Abstract

The challenge was to test instance segmentation methods on transmitted light (bright-field) confocal microscopy images of cells. Instance segmentation is de-sirable because it allows for automatic cell counting and quantification of indi-vidual cell morphologies. Typically instance segmentation is done using fluores-cence microscopy, and nuclear staining, but this has several drawbacks including photo-toxicity for live cell assays. Bright-field microscopy is used extensively in biomedical research hence the applications for the challenge are extensive. For example Dstl in particular is interesting in studying models of infection where segmentation of cells is an important step in the processing pipeline.

The group tested several different machine learning based approaches aiming to perform either semantic, or full instance segmentation, of cells within bright-field images.

Citation information

Data Study Group team. (2019, December 8). Data Study Group Final Report: Dstl – Bright-field image segmentation. Zenodo. http://doi.org/10.5281/zenodo.4452761