Search for a command to run...
The precise assessment of embryo viability is an extremely important factor for the optimization of IVF treatments. In order to assess embryo viability, several embryo scoring systems have been developed. However, they rely mostly on a subjective visual analysis of embryo morphological features and thus are subject to inter- and intra-observer variation. In this paper, we propose a method for image segmentation (the dividing of an image into its meaningful constituent regions) and classification of human blastocyst images with the aim of automating embryo grading. The delineation of the boundaries (segmentation) of the zona pellucida, trophectoderm (TE) and inner cell mass (ICM) were performed using advanced image analysis techniques (level set, phase congruency and fitting of ellipse methods). The fractal dimension and mean thickness of TE and ICM image texture descriptors (texture spectrum and grey-level run lengths) were calculated to characterize the main morphological features of the blastocyst with the aim of automatic grading using Support Vector Machine classifiers. The fractal dimension calculated from the delineated TE boundary provided a good indication of cell number (presented a 0.81 Pearson correlation coefficient with the number of cells), a feature closely associated with blastocyst quality. The classifiers showed different accuracy levels for each grade. They presented accuracy ranges from 0.67 to 0.92 for the embryo development classification, 0.67–0.82 for the ICM classification and 0.53–0.92 for the TE classification. The value 0.92 was the highest accuracy achieved in the tests with 73 blastocysts. Semi-automatic grading of human blastocysts by a computer is feasible and may offer a more precise comparison of embryos, reducing subjectivity and allowing embryos with apparently identical morphological scores to be distinguished.