Neural Implicit 3D Cardiac Shape Reconstruction from Sparse CT Angiography Slices Mimicking 2D Transthoracic Echocardiography Views

TL;DR

This work tackles 3D cardiac shape reconstruction from sparse 2D slices that mimic apical TTE views by learning a neural implicit shape prior from CTA-derived segmentations. It jointly optimizes a per-subject latent code and rigid view poses at test time to recover full 3D multi-class cardiac anatomy, achieving a mean Dice score of about $0.86\pm0.04$ across structures and substantially lower LV and LA volume errors than Simpson’s biplane method. The approach remains robust to pose perturbations and offers a promising route to more accurate 3D chamber quantification from 2D echocardiography. The method highlights the value of combining implicit representations with pose-aware test-time optimization for medical image translation between CTA priors and 2D ultrasound-like views.

Abstract

Accurate 3D representations of cardiac structures allow quantitative analysis of anatomy and function. In this work, we propose a method for reconstructing complete 3D cardiac shapes from segmentations of sparse planes in CT angiography (CTA) for application in 2D transthoracic echocardiography (TTE). Our method uses a neural implicit function to reconstruct the 3D shape of the cardiac chambers and left-ventricle myocardium from sparse CTA planes. To investigate the feasibility of achieving 3D reconstruction from 2D TTE, we select planes that mimic the standard apical 2D TTE views. During training, a multi-layer perceptron learns shape priors from 3D segmentations of the target structures in CTA. At test time, the network reconstructs 3D cardiac shapes from segmentations of TTE-mimicking CTA planes by jointly optimizing the latent code and the rigid transforms that map the observed planes into 3D space. For each heart, we simulate four realistic apical views, and we compare reconstructed multi-class volumes with the reference CTA volumes. On a held-out set of CTA segmentations, our approach achieves an average Dice coefficient of 0.86 $\pm$ 0.04 across all structures. Our method also achieves markedly lower volume errors than the clinical standard, Simpson's biplane rule: 4.88 $\pm$ 4.26 mL vs. 8.14 $\pm$ 6.04 mL, respectively, for the left ventricle; and 6.40 $\pm$ 7.37 mL vs. 37.76 $\pm$ 22.96 mL, respectively, for the left atrium. This suggests that our approach offers a viable route to more accurate 3D chamber quantification in 2D transthoracic echocardiography.

Figures (3)

• Figure 1: Method overview. Dashed elements represent operations that are exclusive to testing. Training: an MLP learns multi-class occupancies from dense CTA points concatenated with a per-shape latent code. Testing: 2D masks are mapped to 3D via a rigid transform $T_{\text{view}}$; $T_{\text{view}}$ and the latent code are optimized while the MLP is frozen.
• Figure 2: Example test result. (a) Initial orientation of TTE-mimicking slices showing that the outer borders of the LA in the A3C (white dotted line) and A5C view (black dotted line) are misaligned. (b) Slices with optimized orientations, demonstrating improved alignment of the LA borders. (c) Our 3D shape reconstruction from sparse slices shown in (b). (d) 3D rendering of reference CTA segmentation. LV = left ventricle, LA = left atrium, RV = right ventricle, RA = right atrium, LV-MYO = left-ventricle myocardium.
• Figure 3: Heatmap visualizations of surface distance errors of reconstructed whole-heart and left-ventricle shapes. Errors are computed between the reconstructed shape and reference CTA segmentation. In this comparison between two experiments, the cases are determined based on the average symmetric surface distance (ASSD) of the proposed method (left).