(E-169) In-silico benchmark to validate four-dimensional cardiac motion quantification in small animals.

Introduction:: The quantification of myocardial deformation via medical imaging techniques has a growing prevalence in diagnosing cardiovascular diseases. For instance, the global longitudinal strain has shown great promise as a sensitive marker of disease progression. Cardiac magnetic resonance (CMR) imaging, integrated with different image registration techniques, has proven to be a rigorous modality to estimate regional strains. However, the complexity of four-dimensional (4D) cardiac motion confounds the accuracy of strain estimation, often leading to significant differences in estimated strain quantities depending on vendor, software, and operator. Moreover, these methods often rely on tissue tagging CMR, which is not readily available for small animals. Although essential in-vitro benchmark phantoms have been offered and implemented to evaluate the reliability of strain calculation schemes, the existing phantoms offer motion and geometries simpler than those in real hearts. Therefore, there remains a crucial need for an in-silico benchmark that replicates complex 4D cardiac motions and can be used as ground truth to validate any strain quantification techniques. In this study, we propose an in-silico benchmark to validate the quantification of 4D cardiac strains using a finite element (FE) approach. Synthetic CMR images obtained from two FE solutions were used to validate our modality-agnostic non-rigid image registration (NRIR) framework [1]

Materials and Methods:: Geometry Reconstruction: (i) an idealized geometry of a three-dimensional hollow tube was created, and (ii) a murine left ventricle (LV) was reconstructed from short-axis CMR images using Delaunay tetrahedralization. FE simulation: (i) the base of the cylinder was fixed, and a simple twist was applied at the top (Fig. 1A), (ii) a full inverse model [2] was conducted to replicate the pressure-volume measurements in the same mouse for which the CMR images were obtained (Fig. 1B). Synthetic MR image generation and strain calculation: Using the deformation across each increment, image slices were synthesized as uniform grids across six short-axis planes. Noise was added via cubic interpolation and Gaussian filtering, and the images were subjected to NRIR to calculate 4D strains. (Fig 2A, B). The strains were validated against original FE strains (ground truth).

Results, Conclusions, and Discussions:: For the first problem (torsion in a cylinder), FE strains were in agreement with the analytical solution. Mean squared (MSE) analysis at six different segments of the mid-section revealed good agreement with an error of 7% between the analytical/FE strains and the NRIR circumferential strains. In the second problem, strains were compared at the mid-slice of the LV. Preliminary comparisons point to a larger error (~ %10), which likely is due to a larger “out-of-plane” pixel motion. Increasing the number of slices and improving the interpolation between slices is expected to reduce the error. In this work, we have presented an in-silico benchmark pipeline to validate imaging-based strain calculations. The ubiquity of research on assessing the functional relevance of cardiac strains necessitates the standardization of the various calculation strategies. To our knowledge, this is the first implementation of an FE-based benchmark for small animals using standard CMR imaging sequences.


Acknowledgements (Optional): : RA was supported by the NIH, under grant No. R00HL138288.

References (Optional): : [1] M. Keshavarzian et al., "An Image Registration Framework to Estimate 3D Myocardial Strains from Cine Cardiac MRI in Mice," in Functional Imaging and Modeling of the Heart, (Lecture Notes in Computer Science, 2021, ch. Chapter 27, pp. 273-284.

[2] R. Avazmohammadi et al., "A Computational Cardiac Model for the Adaptation to Pulmonary Arterial Hypertension in the Rat," Annals of Biomedical Engineering, vol. 47, no. 1, pp. 138-153, 2018, doi: 10.1007/s10439-018-02130-y.