Automatic Methods for Human Embryo Component Extraction

In this thesis, we proposed two novel methods for blastomere extraction and trophectoderm segmentation in an attempt to aid physicians in determining embryo’s viability. Accurate assessment of embryo’s viability can play a vital role towards optimization of in-vitro fertilisation (IVF) treatment outcomes. The first proposed automatic method is developed to identify blastomeres in human embryo HMC (Hoffman Modulation Contrast) images of day-1 to day-2. Our algorithm applies isoperimetric graph partitioning, followed by a novel region merging algorithm to approximate blastomeres positions. Ellipsoidal models are then used to approximate the shape and the size of each blastomere. The proposed algorithm is evaluated on a dataset of 40 embryo images and it exhibits an average blastomere extraction accuracy of 80%. The second method segments Trophectoderm (TE) regions in embryos of day-5 (also known as blastocysts) by first eliminating the inhomogeneities of the blastocysts surface using Retinex theory. A level set algorithm is then used to segment TE regions. We have tested our method on a dataset of 85 images and have achieved a segmentation accuracy of 85% for grade A, 89% for grade B and 92% for grade C.