Automated Insulin Granule Segmentation from Electron Photomicrographs of Rat Pancreatic β-Cells

摘要

Increased blood glucose stimulates pancreatic β-cells and induces an exocytotic release of insulin. The β-cell, whichcontains ~10^4 insulin-containing granules, releases only a few percent of the granules during a given stimulus such as ameal. The temporal response function to a square wave increase in the concentration of glucose is characteristicallybiphasic. It is not known whether the granules exhibit random or directed migration patterns as a function of phase.Directed migration would suggest the development of an intracellular gradient directing the path and velocity of insulingranule movement. Our ongoing research investigates this process using manual morphometric analysis of electronmicrographs of rat pancreatic β-cells. This is a tedious and time-consuming stereological process. Consequently, wehave developed an automated algorithm for accurately segmenting and deriving granule counts, areas, and measuringdistance to the plasma membrane. The method is a data-driven image processing approach that implementsMahalanobis classifiers to hierarchically classify pixel candidates and subsequently pixel aggregates as insulin granules.Granule cores and halos are classified independently and fused by intersecting the convex difference of granule haloswith core candidates. Once fused, total and individual granule areas and distance metrics to the β-cell plasmamembrane are obtained. This algorithm provides a rapid and accurate method for the determination of granulenumbers, location, and potential gradients in the pancreatic β-cell under different experimental conditions.