Skip to main content
. 2020 Nov 27;11:593177. doi: 10.3389/fimmu.2020.593177

Figure 3.

Figure 3

CytoSkaler for automated quantification of subcellular AVA binding. (A) Automated whole cell area segmentation was enabled by incorporating an additional fluorescence channel for the anti-α enolase 1 antibody (anti-ENO1) stain, applying autothresholding, despeckling, largest zone selection and subsequent space filling. (i) An example of a single cell containing FOV is given. A manually segmented whole cell area is given together with an approximation based on thresholding of the anti-ENO1 fluorescent signal. Vimhi and nuclear regions were segmented by autothresholding as for Figure 1 . (ii) Correlation of autothresholded anti-ENO1 signal to manual segmentation of a whole cell area (36 single cell-containing FOVs tested, Spearman r = 0.957 (95% C.I. = 0.915 to 0.978), mean average intersection of unions (IOU) = 0.80. (B) Schematic of machine learning based method to train the CytoSkaler program to segment individual cells and respective subcellular areas. (i) Acquired FOVs, each containing multiple HEp-2 cells were inputted as respective raw color channels. (ii) Channels were autothresholded to yield binary images for each channel based on pixel intensity. (iii) Separated nuclear boundaries from the binary Hoechst channel were used as markers to produce RGB images for the V9 and anti-ENO1 channels, yielding a single cell’s nuclear boundary (colored green) and all other cells’ nuclear boundaries colored red. This process was repeated for every nucleus in the Hoechst channel. (iv). Two neural networks were trained to separately classify subcellular pixels around Hoechst green boundaries for the V9 and anti-ENO1 channels. Trained on manually segmented cells from 189 ground truth FOVs (2094 cells) with training over 2,300 iterations. These yielded mean IOU scores of 0.96 and 0.89 for the V9 and ENO1 networks, respectively. (v) Whole cell segmentations were produced following union of individual channel segmentations and space filling. (vi) All segmented cells in a single FOV are marked in a different color to show complete CytoSkaler output of a multicellular zone.