Segmentation and thickness measurement of glomerular basement membranes from electron microscopy images

An algorithm for segmentation and thickness measurement of the glomerular basement membranes (GBM) in electron microscopy kidney images is presented. Differences in intensities and variations between GBM and other components in the image are employed. Regions of extreme intensities such as the black area of blood cells and white areas of urinary spaces are pre-excluded. Areas of sharp edges are either at the GBM borders or unrelated to GBM regions. These non-GBM sharp edges, along with the pre-excluded regions, are used as barriers limiting the size of the fitting circles centered at a location in the image domain to form a two-dimensional function, proportional to the radius of the largest fitting circle, at the location. A local peak in the radius function corresponds to the largest circle in the local area. The set of the combined peaks in two perpendicular directions is calculated before a thinning procedure is applied. After removing the unwanted branches, a centerline of the GBM is produced. The segmentation of the GBM is then straightforward from expanding each point in the centerline to a circle of radius defined by the radius function. The average of the diameters of the circles gives the average GBM thickness. Results of the real GBM images are provided. Visual comparisons from the superimposed GBM boundaries show that the algorithm provides accurate GBM segmentation. The evaluations of the average GBM thicknesses are also compared to those from the manual tracing method.