Lighting-aware segmentation of microscopy images for in vitro fertilization

Abstract

We present a practical graph-based algorithm for segmenting circular-shaped structures from Hoffman Modulation Contrast images of human zygotes. Hoffman Modulation Contrast is routinely used during In Vitro Fertilization procedures, and produces images with a sidelit, 3D-like appearance; our algorithm takes advantage of such peculiar appearance in order to improve the robustness of segmentation. The task is not straightforward due to the complex appearance of the objects of interest, whose image is frequently affected by defocus, clutter, debris and other artifacts. We show applications of our technique to the unsupervised segmentation of the zygote oolemma and to the subsequent supervised segmentation of its pronuclei. Experiments are provided on a number of images with different characteristics, which confirm the algorithm's robustness with respect to clutter, noise and overexposure. © 2009 Springer-Verlag.