Piotr Sankowski was quoted in an article from Polityka magazine about IVF procedure. For all our followers who do not speak Polish, we translated the text below! To read the original text, check out this link.
In Vitro Ex Machina. How can AI increase the chance of success in fertility treatment?
Translation of article by Paweł Walewski from Magazyn Polityka no 46 (3389), 9.11-15.11.2022. Original publication: https://www.polityka.pl/tygodnikpolityka/nauka/2188634,1,in-vitro-ex-machina-tu-tez-wkracza-sztuczna-inteligencja.read.
In many areas of healthcare and medicine, algorithms are already used to support doctor’s decision-making processes. We can not only use them to detect cancerous tissues based on medical images, but also patient’s medical history which includes heart attacks, epilepsy or a stroke. Reading a medical image or analysing a series of brainwaves is often much faster and more precise when done by a computer than by a, often exhausted, doctor.
A rapid technological progress allowed us to introduce innovations such as electro prosthetics, robots or virtual reality to multiple healthcare areas. It also opened an opportunity to use technology in fertility treatment. However, similarly to 44 years ago, when the first child was born with the help of IVF procedure, the biggest factor of successful fertility treatment is still biology and the patient’s age. For 25 year old women, the success rate of IVF can be as high as 85%, but this decreases up to 25% every 5 years. A biological statistic that cannot be changed even by the best tech specialists.
The matter of choice
“New technologies are not particularly useful in understanding biology. We firstly have to understand it on our own”, argues prof. Waldemar Kuczyński, one of the most experienced medical doctors in Poland specialising in fertility treatment. In 1987, he worked in the team of prof. Marian Szamatowicz in Białystok, where he supervised the birth of the first so-called “in-vitro child” in Poland. “Back then, we knew far less about natural processes leading to embryos’ creation and growth, we also weren’t aware about many problems. Even now, we still don’t have all of the answers”, he admits, based on his 25-years-long experience leading the Kriobank Center for Treating Infertility.
Increasing the success rate of fertility treatment must go hand-in-hand with patient’s safety. “Success rate is the price we pay for safety”, explains prof. Kuczyński. “If success rate was our only goal, we would implant as many embryos as possible in the womb, risking multiple pregnancy. This is not acceptable anymore – we can use only one embryo for women up to 35 years of age, and two embryos when she’s 35 and older.
This is why the core task to tackle is to assess the chance of success of embryos, both in terms of implantation and further growth. Such an assessment is based on analysis of medical images and videos and hence it is most suitable to be automated with the help of artificial intelligence. Why not teach computers to learn to monitor embryos and deduce what their chances are?
The matter of attention
Currently, all of those tasks are done by embryologists. Firstly, they have to fertilise the egg cell with a chosen sperm. The first approach is to allow the sperm to reach the egg cell on its own by imitating the natural environment in a Petri dish. If the sperm is not suitable for this procedure, the embryologist injects it directly into the cell. In the next step, the fertilised egg grows in an incubator which, again, imitates the natural environment with a proper temperature, carbon dioxide concentration and pH level. After 3-5 days, the embryo becomes a blastocyst and, if grown correctly, it is transferred to the womb.
How does an embryologist assess the success rate of an embryo? They check if the cell division is regular and if the sub-cells are symmetrical, how they look and what is the time between consecutive divisions. To do that, they use so-called embryo scopes, which make a video of an embryo’s growth and classify it. “Unfortunately, none of this can be used to automatically assess the growth potential of an embryo”, shares our expert. “Sometimes the embryos that do not have a high chance of success, in our opinion, can still result in a healthy pregnancy”.
The matter of collaboration
Piotr Sankowski, professor at Institute of Computer Science at University of Warsaw and co-founder of MIM Solutions, sees the potential for the use of artificial intelligence in the IVF procedures. MIM Solutions creates the algorithms supporting fertility clinics and competes with six similar startups in the world in this area. “In medical imaging and diagnostics we can provide the machine with a lot of images, and still we will find patients that will be too nuanced for artificial intelligence”, he says. The advantage of MIM Solutions and its system EMBRYOAID is that it doesn’t only analyse images, but also full movies from the embryo’s growth process. “It can also explain why the AI chooses a specific embryo, which is a value added to an embryologist from an educational point of view”, he explains.
However, for prof. Kuczyński, the use of AI in fertility treatment is crucial not only for choosing a specific embryo, but for understanding why some IVF treatments are more successful than others. “When the procedure results in a healthy pregnancy, no one analyses why it was successful. But when the process fails, we often can’t provide answers to patients and explain to them the reasons behind the failure.”
The ability to analyse 10-minutes-long video of an embryo frame by frame allows it to pinpoint things that a human could simply miss. This is also just one step in the whole IVF procedure – to produce a healthy pregnancy, all the elements must be done correctly. Prof. Kuczyński lists all the steps that need to be successful, such as hormonal therapy, time when the egg cells are collected, the fertilisation with a sperm and the final transfer. “Artificial intelligence can be 100% correct in assessing which embryo has the highest success rate. However, if a doctor who oversees the procedure makes a mistake, it will be for nothing,” he sums up.
This is why additional work is done to create a system for assessing ovarian reserve of a woman. This could be useful to better adjust times when she should do hormonal therapy or egg cells collection.
In order to improve AI systems, we can and should look for inspiration in the human brain, even though the algorithms themselves work completely differently than humans. Our brains integrate memory with the ability to analyse data, whereas in computers those areas are separate. Teaching the machine is somewhat similar to how the doctors learn to distinguish between healthy and ill areas in an image. During their course, the best experts can analyse hundreds of images with breast cancer, strokes or healthy and problematic embryos. On the other hand, the computer can process tens of thousands of examples in a mere couple of hours.
The matter of trust
But can we trust AI? Or maybe we expect too much from it? Those are still the open questions. Deep learning, which allows the creation of efficient algorithms for image analysis, is a relatively new research area. Unfortunately, this is also the area which heavily relies on computer science, meaning that the experts working in the field learn about healthcare and medicine together with the algorithms they create.
And what if we can never have the answers for the questions we’re asking? The same concerns were raised a couple of years ago when the first robots were introduced to surgery blocks. Nowadays, many surgeons use them despite the concerns about efficiency and safety compared to the traditional methods.
Prof. Grzegorz J. Nalepa from the Faculty of Physics, Astronomy and Computer Science at Jagiellonian University explains that AI still doesn’t have a good definition and it’s still being researched and formed. Prof. Nalepa is a laureate of Science Award of Polityka Magazine in 2012 and he made this comment during the 150th anniversary of Polish Academy of Learning. “As of now, we are building systems that work better than we expected and we want to understand why this is the case”, he points out. He simultaneously calmed us down, as, in his opinion, AI results in less damage to the world than a normal stupidity.