Automatic Detection of Embryo Using Particle Swarm Optimization based Hough Transform Ikhsanul Habibie ∗ , Anom Bowolaksono † , Robeth Rahmatullah ∗ , Muhammad Nanda Kurniawan ∗ , Muhammad Iqbal Tawakal ∗ , I Putu Satwika ∗ , Petrus Mursanto ∗ , Wisnu Jatmiko ∗ , Adi Nurhadiyatna ∗ , Budi Wiweko ‡ , and Adi Wibowo § ∗ Faculty of Computer Science, Universitas Indonesia † Faculty of Mathematics and Natural Science, Universitas Indonesia ‡ Faculty of Medicine, Universitas Indonesia § Department of Micro-Nano Systems Engineering, Nagoya University Kampus Baru UI, Depok, 16424 Indonesia Abstract—In-Vitro Fertilization (IVF) is a procedure to obtain embryo by inseminating oocyte and sperm outside human body. Several embryos are produced at the end of this procedure and it remains a problem to select the most appropriate embryo to be implanted into uterus. Many strategies have been proposed for selection of the embryo. The latest is time-lapse microscopy which monitors the embryo development continuously. An automatic method using computer to detect and locate the position of the embryo is thus needed. In this paper, an approach based on a modification of Hough Transform using Particle Swarm Optimization (PSO) is proposed to approximate the embryo as a circle. Each PSO particle represents a circle in the parameter space and mainly used to reduce the computational complexity of Hough Transform. Experiment result showed that the proposed method is able to detect the position of the embryo accurately. The result from this method can be used to extract criteria for embryo transfer purpose. 1. I NTRODUCTION In-Vitro Fertilization (IVF) is one of the methods to treat infertility by inseminating oocyte with sperm outside human body, thereby eliminating several factors which causing infer- tility in the first place such as low number of sperm in male ejaculation fluid or obstruction in the female fallopian tube. Several oocytes are fertilized and cultured simultaneously and at the end of the program, one or more embryos are picked to be transferred back into uterus. One of the major concerns of the IVF program is to choose the most viable embryo, the embryo with the highest quality, that are most likely to give clinical pregnancy [1]. Many strategies have been proposed for the selection of viable embryos in human assisted reproduction. Several criteria from different approaches have been reported in the literature. Wharf et al. [2] reported that early cleavage division at 2- cell stage is able to identify embryos with high implantation potential. Balaban et al. [3] used pronuclear morphology to predict subsequent embryo development. A different scoring at the blastocyst stage with reduced risk of aneuploidy has also been reported by Borini et al. [4]. Another research by Rienzi et al. [5] combined the previously proposed scheme into a cumulative embryo scoring system and reported to have a high rate of successful implantation. Computer based approach has also been reported in the literature. Morales et al. [6] used different families of Bayesian classifier and compared their viability in selecting best embryos using central moments of embryo image as its descriptors. It is reported that tree augmented naive Bayes (TAN) performed the best. Manna et al. [7] proposed the use of neural network ensemble with local binary pattern as its features to discriminate viable embryo. However, in their researches, the features are extracted manually. An automated system surely would benefit as it can be installed right in the incubator, and choose the embryo automatically. An automated system has also been proposed in similar but different application. Ning et al. [8] proposed a machine learn- ing approach using convolutional neural network and energy based model to classify C. elegans embryos. Another research by Jiang et al. [9] used Scale Invariant Feature Transform (SIFT) to track the motion of cell if several cells are cultured together in the same petri dish. Recent technology which utilize microfluidic chip test the mechanical impedance to measure the age of oocyte [10]. A new approach based on time-lapse microscopy is cur- rently an active research area. Its viability for selecting embryo with implantation potential has been reported in the publication, for example, by Meseguer et al. [11] and Chamayou et al. [12]. Time-lapse observation offered attractive advantages compared to traditional observation because it can pinpoint the exact timing of embryo vital development. A new descriptor derived from this observation, morphokinetic feature, is then can be used to augment previously proposed parameters and improve the embryo selection accuracy.