Fig. 11

Manual versus semi-automatic segmentation. Shown above is the DA/ZC dataset (grayscale) overlaid with segmentations (orange) resulting from thresholding at intensity values 130, 140, and 150 (arb. units), as well as the result of semi-automatic region-growing segmentation using the ITK-SNAP software [43]. Thresholding at too low an intensity value (a) under-segments the interior cell wall, leaving much of the biomass in the cell wall mislabeled as void space. As the threshold intensity is increased (b), this under-segmentation problem is reduced. However, in order to eliminate under-segmentation of the interior of the cell wall, one must choose a threshold value so high that the lumen is over-segmented (c). Through user intervention using the convenient ITK-SNAP tool, a segmentation (d) is produced in which biomass in the interior of the cell wall is correctly identified, while avoiding mislabelling dark regions in the lumen as biomass