SC-Track: a robust cell tracking algorithm for generating accurate single-cell lineages from diverse cell segmentations

Abstract

Computational analysis of fluorescent timelapse microscopy images is a powerful approach to study biological processes in detail. Core to this approach is the generation of accurate single cell linages from cell segmentations for reliable quantitative analysis. Convolutional neural networks (CNNs) are increasingly being used to segment and classify cells in microscopy images, but current cell tracking solutions are sensitive to inaccurate cell segmentations from CNNs. We present SC-Track, a cell tracking algorithm that employs a hierarchical probabilistic cache-cascade model. Our results show that SC-Track generates accurate single cell linages without parameter tuning, from cell segmentations of varying qualities, morphological appearances, and imaging conditions. Furthermore, SC-Track is equipped with a cell class correction feature to improve the accuracy of multi-class cell classifications in a time series. These features make SC-Track a robust generalist cell tracking algorithm that works with diverse segmentation outputs from CNNs to generate accurate cell linages and classifications.