Automatic Image Analysis Pipeline for Studying Growth in Arabidopsis

Abstract

The need for high-throughput quantification of cell growth and cell division in a multilayer, multicellular tissue necessitates the development of an automated image analysis pipeline that is capable of processing high volumes of live imaging microscopy data. In this work, we present such an image processing and analysis pipeline that combines cell image registration, segmentation, tracking, and cell resolution 3D reconstruction for confocal microscopy-based time-lapse volumetric image stacks. The first component of the pipeline is an automated landmark-based registration method that uses a local graph-based approach to select a number of landmark points from the images and establishes correspondence between them. Once the registration is acquired, the cell segmentation and tracking problem is jointly solved using an adaptive segmentation and tracking module of the pipeline, where the tracking output acts as an indicator of the quality of segmentation and in turn the segmentation can be improved to obtain better tracking results. In the last module of our pipeline, an adaptive geometric tessellation-based 3D reconstruction algorithm is described, where complete 3D structures of individual cells in the tissue are estimated from sparse sets of 2D cell slices, as obtained from the previous components of the pipeline. Through experiments on Arabidopsis shoot apical meristems, we show that each component in the proposed pipeline provides highly accurate results and is robust to `Z-sparsity' in imaging and low SNR at parts of the collected image stacks.