Assisted reproductive technologies (ARTs) are a set of invasive medi- cal techniques that attempt to induce a pregnancy. In vitro fertilization (IVF) is the most common and effective type of ART. Embryo selection is a difficult and complex task. There is a morphological evaluation crite- ria and a categorization into scales for each of the various embryo stages. From these results, clinicians have to select which embryos to transfer, as the clinical procedure can produce excess embryos. The transferred em- bryos have to be carefully selected among the ones that show best quality according to this morphological classification, as the aim of the process is to achieve a pregnancy. In this project, we present a novel probabilis- tic graphical model that, for the first time, accounts for the uncertainty that represents all the unknown factors that can drive to a failure even though all the components that take part in the ART process seem to be favorable. In an ARTs' dataset it is not always possible to know which embryo was implanted. Among others, this uncertainty source forces us to use an EM strategy, as well as the consideration of hidden variables in our model. The experiments carried out show that much more informa- tion can be obtained from this type of model than from previous simpler approaches. The database for this work have been collected by the Unit of Assisted Reproduction of the Hospital Donostia (Spain) throughout 18 months (January 2013¿July 2014) where 604 patients participated in the IVF-ICSI program compiling a total number of 3125 embryos.
