CardioSpectrum: Comprehensive Myocardium Motion Analysis with 3D Deep Learning and Geometric Insights

TL;DR

3D deep learning architecture based on the ARFlow model, based on the ARFlow model, is optimized to handle complex 3D motion analysis tasks and can contribute to improving cardiovascular diagnosis and treatment.

Abstract

The ability to map left ventricle (LV) myocardial motion using computed tomography angiography (CTA) is essential to diagnosing cardiovascular conditions and guiding interventional procedures. Due to their inherent locality, conventional neural networks typically have difficulty predicting subtle tangential movements, which considerably lessens the level of precision at which myocardium three-dimensional (3D) mapping can be performed. Using 3D optical flow techniques and Functional Maps (FMs), we present a comprehensive approach to address this problem. FMs are known for their capacity to capture global geometric features, thus providing a fuller understanding of 3D geometry. As an alternative to traditional segmentation-based priors, we employ surface-based two-dimensional (2D) constraints derived from spectral correspondence methods. Our 3D deep learning architecture, based on the ARFlow model, is optimized to handle complex 3D motion analysis tasks. By incorporating FMs, we can capture the subtle tangential movements of the myocardium surface precisely, hence significantly improving the accuracy of 3D mapping of the myocardium. The experimental results confirm the effectiveness of this method in enhancing myocardium motion analysis. This approach can contribute to improving cardiovascular diagnosis and treatment. Our code and additional resources are available at: https://shaharzuler.github.io/CardioSpectrumPage

Figures (5)

• Figure 1:
• Figure 2: CardioSpectrum Architecture: The NN analyzes 3D image pairs of cardiac cycle timesteps, incorporating 2D constraints from ZoomOut. These constraints, derived from segmentations converted into meshes, result in a 3D optical flow.
• Figure 3: Comprehensive comparison of proposed model and two baselines in accurately capturing cardiac deformation across a range of torsion angles. The metrics evaluated included the mEPE over the myocardium volume, the mEPE in locally-radial and locally-tangential directions within the segmentation hull. As depicted, our model exhibited strengths that were particularly evident in \ref{['fig:sub1']} representing the overall mEPE, and pronounced advantages in locally-tangential components \ref{['fig:sub3']}, which aligns with our method’s goal of addressing the aperture problem, a challenge that was difficult to face by the two baseline models. Error bars represent the standard errors (SE). See Fig. 1 in supplementary material for the component-wise mEPE in radial, circumferential, and longitudinal directions and mean angular error.
• Figure 4: Additional comparison of the proposed model and two baselines in capturing cardiac deformation across torsion angles, breaking down mEPE into radial (\ref{['fig:sub1']}), circumferential (\ref{['fig:sub2']}), and longitudinal (\ref{['fig:sub3']}) components. Radial and circumferential mEPE consistently showcase our model’s superior performance. Although performance relatively decreased at lower torsion angles in the longitudinal component, all models had lower errors than in other components. Subfigure \ref{['fig:sub4']} shows dissimilarity between predicted and true flow directions based on angular error, with our method outperforming baselines particularly at higher torsion angles. Error bars represent standard errors (SE). These figures supplement Figure 3 and Section 5.1 of the main paper.
• Figure 5: (\ref{['fig:sub5']}) The cardiac cycle involves radial (LV cavity center to myocardial wall), circumferential (tangential along the epicardial wall), and longitudinal (along the LV’s long axis) LV movements. Locally-tangential and locally-radial movements refer to projections onto and perpendicular to the myocardial surface’s tangent plane. (\ref{['fig:sub6']}) A sample before (top) and after (bottom) deformation. Arrows represent selected ground truth annotations. Sample from the 3D Slicer library (http://www.slicer.org). (\ref{['fig:sub7']}) Comparing mEPE of CardioSpectrum, baselines, and ZoomOut-derived constraints across torsion angles, evaluated over the segmentation map hull. "Constraints" is composed of errors from ZoomOut and voxel-mesh conversions, impacting CardioSpectrum’s performance, especially at lower angles.