Generation of single-cell lineages from segmentation masks

The segmentation results from the various evaluation datasets were used to measure the cell-tracking performance of SC-Track and three other trackers: pcnaDeep [28], Deepcell-tracking [29] and TrackMate [30, 31]. For cell-tracking experiments involving in-house generated testing datasets, the segmentation results in the form of a VGG image annotator (VIA2) compatible JSON file containing cell cycle class information of each segmented cell was used [26]. The data in the JSON files were read directly by SC-Track and pcnaDeep to generate the cell lineage tables. The cell segmentation data in the JSON files were converted into greyscale multi-TIFF image files before being read by TrackMate and Deepcell-tracking, as both software packages lack the function to read JSON files directly.

Default tracking settings were applied to SC-Track, pcnaDeep and Deepcell-tracking. The overlap tracker algorithm was used with default tracking settings with TrackMate. Details summarizing the default settings used by all trackers are detailed in Table 1. We could not perform the cell-tracking experiments with the CTC dataset on pcnaDeep as it requires cell cycle class information to function [28]. The scripts used to generate the cell-tracking results are available from GitHub: https://github.com/chan-labsite/SC-Track-evaluation.

A summary table detailing the key information of each cell-tracking software used in this publication