ABSTRACT
In-Vitro Fertilization (IVF) is an infertility procedure where sperm and eggs are inseminated outside the body (“in-vitro”). One of the main factors influencing the success rate of the IVF is the quality of the embryo. In fact, one-thirds of the implantation failures are caused by embryo grading issues. The tedious grading task is usually performed by an expert, which may result in inaccuracy result depending on his/her experience. In this chapter, we propose an automatic embryo grading procedure using the convolutional neural network (CNN). The performance of the CNN is often supported by the preprocessing method which extracts the features of the input image. In this work, we focus on image processing filters as a preprocessing method and develop a joint optimization method of the image processing filter selection and weights of the CNN using gradient-based method. In particular, the image processing filters are selected according to a binary vector drawn from the Bernoulli distribution whose parameter is optimized by a stochastic natural gradient method. The simulation result shows that the proposed method results in 8% reduction in the test error compared with the CNN without image processing filters. A deep analysis of the proposed system will also be presented.
