Automatic segmentation of scanning electron microscopy images for molecular aggregation profiling

Scanning electron microscopy (SEM) imaging has made it possible to obtain accurate inferences about functional intra-cellular events by the indirect observation of cell surface receptors' aggregation. Such observations are made with the use of colloidal gold particles that bind to the surface of specific ligands that cross-link to the desired receptors. Since the reliability of statistical inference methods about intra-cellular activities based on the observation of the receptors depend on the quality of the data set in hand, we propose in this paper a method to automatically segment gold labels in SEM images. This method includes an automatic thresholding operation aiming at simultaneously minimizing probabilities of false alarm and missed detection errors. Morphological filtering is used to deal with the partial superposition cases of structures of interest and additional median filtering is used for noise reduction purposes. We validate the method against a set of manually processed images of rat basophilic leukemia cells and we present results for the automatic estimation of aggregation density.