Semi-automated segmentation of pollen grains in microscopic images: a tool for three imaging modes

Abstract

Image analysis methods have the potential to increase the accuracy and rates of data collection in palynological research. Automated segmentation of pollen grains is a method that would facilitate image-based palynological analysis by creating large reference image libraries. We developed an executable for the automated segmentation and cropping of pollen grains from microscope images based on pixel intensity values. Our method works with images taken using transmitted-light, widefield-fluorescence, structured illumination (Apotome), and includes a novel approach for cropping the Apotome Z-stack. The system crops pollen grains from sampled fields of view with ~97% recall and ~97% precision for transmitted-light and widefield-fluorescence images, and ~90% recall and ~89% precision for Apotome fluorescence images. Results differed between different imaging wavelengths for fluorescence images, with Apotome images showing the greatest difference between red and green emission wavelengths. Recall in cropping of transmitted-light images was comparable to previous segmentation efforts.