Fig 2.

Timelapse images of hMVECs reveal four major phenotypes: proliferative, apoptotic, migratory, and sessile. (a) Example contours outlining hMVEC in sessile, proliferative, migratory, and apoptotic states as detected by level set active contour. Proliferative instances are characterized by contour splitting, while apoptotic instances by contour collapsing. Scale bars are 20 μm. (b) Semi-hierarchical scheme for state classification. Contours are first classified based on whether there is a change in contour topology. If topology is conserved, they are further classified into sessile or migratory instances based on contour morphology and motility-related features. (c) Hierarchical clustering reveals two clusters within the subset of non-proliferative, non-apoptotic instances. The main discriminatory features between the two subsets are the mean and the variance of VACFs over 1 – 6 hour track interval. Sessile instances (blue) are characterized by low μ_VACF, high σ²_VACF while migratory instances (orange) are characterized by high μ_VACF, low σ²_VACF. (d) PCA projection of the motile vs. sessile instances as classified by an optimized ensemble of sessile vs. motile base classifiers (details in Supplementary Method). (e) Example instances of single cell contour and centroid tracks labeled to show the sessile vs. motile state identification results.