How to use AI for IVF
The first IVF baby, Louise Brown, was almost ten years in the making. Back in the 1970s, embryologists and clinics couldn’t possibly envision how much IVF would evolve and its impact on the fertility care industry and millions of families.
Worldwide, an estimated one in six people of reproductive age experience infertility (approximately 17% of the adult population). But it’s not just infertility that makes IVF such a popular solution. Women are getting married and having their first child later in life, when the success rates for conceiving are notoriously lower, and they may have a greater availability of funds for IVF. Equally, some women may want to start a family on their own.
Regardless of what makes individuals pursue IVF treatment, many clinics and embryologists are unprepared to handle this rising demand. From limited resources to reliance on manual processes, clinics face challenges that inevitably impact the quality of care they offer their patients.
So, what needs to change in fertility treatments?
IVF today faces many challenges. In a world where ‘innovation’ is not only valued but expected, we also expect positive changes in IVF. Take, for example, the pregnancy rate per cycle. According to data published by the UK’s Human Fertilisation & Embryology Authority, the IVF success rate has remained stable but has been at a low 32% since 2014. That’s nearly ten years without improvement, during which women may need to go through three or even five cycles to become pregnant with a viable embryo.
If we look at the IVF process, it is easy to understand this plateau. Before the embryo is implanted, the embryologist must evaluate its development probability. The embryologists who work in this field are specialized, but there are not enough embryologists with this specific skill set to meet the ever-growing need.
In addition to the shortage of specialized embryologists, most IVF clinics still use very time-consuming, manual processes such as morphological assessment and manual grading to choose the embryo. These methods are subject to human error and selection bias. Clearly, it is time for innovative change in IVF.
Is AI the missing piece in the IVF evolution?
With the increasing number of women seeking IVF, the pressure to improve success rates is higher than ever. Why choose a lab with a success rate of 32% when there are clinics that do better?
Traditionally, in the IVF lab, embryologists use the ‘Three time points’ paradigm for embryo assessment. These three time points are the fertilization check, embryo quality check on day 3, and the blastocyst grade. The embryologist alone assesses the embryo’s quality (euploid or aneuploid).
While IVF clinics employ embryologists with the unique skill set to evaluate the embryo, research shows that even the most experienced embryologist is subject to decision bias when choosing the embryo.
Another facet of traditional IVF is the current status quo for preimplantation evaluation, the genetic diagnostic test, PGT-A. While PGT-A testing does have its advantages, it is an invasive procedure, and like any invasive procedure, there are risks.
The embryo may not survive the biopsy, there is the risk of cryogenic damage, there is the risk of false positive/false negative results, and it is not suitable for patients with a history of embryo damage or those who undergo fresh transfer. Additionally, PGT-A is a costly procedure, and because there is significant turn-around time, the amount of time to pregnancy is increased.
How to use AI for IVF
There are multiple use cases for artificial intelligence in embryology. AI is excellent for data processing, providing objective analysis, and discerning things the human eye cannot. Moreover, AI can evaluate embryos in a fraction of the time it takes in the traditional IVF lab. Let’s break it down.
Higher accuracy in early-stage testing
In the traditional IVF lab, sans AI, the embryologist literally has the weight of someone’s world on their shoulders. For a pregnancy to happen, the embryologist must pick out the most viable embryos out of 10 or 12 (sometimes less) that may look very similar. This task inevitably involves a certain level of subjectivity, which can lead to human error and varied results. Enter computer vision, and that weight translates into confidence.
While no one can deny the vital role of the experienced embryologist, a recent study demonstrated that computer vision can allow non-invasive differentiation between euploid and aneuploid embryos during the high-intensity stage of early blastulation. While these differences are too slight for the human eye to discern, AI can benefit embryologists by choosing embryos for transfer and implantation. With AI, the embryologists’ decisions are backed by objective analysis, saving valuable time, reducing stress and improivng patient outcomes.
Clinic-patient relationship
For many people undergoing IVF in a traditional clinic, the process is a mystery. They go into the clinic and meet the specialist but are rarely part of the process. Plus, they may not know why the embryologists choose ‘the embryo.’
With AI, the patient can be included in the process and even see the chosen and discarded embryos. The patient’s involvement in the process can strengthen clinic-patient communications and perhaps alleviate some of the anxiety that comes with the unknown. Transparency and timely information are ideal for fostering confidence, well-being, and control.
Another option for invasive tests
For patients for whom the invasive genetic diagnostic test, PGT-A, is not an option or who don’t want the added expense and risk of PGT-A, a non-invasive AI screening test can be the ideal solution.
A screening test can detect potentially abnormal embryos while minimizing unclear or unknown results. Instead of taking an embryo biopsy, the screening test utilizes advanced Time-Lapse Imaging (TLI) systems combined with cutting-edge AI algorithms. Intended to characterize genetic risk, screening test technology like AIVF Genetics, offered by AIVF, provides immediate, accurate evaluations of the embryos’ genetic integrity in real-time.
Based on a robust, multicentric, heterogeneous dataset from established data on live births plus results from PGT-A outcomes, AIVF Genetics was trained to identify the crucial embryonic features most significant to genetic integrity while minimizing unclear or unknown results. The tool produces a ranking scale (0-99) reflecting probabilistic estimates of each embryo’s euploidy (genetic normalcy).
Continuous monitoring
A distinct benefit of AI in the IVF lab is that AI doesn’t go home at the end of the day. Embryo assessments are an ongoing process, requiring 24/7 attention. Continuous monitoring means that conditions in the incubator, such as temperature and humidity/stability, can be monitored and maintained to ensure the most stable conditions for optimal outcomes. Continuous monitoring, visual inspections, and the evaluation process can occur on time, regardless of the hour or the workload of the embryologists.
Integrated communications
AI technologies can make embryo screening faster, more accurate, and more seamless for patients and embryologists. But this is just one piece of the puzzle. Another critical component is data governance, storage, and secure communications so clinicians can make the most informed patient decisions.
Integrated communications and a central data storage, sharing, and messaging hub seem to be the most logical next steps for clinics worldwide—and AI makes it all possible. For data security reasons, all third-party solutions should comply with relevant standards and regulations, particularly those concerning healthcare cybersecurity.
AIVF’s Communication and Integrated Messaging Dashboard seamlessly connects with time-lapse incubators and EMR data for data sharing. It allows team members to instantly and safely message colleagues and share reports, optimizing patient care.
Where to from here?
Here we are, 46 years after the birth of the first IVF baby, witnessing how AI transforms the IVF journey for patients, embryologists, and reproductive medical experts.
In a recent retrospective review of 20 articles, M Salih et al. took it upon themselves to look at the accuracy of predicting clinical pregnancy in IVF by AI versus clinical embryologists. For predicting clinical pregnancy using patient treatment information, the AI models had a median accuracy of 77.8%, while embryologists had a median accuracy of 64%. When TLI was added to the clinical information, the AI’s median accuracy was 81.5%, while embryologists’ median accuracy was 51%.3
When you think of the growing demand for IVF and the stagnant, pre-AI success rate of 32%, with women enduring 3 to 5 cycles of IVF treatment before becoming pregnant, what do you think patients would choose? A clinic with AI incorporated into its IVF treatment program, or one that relies solely on the experience of a dwindling number of embryologists? AI is an easy choice.
If you are interested in seeing AIVF in action, book a demo today.
- Submitted to ASRM 2024 (official reference available after 10/2024)
- Presented with Embryolab: IFFS World Congress 2023ESHRE 2023
- M Salih, C Austin, R R Warty, C Tiktin, D L Rolnik, M Momeni, H Rezatofighi, S Reddy, V Smith, B Vollenhoven, F Horta, Embryo selection through artificial intelligence versus embryologists: a systematic review, Human Reproduction Open, Volume 2023, Issue 3, 2023, hoad031, https://doi.org/10.1093/hropen/hoad031