Leveraging the Rich Spatiotemporal Features of Lattice Light-sheet Microscopy with Machine Learning and AI

Abstract

Live-cell imaging allows scientists to observe the dynamics of living cells across time. Lattice Light-sheet (LLS) microscopy is one such method that captures these processes at high spatiotemporal detail in 4D. LLS enables us to observe previously unknown events, however, the large data size, specialized processing needs, and the complexity of the feature rich datasets pose significant challenges for maximizing the utility of this technology. To this end, we developed napari-lattice, a python plugin within napari, an n-dimensional viewer that streamlines LLS analysis. It enables users to extract specific regions of interest within LLS data without processing the entire volume. Furthermore, napari-lattice integrates seamlessly with standard image analysis pipelines, enabling segmentation and feature extraction in a single end to end workflow. We applied the napari-lattice workflow to live-cell imaging of Neutrophil extracellular trap (NET) formation, a form of programmed cell death exhibiting dynamic changes in cell shape, topology and nuclear DNA conformation, as multilobular nuclei decondense and DNA is extruded extracellularly. Using primary human neutrophils, we study how cells from different donors behave under various NET-inducing stimuli. To enable this, we developed an end-to-end workflow that extracts morphological information in 2D and 3D for live cells over time, which is modular and scalable. Traditionally, time series data is summarized using basic statistics such as mean, maximum, number of peaks and area under the curve. However, this approach fails to capture the full complexity and dynamics of the temporal changes. We address this limitation by using tsfresh, a python package that computes multiple statistical properties to summarize temporal changes.