Motion-based structure separation for label-free, high-speed, 3D cardiac microscopy

Capturing the dynamics of individual structures in the embryonic heart is an essential step for studying its function and development. Label-free brightfield (BF) microscopy allows for higher acquisition frame-rates than techniques requiring molecular labeling, without interfering with embryo viability or needing complex equipment. However, since different structures contribute similarly to image contrast, label-free microscopy lacks specificity. Here we mitigate this problem by separating a single-channel image series into multiple channels specific to different cardio-vascular structures, based only on their motion patterns. The technique combines images from multiple cardiac cycles and z-sections after non-uniform temporal registration to produce 3D+time image volumes of one full cardiac cycle with separate channels for static, transient and periodically moving structures. The resulting data is well-suited for velocity analysis and 3D-visualization. We characterize the separating capabilities of our technique on a synthetic cardiac dataset and demonstrate its practical applicability, by reconstructing three-channel views of the beating embryonic zebrafish heart with an effective frame rate of 1000 volumes (256×256×20 voxels each) per second. This technique enables quantitative characterization of dynamic heart function during cardiogenesis.