Feature Identification

Segmenting Cells

Cell segmentation is a very challenging task for image-based transcriptomics experiments. There are many approaches that do a very good job of segmenting nuclei, but few if any automated approaches to segment cells. Starfish exposes the watershed segmentation method from classical image processing, which inverts the intensities of the nuclei and spots and treats them like a literal watershed basin. The method then sets a threshold (water line), and each basin is treated as its own separate segment (cell).

This approach works fairly well for cultured cells and relatively sparse tissues, but often cannot segment denser epithelia. As such, starfish also defines a simple segmentation format to enable it to read and utilize segmentation results derived from hand-drawing or semi-supervised drawing applications.

TODO cell segmentation demo