In Feb 2023 Dr Jonathan Hall presented at the Indian Society For Assisted Reproduction (I.S.A.R) the presentation “AI Enhanced Fertility”. A video of the presentation, the presentation slides and transcript are available below.
Hi everyone, Thanks for joining me. I’m Jonathan from Presagen.
Before we start, some disclosures and disclaimers.
I’m a cofounder of Presagen, which manufacturers Life Whisperer, a commercial AI tool that assesses embryo viability in IVF.
The research and clinical studies relating to this AI will be the topic of this presentation.
Life Whisperer AI tool itself is not authorized for use in all countries. The countries where it is authorized are listed here. Its is not for sale in the USA.
And Life Whisperer is only available for sale or use by healthcare professionals.
Today, I will be talking about how AI can enhance fertility, help improve outcomes for couples wanting to have children, and provide insights into embryo quality.
IVF is a crucial process for couples struggling with infertility.
It is a type of assisted reproductive technology.
In IVF, an egg is fertilised, and develops into an embryo.
The embryos are assessed in terms of their quality, and then transferred back to the patient, with the hope of a pregnancy or a live birth.
Artificial Intelligence is making a big difference in IVF.
AI can be used to look at human egg quality, sperm quality, and track developmental milestones.
AI can also look at the genetic integrity of a developed embryo, and predict the likelihood it will implant, or the likelihood of having a successful live birth.
In this talk, we are going to focus on the Embryo Evaluation and Selection process.
Using Computer Vision, Life Whisperer AI are able to assess embryo quality from a single image.
To describe the process, a blastocyst stage embryo is imaged on Day 5 after IVF. This is an endpoint analysis, and we are interested in the quality of the embryo immediately before transfer to the patient.
The image is then drag-and-dropped onto the Life Whisperer website.
A sequence of validated AI systems then run. Using object detection, and focusing on different regions of the image, there are two AI scores that are displayed.
Life Whisperer Viability gives a score from 0 to 10, which is related to the likelihood the embryo will lead to a clinical pregnancy. This means, a foetal heartbeat is measured at the 6 week ultrasound scan.
Life Whisperer Genetics gives a separate score from 0 to 10, which is related to the likelihood that the embryo has high genetic integrity, meaning it has the correct number of chromosomes, or is “euploid”. This technology is objective, is able to run instantly, and is completely non-invasive, unlike biopsies.
In fact, the Computer Vision AI are able to detect features in the embryo that may be difficult or even impossible for the human eye to detect, and outperform a person at embryo quality assessment.
Let us look at an example.
Before choosing an embryo, a person in the lab must look down a microscope, and assess the embryo quality.
By using Life Whisperer Viability, a simple drag-and-drop results in a viability score from 0 to 10, in this case, 9.1.
Each score tells you about the likelihood of a pregnancy. For low scores, there is less chance (but still some chance) of a pregnancy. This is useful for expectation setting and planning with the patient.
As the score increase, the chance of pregnancy increases. A high score of 9.1 out of 10 means a 65% chance of pregnancy – the highest chance possible without knowing more about patient factors, such as endometriosis.
Using Life Whisperer Genetics, a drag-and-drop gives a Euploid score from 0 to 10.
In this case, it was 9.5 out of 10.
Each embryo’s Euploid score tells you about the likelihood that the embryo is chromosomally normal – non-invasively, instantly, and purely through imaging.
Low scoring embryos have a much higher likelihood of aneuploidy – these embryos are riskier to transfer, carrying a higher likelihood of disease or miscarriage. This is useful for counselling the patient, planning for risks and potentially considering further genetic testing to learn more.
High scoring embryos, such as this one, carry a high chance of being euploid. (I won’t say ‘which could be used as a pre-screen and avoid costly and invasive genetic tests’ re FISI?)
Importantly, Life Whisperer Genetics may be able to account for Mosaicism, which is an effect where an embryo is partly euploid and partly aneuploid. The AI takes the whole morphology into account before presenting the overall score.
Life Whisperer AI have been tested in a range of highly-regarded international studies and published in several prestigious journals, including Nature Scientific reports, Oxford University Press ‘Human Reproduction’, and Reproductive BioMedicine Online.
It was found that Life Whisperer Viability score offers up to 25% increased accuracy in predicting pregnancy outcome compared to traditional methods. It can also rank order embryos to minimize the time to a first pregnancy by 12%, compared to manual grading methods, preserving morale that all-too-often can result in patients dropping out of their therapies.
Lastly, looking across many patients’ groups of embryos, the probability the top embryo is a euploid was found to be 82%.
But what is it, exactly, that the AI is seeing, to result in such performance improvements?
Each AI system is a series of linked, neural networks, which perform object detection, and then looks at different parts of the embryo all at the same time, before combining this knowledge together for a final score.
The AI not only takes into account the overall look of the embryo as a whole, but has shown to be correlated to known features that scientists have long-studied in embryos – the expansion grade, which can be seen by focusing on the egg shell, or zona pellucida region, and the quality of the Inner Cell Mass and Trophectoderm regions of the embryo, as marked.
What Life Whisperer AI are able to do is to go beyond the known features of the embryo, and more precisely identify the features relating to embryo quality.
In this table, the standard embryo grades are listed on the left column. There are only a few combinations.
However, within each category, Life Whisperer AI are both consistently able to provide a high Sensitivity or Recall accuracy.
This means that AI can distinguish embryo qualities that a skilled professional would grade the same.
Clearly, the AI is seeing more than the human eye.
Human grading can be fraught with bias, and is well known to be subjective – varying from person to person, day to day, and lab to lab.
In a recent survey, 158 Embryologists took the ‘Life Whisperer Viability AI Challenge’!
Their goal was to grade 20 embryos, using the best of their skills and knowledge, and see if they could beat the AI.
What we found was that embyrologists had a wide range of performances. There was little consistency in terms of the kinds of grades given to the embryos, with most people scoring around 50%.
Very few, however, were able to beat the AI. While no one got 100% (not even the AI can see everything from what we give it), the AI outperformed or did as well as 99% of the participants.
So if the AI is seeing more than the traditional features people look at, what new features is it seeing?
Life Whisperer AI can detect features both known and novel.
Consider a blastocyst that was graded a ‘4BA’ manually, and Life Whisperer Viability gave a score of 9.1 out of 10.
This embryo resulted in a clinical pregnancy.
Life Whisperer Viability uses a combination of 4 feature classification neural networks, and we can look at heatmaps showing the areas of relevance to viability in the image.
Here, gold represents areas of high relevance, and purple areas of low relevance.
We can see that there are different types of features that are deemed relevant to viability.
Overall, gross morphological shapes, and specific targeted areas, particularly around the egg shell, or zona pellucida region, are relevant.
We can also measure areas of relevance to non-viability in the image.
Here, bright represents areas of high relevance, and grey areas of low relevance.
A single image can have some regions that are correlated with high quality, and some with low quality, and the AI takes this all into account to give a statistically reliable score.
Here, the features that are correlated with non-viability are more sharply defined, and seem to involve the trophectoderm region of the embryo.
Life Whisperer Genetics, on the other hand, uses a combination of 3 feature classification neural networks, and can be used to map out areas of relevance to genetic integrity, in the image.
Here, yellow represents areas of high relevance, and blue areas of low relevance.
What we see is that, for this image that scored a 9.9 out of 10 Euploid score, a wide and distributed region was considered relevant in predicting the embryo was chromosomally normal.
On the other hand, we can look at the features that correspond to aneuploidy even though this embryo was likely to be a euploid.
This helps us understand what the AI might be looking at, and deeming relevant for assessing if there are any features that are correlated with lack of genetic integrity.
Here, red represents areas of high relevance, and dark purple areas of low relevance.
The areas of relevance here are more precise and targeted. The AI looks at regions around the embryo as well, for features that might have to do with the egg shell, or hatching.
While we have only begun to explore the features that the AI might be ‘seeing’, it is important to note that these findings are preliminary, and that the AI doesn’t see the way a human sees things.
Now, we will focus on synergies that can come from using multiple AI technologies together to improve outcomes.
New studies presented at the American Society for Reproductive Medicine annual conference in 2022 in Anaheim, California, showed how AI can add synergy in IVF.
On the left, it was explored how genetics can play a big part in failure to reach a live birth outcome.
By examining live birth data using Life Whisperer Genetics, it was found that the AI was highly predictive of Live Birth.
On the right, Life Whisperer Viability and Genetics were used together.
It was found that combined use can in some cases double the improvement in time-to-pregnancy, beyond using Life Whisperer Viability alone compared with manual methods.
Further AI synergies are about to be explored with Presagen’s AI Open Projects, where our collaborative network of clinics and hospitals work together to co-create new AI products for IVF and beyond.
By using our unique suite of AI technologies, our AI consistently demonstrates increased accuracy, and increased robustness during training.
This is because of our AI training system for removing errors from medical data, and for accessing distributed and diverse data using Federated Learning.
These methods have shown to increase AI performance dramatically, and were recently published in Nature Scientific Reports.
The newest addition in development is Life Whisperer Oocytes.
A new AI system for assessing human eggs, prior to IVF, can help select the ones most likely to result in good quality embryos, and help the patient make decisions about how many times they might need to go to a clinic for egg collection.
In summary, AI can be used to enhance fertility and assist in IVF in a number of ways.
Computer Vision AI is objective, non-invasive, and has improved performance compared to traditional embryo grading.
It can identify known and even unknown features embryos that impact on their quality.
Features that relate to viability, non-viability, euploidy and aneuploidy are all different and both are useful in embryo selection.
Combining multiple AI technologies together can lead to synergies across the whole IVF workflow.
I would like to thank you for joining this webinar.
This presentation is available on our websites at Presagen.com and LifeWhisperer.com