Many quantitative digital pathology tasks require the segmentation of cells or other histological components of an image. Variation in cell shape and morphology is a feature that is strongly associated with many types of cancer.
Sometimes it may not be enough to simply detect the center of a cell, as one pair of coordinates cannot sufficiently describe the shape of a cell. Cell segmentation combines the location and appearance of cells in an image into a numerical description of its shape.
A segmentation is often represented as a “mask,” a binary image indicating where a cell is (1) and where a cell is not (0), or as an ordered list of x,y coordinates that describe the outermost perimeter of the cell boundary. With such cell segmentations, one can calculate the total area of the cell, the minimum and maximum Feret diameters, convexity measures such as extent and solidity, and even generate a set of typical cell shapes within the given dataset.
Much like detection, cell segmentation is generally accomplished using variants of the CNN architecture, in particular the recent fully convolutional network (FCN) models. At CytoInformatics, our proprietary cell segmentation takes advantage of the state-of-the-art methods, ensuring that your results are as robust and accurate as can be.
Cell segmentation is one of our most commonly requested services at CytoInformatics. To see how we can help you achieve accurate cell segmentation, submit a quote today.