A Two-Phase Segmentation Method For Drosophila Olfactory Glomeruli

In order to understand olfactory coding and functional connectome within the fly brain, scientists need not only a 3D stereotypical brain atlas but also anatomical local landmarks to describe hard-wiring circuits of olfactory neurons. Olfactory glomeruli are such indispensable local landmarks; however, it is hard to segment them from confocal microscopy images. We propose in this paper a systematic approach for semi-automatic olfactory glomerulus segmentation. This method consists of two phases. The former phase aims to highlight within-glomerulus regions, appraise the amount of recognizable glomeruli in each z-slice, and to suggest a seed used to generate an initial contour for every recognizable glomerulus. The latter phase, starting from the initial contour of each recognizable glomerulus, propagates the initial contour information to adjacent slices recursively until the glomerulus' contours on all z-slices are obtained. Experiment results shows that our results are similar to those of manual segmentation. The proposed strategy is effective.