Boundary Element Method-Based Scattered Feature Interpolation Algorithm in the Analysis of LV Deformation

Abstract

The quantification of left ventricular (LV) deformation from noninvasive image sequences is an important clinical problem. The accurate estimation of this information is often hindered by the following factors: i.) the imagederived features are found at sparse and unevenly sampled spatial positions; ii.) the currently available algorithms for interpolating features have tradeoffs related to accuracy, lattice density, physical plausibility and computation time. This paper introduces a new interpolation algorithm based on Boundary Element Method (BEM). By Comparing with more standard B-Spline based interpolation method: Free Form Deformation (FFD) with different lattice density, this new algorithm is more accurate and physically plausible, however, doesn't lose the computational efficiency. We then employ the BEM-based interpolation algorithm with the Generalized Robust Point Matching (GRPM) to derive the dense displacement fields from 3D LV image sequences. The approach is evaluated on 4 in-vivo cardiac magnetic resonance image sequences. The results are compared to the displacements found in implanted markers, used as a gold standard.