Segmentation and analysis of insulin granule membranes in beta islet cell electron micrographs

Quantification of sub cellular structures is necessary in understanding how cells function. This paper presents a segmentation algorithm for transmission electron microscopy images of insulin granule membranes from beta cells of rat islet of Langerhans. Granules are described as having a dense core and a surrounding halo. We use a mixed vector field convolution snake to segment the granule membranes. We also present a novel contribution to the convergence filter family, which uses an adjustable region of support. The filter is used to verify our segmentation. We calculate pixel error by comparing the membrane areas from our method with a manually defined ground truth. 1300 granules are used in our test and an average area difference of 7.54% is observed.