Automatic Classification of Human Embryo Microscope Images Based on LBP Feature - Advances in Image and Graphics Technologies - page 146

embryo to transfer is currently a hot research field for human reproductive [2-4].
The embryo morphology during the third day or the fifth day after fertilization can be
used to evaluate the implantation potential[5-7]. Fig. 1 shows the morphology of an
embryo during the third day. Fig. 1(a) shows an embryo that successfully becomes a
baby, and Fig. 1(b) shows an embryo that fails to develop into a baby. Usually, the
texture of successful pregnancy embryos is relatively regular for embryo images, and
the texture of the embryos that fail to be implanted is rather disorganized.
(a) an embryo leading to successful pregnancy (b) embryo that fails to give birth
Fig. 1 . Embryo images in the third day after fertilization
In this essay, we mainly focus on the embryo viability by exploiting their morpho-
logical features during the third day after fertilization, which can be considered as the
classification of embryos, and feature extraction of embryo images and classifier de-
sign are involved[8]. Özkaya[9] studied the impact of maternal age, the number of
eggs, the size of blastomeres and thickness of zona pellucida on embryo's classifica-
tion. Scott et al[10] used the nuclear morphology of embryos to do classification.
Hnida et al[11] made use of multi-level morphological analysis method to classify
embryos. However, these methods are difficult to extract and quantify feature accu-
rately. Due to the abundant texture of embryo images, it is expected to be more rea-
sonable to quantify the characteristics of embryo images. As a texture descriptor, the
LBP descriptor is effective for delineating the local texture feature of images. The
descriptor compares each point with its neighboring pixels, and the comparing result
is saved as a binary number. Due to its powerful discrimination and simple calcula-
tion, the LBP operator has been applied in fingerprint recognition, face recognition,
license plate recognition and other areas[12,13]. The most important advantage of the
LBP is its robustness to the changes of image intensities; this advantage makes the
LBP more suitable for embryo images taken by HMC (haffman modulation contrast).
Moreover, thanks to the simple calculation, this operator can be used for real-time
analysis. Therefore, we attempt to employ this operator to do feature extraction on
embryo images.
In addition, how to design the classifier is another key issue for the embryo image
evaluation[14]. The SVM (support vector machine) is a supervised learning method,
and it tries to establish a hyper-plane with the maximum interval in the feature