Development of a 3D nnU-Net-based cell tracking platform for quantifying myocardial deformation in zebrafish

Abstract

Quantifying three-dimensional preclinical myocardial deformation in vivo currently remains a challenging task, as non-invasive cardiac imaging methods are often restricted to two-dimensional projections suffering from limited reproducibility. Here, we introduce a 4D (3D + time) imaging and analysis platform that integrates a benchtop high-frame rate, modular light-sheet microscope (LSM) integrated with a deep-learning-enabled cardiomyocyte tracking pipeline. Our LSM achieves cellular resolution up to millimeter-scale field-of-view (FOV) at millisecond temporal sampling, allowing reconstruction of 4D trajectories of individual cardiomyocyte nuclei within embryonic zebrafish. To address segmentation failures arising from spatially overlapping nuclei within multilayered myocardium, we incorporated a difference-of-Gaussian and watershed-based preprocessing module into a 3D nnU-Net framework, enabling robust separation of cell-cell contact and mitigating nucleus-merging artifacts. This workflow supports reproducible extraction of myocardial deformation metrics and provides a scalable foundation for disease-specific biomechanical modeling. Together, our approach demonstrates how combining advanced LSM instrumentation with deep-learning-driven reconstruction pipelines can accelerate autonomous, high-throughput imaging of cardiac microstructure in vivo.