Chalmers Conferences, 9th European Conference on Mathematical and Theoretical Biology

Automated multi-cell segmentation and tracking from live bioimaging datasets
Yanthe E. Pearson, Jeremy Teo Choon Meng

Last modified: 2014-03-27


Live cell bioimaging is an important tool for acquiring datasets necessary for dynamic studies of biological mechanisms. While imaging is a vital step in unraveling the dynamics and functions of cellular processes, datasets, especially that of cells in 3D matrices possess inherent noise. Noise is a consequence of several experimental factors such as a background drift, cell traction on matrices, and interference on cells of interest by neighboring cells.

Data extraction, either manual or semi-automated, remains a challenge and the ability to distinguish artifacts from biology is highly dependent on the quality and resolution of each image. Manual tracking of identical cells is still the most reliable technique for extracting biological information from a time series of automated microscopy images, however, several studies have shown automated cell tracking to be possible and necessary, but restricted to specialized data sets to explore specific biological phenomena.

With this this in mind, our study focuses on the development of a user friendly, highly flexible algorithm that will import a sequence of time lapse images, parsing each image through an automated filtering and segmentation scheme with an option for either Morphological filtering or Edge detection, followed by threshold segmentation.

Initial stages of the algorithm uses two consecutive images by comparing the current image to the previous image, measuring displacement of all identifiable objects from time t to t+1. Minimum Euclidean distance criteria are used for matching (biological) objects from frame to frame. Unmatched objects are treated as either new or old. An old object implies it has appeared in earlier images, while new objects have not.

Algorithmic developments are underway which will allow for user specified threshold levels, morphological structuring element and expected (or maximum) number of biological objects to be identified and tracked. Automatic data extraction and (organized) storage of both migratory and spatial variables, such as Centroid coordinates, Area, aspect ratio are implemented.

Reliability of this framework was measured through evaluation and crossvalidation by comparing to manual tracking across immune cell migration datasets (See poster titled: Extracting T-lymphocyte cell dynamics in biommetic microenvironments, Aysha Alsuwaid) and sensitivity analysis to varied levels of filtering thresholds.


Bioimaging; Multicell Tracking; Automated data storage