Assessment of Left Atrium Motion Deformation Through Full Cardiac Cycle

TL;DR

This work tackles the lack of 3D regional metrics for left atrium (LA) deformation by introducing an end-to-end 4D registration framework that combines high-resolution LA segmentation with displacement field (DVF) estimation across the full cardiac cycle. The approach fuses a 3D segmentation network (hybrid CNN-transformer) and a 4D registration model, enhanced by a CVAE-based motion-tracking module and a spatio-temporal Mask Autoencoder to map deformations into a low-dimensional latent space and forecast DVFs. Key contributions include (i) a 3D LA segmentation model trained on end-diastole and end-systole frames, (ii) a 4D registration pipeline producing accurate DVFs and high-quality meshes, and (iii) a motion-tracking framework that improves LA deformation estimation and enables regional fibrosis/strain analysis. On 41 subjects with 24-frame cycles, the Motion+MAE encoder achieved Dice around $0.88$ and Hausdorff Distance around $6.59\text{ mm}$, outperforming VoxelMorph and other baselines, highlighting potential clinical utility for AF management and personalized ablation planning.

Abstract

Unlike Right Atrium (RA), Left Atrium (LA) presents distinctive challenges, including much thinner myocardial walls, complex and irregular morphology, as well as diversity in individual's structure, making off-the-shelf methods designed for the Left Ventricle (LV) may not work in the context of the left atrium. To overcome aforementioned challenges, we are the first to present comprehensive technical workflow designed for 4D registration modeling to automatically analyze LA motion using high-resolution 3D Cine MR images. We integrate segmentation network and 4D registration process to precisely delineate LA segmentation throughout the full cardiac cycle. Additionally, an image 4D registration network is employed to extract LA displacement vector fields (DVFs). Our findings show the potential of proposed end to end framework in providing clinicians with novel regional biomarkers for left atrium motion tracking and deformation, carrying significant clinical implications.

Figures (6)

• Figure 1: overall pipeline of proposed method for LA segmentation and registration
• Figure 2: 4D Registration method to estimate DVF for LA motion tracking
• Figure 3: Validation results for subject 1 over the entire cardiac cycle, comparing the performance of the proposed and existing deep learning models.
• Figure 4: HD for full cardiac cycle using validation subject1 based on proposed and existing registration methods.
• Figure 5: Dice