mobstr3D: Open-Source Software for Analysing Cine DENSE CMR to Quantify Regional Heart Motion and Strain

Description: Quantifying cardiac wall motion is critical for understanding normal and abnormal heart function. Regional left ventricular (LV) functional measures, such as strain and torsion, have been highlighted as early indicators of cardiovascular disease, such as ischemic heart disease, in which regional wall motion abnormalities are common [1-4].

Medical imaging methods allow for non-invasive assessment of LV wall motion. Such methods include speckle-tracking echocardiography [5], feature tracking from routine clinical cardiac magnetic resonance (CMR) [6], CMR myocardial tagging [7,8], and CMR displacement encoding with stimulated echoes (DENSE) [9]. Accurate kinematic assessment requires reliable reconstruction of 3D deformation fields from the acquired displacement information. Existing tools are generally restricted to 2D in-plane quantification of strain, which limits their ability to describe the complex 3D motion of the heart.

In this abstract, we present a novel, open-source software package, mobstr3D (MOdel-Based STRains in 3D, v1.0.0) for analysing myocardial mechanics by reconstructing 3D patient-specific biophysical kinematic models from acquired 3D displacement information, from which regional strains are estimated. mobstr3D is developed in Python (v3.13.2) and utilises the open-source finite element meshing library, HOMER (High Order MEsh Representations) [10].

mobstr3D takes inputs of mid-LV myocardial contours and the masked 3D displacement components, representing local motion of the myocardium accrued since phase tagging (typically at end-diastole). For DENSE with 3D displacement encoding, the input data can be accessed via preprocessing of the MATLAB files generated by the existing 2D analysis tool, DENSEanalysis [11,12]. Firstly, the surfaces of a bi-cubic Hermite quadratic Lagrange 8x1x1 element cylindrical template model are fitted to the contours, providing a geometric mesh of LV anatomy at the frame of interest. Displacements then inform a free-form deformation (FFD) fit to define the associated deformation model at the frame of phase tagging. Strain analysis is then performed to calculate the local cardiac wall coordinate Green-Lagrange strain tensor (circumferential, longitudinal and radial strains, and the associated shear strains). RV inserts and LV centroid selection allow for AHA segmental and transmural regional analyses. If multiple frames are fitted for a single case, strain-time analyses can also be generated.

Work In Progress:

• Extension to a full truncated ellipsoid model of the LV.


• End-to-end DENSE Pipeline: data pre-processing of DENSE DICOMs, including automatic segmentation, slice correction and phase unwrapping.

Access Instructions: mobstr3D can be accessed through the University of Auckland Heart Mechanics Research Group GitHub:

https://github.com/UOA-Heart-Mechanics-Research/mobstr3D



The repository README.md contains detailed instructions on how to install, setup and run the software using the provided example data.

Figure 1. Schematic of the mobstr3D software package pipeline for generating mid-LV models, using cine DENSE CMR with 3D displacement encoding acquisitions as an example. Preprocessing of DENSE CMR, including segmentation and phase unwrapping is performed ahead of fitting in mobstr3D. 1) The surfaces of a cylindrical template model are fitted to the input contours, providing a geometric model representing the LV anatomy at the time frame of interest (e.g. end-systole). 2) A deformation model is fit using the displacement information to define the associated model at the time frame of phase tagging (e.g. end-diastole). 3) Regional strain analysis is performed to calculate the local cardiac wall coordinate Green-Lagrange strain tensor, and strain versus time analyses.