Embryo quality analysis from four dimensional microscopy images: A preliminary study

Automated selection of a healthy embryo is very important for improved success rate in the In-Vitro-Fertilization (IVF) treatment. Previous methods use morphological signatures from a short sequence of images from day-1, day-3 or day-5 embryos. However, the grading score based on short sequence of images may not produce consistent outcome. We therefore propose a method for assessing the quality of mouse embryos by analyzing long sequence of images up to the blastocyst stage. Two features viz. nuclear frequency and nuclear volumes are computed automatically from 4D time-series. A spatiotemporal adaptive technique is applied for counting nuclear centroids, while a centroid driven segmentation method is proposed for extracting nuclear volumes. A simple supervised classifier using normalized cross-correlation is then applied to discriminate healthy embryos based a proposed healthiness index measure. An experiment with ten sequences of embryo images, where five sequences for training and the rest five is for testing, shows promising performances of the proposed method.