What is artificial intelligence?
Artificial intelligence is widely used and affects our daily lives even if we do not notice. From smart phone tools and personalised shopping experiences to assisted airline operations and medical diagnosis. When the appropriate method is chosen and trained on data that sufficiently represents the intended use, artificial intelligence makes our life easier and more comfortable.
The term artificial intelligence is used to describe computer-based systems or machines, which have been programmed to process and analyse information in an intelligent manner, akin to human reasoning. Artificial intelligence covers a broad range of methods that can be used for this purpose and often when we talk about artificial intelligence we use the analogy of neural networks. Such networks are designed to mimic the human brain which receives multiple inputs of information related to a specific outcome, thereby learning by connecting input and output. Multiple inputs and processing steps occur simultaneously depending on how the network is designed to generate an output. In that sense neural networks are designed as simplified mimics of human brains with neural nodes that signal to other neural nodes in multiple dimensions.
Artificial neural networks need to be trained on large amounts of data to process the type of information that is specific for the application. For instance, neural networks to process spoken words need to be trained for that purpose while networks designed to analyse images need to be trained on the type of images that will later be input for a certain application.
What is the potential of AI in ART?
The potential of artificial intelligence to improve processes in IVF is immense. For example, artificial intelligence can be used to automate repetitive tasks which may be prone to human-derived subjectivity and variability.
A perfect example is the process of embryo selection where embryos in a cohort need to be prioritised for transfer or cryopreservation according to their likelihood of implantation.
Currently, standard embryo evaluation is conducted in a protocolled fashion, gathering data and registering development features on a daily basis. However morphological assessment of embryos is known to be prone to inconsistency mainly due to human subjectivity as well as the limited data input available with gathering only a daily snapshot of information1. Research has shown that different embryologists may rank the very same embryo differently, and that even the same embryologist given the same embryo at two different time-points may offer differing output scores2. Using state-of-the-art time-lapse systems provides much more data and can alleviate inconsistencies and time needed to evaluate a cohort of embryos. However, estimates for viability are still based on manual annotations by the human eye when examining the embryo development videos and are therefore still somewhat subjective.
How is Vitrolife utilising AI in combination with time-lapse?
Vitrolife has worked with AI methods for a long time. From the early days of our time-lapse journey, we have employed advanced computer vision which resulted in the development of the Blastomere Activity Chart as an integrated part of the EmbryoViewer software. This chart was designed based on advanced computer vision and graphically depicts pixel-to-pixel changes from one time-lapse image to the next, guiding the user to find the likely timing of a kinetic change of the embryo.
Later on, using machine learning, large numbers of annotated embryos with clinical outcome data, were used to develop the KIDScore decision support tools (“KID” is short for Known Implantation Data). These are mathematical algorithms to support ranking of embryos monitored by time-lapse and correlating embryo development patterns to their implantation potential. KIDScore decision support tools are designed for use during the embryologists’ decision process of prioritising embryos before transfer. We have continued to strengthen these algorithms as more data has become available.
Guided Annotation is our most recent tool for the embryo assessment process and makes use of deep learning-based algorithms to automatically estimate cell division events, PN number and blastocyst morphology. This has resulted in a streamlined and efficient full embryo assessment workflow. The system was trained on more than 50.000, 13.000 and 8.500 time-lapse sequences to estimate with high accuracy PN number, division times through to blastocysts as well as ICM and TE grades, respectively. During the development of Guided Annotation it became clear that the highest precision of the AI-based algorithms was reached upon training the network with time-lapse sequences rather than single embryo images1. With this development, the Guided Annotation tool allows users to efficiently gather necessary data that can be used as direct input for embryo evaluation scoring systems, including our KIDScore decision support tools.
Development of new AI-based tools by Vitrolife
Vitrolife continues to be at the forefront of developing and implementing new technologies to improve clinical workflow and facilitate consistent procedures in the IVF laboratory. We continuously improve our products – and we have teams of experts for each product line we develop.
We have a dedicated team of software engineers, specialised in deep learning and neural networks for the development of new imaging analysis tools, which will bring time-lapse evaluation of embryos to the next level. Our statisticians thoroughly test and validate our automated analysis tools to ensure they are robust across multiple clinic profiles and embryo culture conditions.
Hereby you will be able to rank embryo cohorts directly based on automatic analysis of time-lapse sequences in our time-lapse platform. A new software tool, iDAScore®, has been developed with the use of deep learning and a neural network trained to analyse time-lapse sequences on more than 115.000 embryos with known clinical fate. iDAScore® takes the entire embryo development history into account to rank embryos according to likelihood of implantation and was developed based on IVY, which in clinical use showed with very promising results3.
The algorithm will integrate directly with the time-lapse platform without the need for manual image processing. For you as the user, it will mean a more efficient and consistent workflow. Embryo evaluation can be performed automatically allowing you to utilise your time more efficiently in other steps of the IVF process that are more complex and require more time and attention.
Such evaluations will support consistent embryo evaluation without any work required by the user except the click of a button.
What is the future of AI in IVF?
As computers become more powerful and as data sets increase in size, the ability of AI-based systems to perform more and more complex functions will increase.
AI-based tools should at a minimum, perform equally as well as top-of-class clinical staff to automate time-consuming processes. In the future, with more data inputs such as patient characteristics and medical history, AI could potentially outperform humans in clinical treatment procedures.
It is important that implementing smart tools does not compromise safety and efficiency in the clinic. The first steps on the way have been taken with iDAScore®, which integrates time-lapse monitoring and AI-based embryo evaluation in one workflow.
As clinics are faced with increasing patient demand, often requiring more and increasingly complex procedures, there is a clear need to automate as many processes as possible to ensure the best treatment quality while optimising clinic resources. Use of artificial intelligence-based systems are already being introduced into IVF clinics and show great potential for streamlining treatment decisions.
As always when applying new technologies, it is important that they are validated, and should be used as supportive tools rather than a replacement for humans until we are confident that they perform in a safe and consistent manner. To this end, properly designed randomised controlled trials (RCT) are the way forward.
iDAScore® is currently being tested in an international multicenter RCT.
Are you interested in learning more about future developments for embryo evaluation?
Get a glimpse into the future of embryo evaluation. Watch this recorded webinar ‘The future of AI methods to automate embryo evaluation’, from the 1st IVF Worldwide Online Congress. In this webinar Dr. Mikkel Fly Kragh presents new developments based on time-lapse and artificial intelligence.
Watch this recorded webinar 'AI in time-lapse - automatic grading of human blastocysts' where Dr Mikkel Fly Kragh goes through how he and colleagues have worked on integrating artificial intelligence into time-lapse and more specifically how they have automated blastocyst grading using AI.
In this recorded webinar 'Workflow efficiency related to iDAScore' Marcos Meseguer will go through the topics of introducing the use of Time-lapse technology in the laboratory and share his experience of applying Artificial intelligence annotation tools, KIDScore and iDASCore, in practice.
References
- Kragh, M.F., et al., Automatic grading of human blastocysts from time-lapse imaging. Comput Biol Med, 2019. 115: p. 103494.
- Adolfsson, E. and A.N. Andershed, Morphology vs morphokinetics: a retrospective comparison of inter-observer and intra-observer agreement between embryologists on blastocysts with known implantation outcome. JBRA Assist Reprod, 2018.
- Tran, D., et al., Deep learning as a predictive tool for fetal heart pregnancy following time-lapse incubation and blastocyst transfer. Hum Reprod, 2019. 34(6): p. 1011-1018.
Topics: Time-lapse