Fig. 1
From: Micro-anatomical quantitative optical imaging: toward automated assessment of breast tissues
Algorithms for nuclear (a-d) and ductal (e-l) segmentation. Nuclear segmentation: a: Raw image acquired from confocal fluorescence microscope with 750 × 750 μm2 field of view. b: Region of interest selected in confocal fluorescence image with 75 × 75 μm2 field of view. c: The maximally stable extremal regions (MSER) algorithm applies thresholds from 0 to 255 to b. d: At each threshold, the MSER algorithm identifies nuclei as connected components and selects “maximally stable” components with the lowest size variation. Ductal segmentation: e: Raw image acquired from confocal fluorescence microscope with 750 × 750 μm2 field of view. f: Wiener lowpass filter and adaptive histogram equalization applied to e. g: The algorithm converts E to a binary image using an interactive threshold tool. h: Objects below range of nuclear area are removed and then user selects a region of interest (ROI) around ducts with an interactive polygon selection tool. i: The algorithm fills boundaries of ducts identified in (h) to segment the outer boundaries of the duct. j: The algorithm selects the complement of (h) to segment the inner boundaries of the duct (lumen). k: Duct wall width is measured by selecting the shortest distance from the outer to the inner duct boundaries (red lines). l: Ellipses are fitted to outer and inner duct boundaries. e-j: Scale bar is 100 μm. k, l: Scale bar is 25 μm