Fig. 2: Pixel classification by machine learning in Ilastik as a pre-step for cell segmentation. | Nature Communications

Fig. 2: Pixel classification by machine learning in Ilastik as a pre-step for cell segmentation.

From: Multiplexed histology analyses for the phenotypic and spatial characterization of human innate lymphoid cells

Fig. 2

a A sum membranes image is created by summing of all membrane stainings in ImageJ in order to have one single image containing the spatial information of all cell membranes stained in a MELC run. The white square represents the region of interest (ROI). b Fluorescence images from the nuclear staining (DAPI, red) and the sum membranes generated in (a) (green) are the input data for Ilastik. Complete images are shown in the upper panel and the ROIs are shown in the lower panel. c A machine-learning-based algorithm for pixel classification (random forest) is interactively trained by manually drawing labels for nuclei (pink), membranes (yellow), or extracellular matrix (ECM, cyan) on an ROI of the overlaid input images (nuclear staining in red and sum membranes in green). d The trained algorithm calculates probability maps for nuclei, membranes, and ECM in the whole image and in other data sets. The maps can be exported and subsequently used for cell segmentation. Complete probability maps are shown in the upper panel and ROIs are shown in the lower panel. A second ROI is depicted as dashed lines and represents the tissue area that will be further used as an example in Fig. 3. Scale bars: 50 µm. a–d (n = 7).

Back to article page