Figure 1
From: Environmental properties of cells improve machine learning-based phenotype recognition accuracy
Segmentation and feature extraction. (a) Segmentation of the MCF-7 breast cancer cell line using CellProfiler 2.2.0. Scale bar: 50 µm (b) SLIC superpixel segmentation of urinary bladder cancer tissue section images. Images were segmented into superpixels of different sizes (25, 50, 100 pixels). Scale bar: 50 µm (c) The K-nearest neighbours (KNN) method, illustrated in a schematic figure and in real cell culture and tissue section scenarios, K = 5. Scale bars: 25 µm (d) The n-distance method, illustrated in a schematic figure and in real cell culture and tissue section scenarios, n = 50 pixels (cell culture: 19.51 µm, tissue sections: 13.5 µm). Scale bars: 25 µm (e) Superpixels containing two different phenotypes (cancer cell and fibroblast) share highly similar regular features, but features of their neighbourhoods differ significantly. Scale bars: 50 µm.
