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

• What's SC-Track?

  SC-Track is an efficient multi-object cell tracking algorithm that can generate accurate single cell linages from the segmented nucleus of timelapse microscopy images. It employs a probabilistic cache-cascade matching model that can tolerate noisy segmentation and classification outputs, such as randomly missing segmentations and false detections/classifications from deep learning models. It also has a built-in cell division detection module that can assign mother-daughter relationships using the segmentation masks of cells.

  SC-Track allows users to use two different segmentation results as input. It can either take:

  1) A greyscale Multi-TIFF image, where every segmented instance of cells is given a unique pixel value and the background set as 0. 2) A VGG image annotator (VIA2) compatible JSON file, containing the segmented instances and class infromation of every segmented cell.

  ---

  Why use SC-Track?

  1) It is generally accepted that segmented instances of objects detected from state-of-the-art deep learning convolution neural networks such as U-net and Mask RCNN will produce low instances of segmentation and classification errors. The most common errors are false detections, failed detections or wrong classification of detected objects. These inaccuracies often confound commonly used tracking algorithms. SC-Track works well with noisy outputs from these deep learning models to generate single-cell linages from these segmented timelapse microscopy images. 2) SC-Track is compatible with the output results from manually segmentated images as well as from most existing mainstream segmentation models, such as Cellpose, DeepCell, and StarDist. 3) SC-Track probabilistic cache-cascade matching model can efficiently track multiple targets between frames and therefore can be used for real-time tracking. 4) SC-Track is implemented in Python, making it easy to be integrated into any deep learning based segmentation pipelines to be used to generate single-cell tracks from timelapse microscopy images.

How to use SC-Track?

If you have a single-channel segmentation result, the segmentation results should be in a grayscale 2D+t tiff file with each segmented cell instance containing a unique pixel value with the beackground given a pixel value of 0. You will need to run SC-Track from the folder containing the image and mask TIFF files. The command to call SC-Track is as follows:

sctrack -i /path/to/image.tif -a /path/to/mask.tif

When the segmentation results are contained in a VIA2 compatible JSON file, you will need to run SC-Track from the folder containing the image and JSON files. The command to call SC-Track is as follows:

sctrack -i /path/to/image.tif -a /path/to/annotation.json

The file "image.tif" corresponds to the microscopy timelapse image stack, "mask.tif" represents the greyscale segmented cell instances and "annotation.json" is the VIA2 compatible JSON annotation files.

Below is the expected outputs from the tracking results assuming that the "image.tif" and "annotation.json" file was provided.

|__image.tif
|__annotation.json
|__tracking_output\
   ├─track.csv
   └─track_visualisation.tif

The track_visualization.tif are greyscale images visualising the tracking results, and track.csv is a detailed table of the tracking results.

For specific information about the track.csv, see https://github.com/chan-labsite/SC-Track/blob/master/notebook/quick-start.ipynb.

SC-Track can run without the corresponding "image.tif" file. In this case, SC-Track will output the tracking results without a corresponding track_visualization.tif folder containing the labelled cell linages.

To access our demo dataset, you can visit here.

Installation

To install the latest version of SC-Track, please download SC-Track from this github repository, navigate to the downloaded SC-Track folder where setup.py is located and run the command:

python setup.py install

Note：

On some systems, the required package pylibtiff cannot be installed directly by pip. Please use the following command instead:

conda install libtiff

or you can download the wheel package from here, and use the command pip install pylibtiff.whl to install the package.

On Linux or macOS, you can use the command pip install pylibtiff.

The installation times on a "normal" desktop computer should not exceed 5 minutes assuming that the computer is connected to a reasonably fast (10 Mbps) broadband connection.

If you encounter the following error ValueError: Failed to find TIFF header file, you can try running sudo apt-get install libtiff5-dev to fix the error.

Usage

To automate SC-Track for batch processing, please refer to our source code for additional documentation. Its basic usage is:

from SCTrack.track import start_track

image = 'path/to/image.tif'

# using mask annotation
annotation_mask = '/path/to/annotation.tif'
start_track(fannotation=annotation_mask, fimage=image, basename='image', track_range=None, fout='/path/to/dir')

# using json file annotation
annotation_json = '/path/to/annotation.json'
start_track(fannotation=annotation_json, fimage=image, basename='image', track_range=None, fout='/path/to/dir')

Using guidance

To see the using guidance, please refer our quick-start.

API Documentation

For more information, please see the reference documents.

Please cite our paper if you found this package useful.

SC-Track: a robust cell tracking algorithm for generating accurate single cell linages from diverse cell segmentations
Chengxin Li，Shuang Shuang Xie，Jiaqi Wang，Septavera Sharvia，Kuan Yoow Chan
https://doi.org/10.1101/2023.10.03.560639