Partial Least Squares An Overview

“…However, until recently and in contrast to linear PLS, nonlinear PLS has been mainly used in the chemical data analysis domain. It was the new concept of nonlinear kernel PLS, representing an elegant way of dealing with nonlinear aspects of measured data, which has considerably extended the applicability of nonlinear PLS into a wider area of research fields (Hardoon, Ajanki, Puolamaki, Shawe-Taylor, & Kaski, 2007;Lee, Wu, Huntbatch, & Yang, 2007;Mu, Nandi, & Rangayyan, 2007;Saunders, Hardoon, Shawe-Taylor, & Widmer, 2008;Trejo et al, 2006). The main reason is the fact that the kernel PLS method keeps computational and implementation simplicity of linear PLS while providing a powerful modeling, regression, discrimination or classification tool.…”

In vivo and in situ image guidance and modelling in robotic assisted surgery

“…To control the deformation in 3D, a volume constraint is introduced, as the volume of cardiac muscle varies little across the cardiac cycle. Lee et al [17] extended that work to incorporate the kernel-partial least squares regression to reduce the search space for the nodal deformation and to integrate the Kriging estimator [18]. Figure 2 shows results from this work as well as examples of the MR velocity images of the myocardium.…”

Cardiovascular magnetic resonance phase contrast imaging

“…Phase contrast sequences have been used as input to biomechanical models of the heart as shown for example by Liu [ 256 ]. Lee et al also used phase contrast myocardial data to validate myocardial contractility modeling [ 257 ]. There have also been a number of studies using myocardial velocity as an aid to segmentation of the left ventricle [ 258 ].…”