Artificial intelligence - a novel tool in assisted reproduction technology

The fertility rates have declined in many industrialized countries during the last decades. This can partly be explained by social and economic conditions, but may also be due to a rise in fertility problems. Between 10 and 15% of couples are involuntarily childless. Over the last decades, there has been a development of assisted reproduction technology, and the use of such treatment, also called in vitro fertilization (IVF), is increasing. Intracytoplasmic sperm injection (ICSI) is a treatment for couples where the male has reduced semen quality, although sometimes also used when the semen quality is normal. By this method, the sperm is injected directly into the egg cell, whereas in ordinary IVF, sperm and egg are mixed in a dish, and one of the sperm cells fertilizes the egg. The decision about which fertilized egg, called embryo, should be transferred to the woman is based on the appearance of the embryo and how it develops during the first few days after fertilization. Likewise, the sperm used for ICSI is selected upon examination of some characteristics, like the movement and appearance. The evaluation of embryo and sperm is performed by embryologists, but there are no clear criteria for prediction of pregnancy. Therefore, information important for achieving pregnancy may not be recognized by this subjective assessment only. Artificial intelligence (AI) methods are more and more accepted as an important tool in medicine and are especially suitable to obtain information from images. In this project, we develop AI methods to analyze videos of embryo development. We have shown that deep learning models are able to characterize cell division and different stages of the embryo development. Data from these image analyses will be coupled to reproductive outcomes, like pregnancy and live birth, to elucidate if specific features in the development of the embryo increase the probability for treatment success. We have also developed neural networks which by analyzing videos can categorize the spermatozoa according to their motility. Such models may contribute to improve the semen analysis, which is important in the investigation of infertility. Videos of spermatozoa prior to selection for injection into the oocyte by the ICSI method are presently being analyzed and will thereafter be related to reproductive outcomes. Both the spermatozoa and the ICSI procedure are evaluated. The results will be used to make a tool for fertility clinics to assist in clinical decisions. The goal is to improve the methods for embryo and sperm selection and thereby increase the chance of pregnancy and ultimately a live born child. Another advantage may be to reduce the number of treatment cycles and to lower cost per treatment.

The fertility rates have declined in many industrialized countries during the last decades. This can partly be explained by socioeconomical factors, but may also be due to biologically related matters. Around 15% of couples will encounter fertility problems. Over the last decades, there has been a development of assisted reproduction technology (ART), and the use of ART treatment is increasing. The method of intracytoplasmic sperm injection (ICSI) was originally a treatment for couples where the male has reduced semen quality, but is often used also when the semen characteristics are normal. ART is to a high degree based on subjective assessments of spermatozoa and embryos, utilizing only a limited set of information. This project aims to develop strategies for making the selection of embryo and spermatozoa based on more objective criteria. AI methods make it possible to analyse large amounts of data from imaging and cell biological examinations to uncover patterns applicable in developing new methods for assessments of spermatozoa and embryos. By relating these patterns to ART outcome, the assessments will be optimized to improve the treatment results. This project also aims to develop tools for clinicians and embryologist to make more evidence-based decisions and thereby improving the ART outcome. AI methods constitute a new approach compared with traditional statistical methods, which are not equipped to reveal nonlinearities and complex relations between factors. A challenge using AI models is that they are more complex to interpret than traditional statistical models. Although the models may fit better to the data, it may be challenging to make them generalizable. Furthermore, the analysis may be difficult to understand by the users, but we will comply with this through a unique support system. Potential impact of the project findings is to reduce the number of treatment cycles and to lower cost per treatment.