I was thinking if this was the scenario some 10 years back, people would have thought AI in reproductive medicine means artificial insemination in reproductive medicine. But we have come a long way. IVF is a very young science, but it is the fastest growing science medical faculty. And within 40 years, it has become a $40 billion industry now. So it's a huge market for AI. I would like to focus on those areas where AI is going to play a pivotal role and decisive role in reproductive medicine and what the IVF in future would look like.
Artificial intelligence is a strong technological way of providing the ability for a machine to perform the human brain function, like perceiving, reasoning, learning, and interacting. Broadly defined, AI here refers to machine making human decisions and tasks. And machine learning is a subset of AI technology which uses statistical method to draw a conclusion or prediction model.
Take care of the time. Because AI is zero without data. So, as data accessibility improves, insights gained may lead to decision support tools and could guide the doctor or the patient whether to continue the IVF cycle or to cancel it.
Deep learning is now a different subset of machine learning. As a subset of artificial intelligence, it enables computers to identify patterns in large, complicated data sets. Large amounts of data are too much for even the CPU to handle. A GPU is necessary. After that, it forecasts.
Why ART?
You can see limited success. We are stuck around 30 percent.
• The LIBOR threat. Reliance on human expertise
• Lack of personalisation
• High-cost ART
• Invasive procedure.
And there are some ethical concerns. And with significant technological advancement, the success rate of IVF has notably increased. But implantation rate without PGT, it is around 50 percent. And with PGT it is around 60 percent. But as far as LIBOR threat is concerned, it's typically hovering around 30, 32 percent, indicating the challenges that persist despite advancement in IVF technology.
Why ART is required?
Fellowship in Reproductive Medicine in India is a complex, multi-phase process that uses a variety of resources, but also has drawbacks, including high inter- and intra-observer variability and labor and time requirements. These difficulties affect ART's efficiency and reproducibility. That issue can be resolved by AI. AI could reduce the amount of effort that embryologists and physicians do.
AI technologies are fast and uniform across all IVF labs. AI lowers the possibility of human error while also ensuring best practices and outcomes. You can now observe how we use reproductive medicine.
We obtain the information from hospital data and computerized medical records. Next are natural language processing and machine learning. Additionally, we employ reproductive medicine in research, experimentation, and clinical practice.
In clinical practice, we are using it for sperm cells, quality detection, oocyte, embryo, then cost effectiveness also. With machine learning, we are using only supervised, mainly supervised and unsupervised.
Why AI in ART? With rapid pace of computer and genomic science, the future of reproductive science is likely to be a personalised digital fingerprint or digital embedded in patient's medical records. With regards to pharmacogenomics, the genetic heterogeneity between each IVF patient is an opportunity to tailor IVF treatment specifically for that individual because we'll have the DNA fingerprinting.
For example, FSH receptor polymorphism, we know there are three varieties. Someone needs more dosage of FSH, but in some cases it requires less.
So that targeted treatment we can do. Startup companies in Israel, such as Fertility, Embryonics, and Alive in California are developing AI systems aimed at clinics which can integrate IVF workflow and offer end-to-end optimisation and cost savings balanced with the promise of personalisation of the treatment. We are striving for a personalised medicine.
Again, workflow optimisation. If someone comes to an IVF lab, if some five, six cases are going on, in one day, you see that the sites are coming here, then some consultant has to do some embryo transfer, then some thawing has to be done. It's really a powerhouse.
How to optimise the whole work schedule?
• It optimises the scheduling of the predicting peak times, when exactly it will be busy. Accordingly, you can arrange your staffs. In AI, we start with outpatient. We'll be using AI in outpatient and before. The advances in natural language processing may lead to, in the future, to the analysis of the clinical notes and ability to prepare reports, engage in conversational AI chatbot for common questions.
• I mean, there are lots of questions they ask, and that can be solved. Some, because they get a lot of information and disinformation from internet, so just to confirm that they will give a call. So, AI will solve that problem. And they come with a bunch of papers. So how many cycles she had. Some clomiphene or gonadotropin, how many failed cycles she had. So this will be dealt with AI. AI will prepare the reports. And that's solving the problem.
• For the clinician, time will be much less. And in the same time, she or he will be able to see more patients. AI chatbot responses may provide answers to simple questions that would reduce the time requirement for the clinic staff, make the clinic available 24 hours a day.
• They are very insecure, infertile couples. They can phone even at 12 o'clock. But consultant will never pick up. So AI can solve that problem. And they are satisfied that clinic is available 24 hours a day. AI feedback and mood tracking, very much defaulted about the emotional part.
• Capabilities can be captured in real time. AI deep learning models have been shown to sense and respond to emotion. There will be social robo.
• They will find out who is depressed, sitting in OPD, or who is crying. They will solve the problem.
• Whether they have the need in one-to-one counselling or group counselling, that will be solved by AI. And automatic speech recognition and real-time language translation with capacity for speech to text translation may reduce language barrier. If any patient from Afghan or Arabic country, you don't need a translator.
Now, we have recruited patient. Now we'll start the IVF. How much of gonadotropin we'll be using.
Normally what we're doing, we take care of the age BMI, AFC, means answer follicle count and image level. And it's 150 to 450 units. However, the potential for follicle recruitment decreases after about 8 days. Suppose we find after 7-8 days follicles are not growing, then even after pumping in more gonadotropin, it's not going to respond. So the critical nature of the starting FSH dose is very important. AI is going to guide us there.
Several machine learning models harnessing historical clinical data are being advanced to streamline and selection of the initial FSH dose for IVF. Providing a standardised framework to enhance the personalisation of treatment protocol, mitigating the variability. Sometimes there will be suboptimal response.
Sometimes there will be over response. So AI will take care of that. Exactly what amount of gonadotropin we have to start.
This machine learning model by Fenton et al, the personalised dose response profile using variables. They have taken care of age, BMI, AMH, and answer follicle count. Could indeed customise the starting dose to enhance IVF outcomes and reduce the overall gonadotropin consumption by 195 units with 1.5 more M to mature sites.
One way reducing the dose and at the same time we are getting more mature sites. Showing the potential of AI to personalise and improve fertility treatment protocol. It's not only the starting dose.
Real-time tailoring, I mean real-time titration. It can do how much? Do I have to reduce? AI will guide. Do I have to increase? AI is guiding.
Do I have to use LH component now? AI will guide. Now when I am giving gonadotropin injection, follicle will grow. Now how to monitor those follicles? Automatic measurement of the follicular diameter. Normally what we do, we take the diameter. To biggest diameter we take the average. But here we'll do the volumetric assessment.
Automatic volumetric assessment. So, using follicular volume obtained with automatic measurements of follicular volume as a measure of follicular growth combined with the volume-based criteria for HCG triggering improves the treatment outcome compared to that achieved with conventional monitoring of the follicular diameter. So, we are doing volumetric assessment, I mean follicular scanning.
Now this is one of the pictures, 3D. Now that follicular scan, what it has shown, AI ERT software can analyse in detail all information by automatic measurements of the follicular volume with 3D ultrasound to monitor the cycle and HCG trigger. And there was only 3.27% discordance between the clinician's finding and AI finding.
It's highly reliable and automate the routine task in IVF centre. Now the follicles are growing. Now our next job is when to give HCG. That is also another part. You know we follow the traditional rule of thumb that when three follicles are 17 mm or more, then we give HCG. That means three follicles is working as a surrogate for the whole cohort of follicles.
Actually the optimal timing of trigger lacks evidence-based consensus and lends itself to AI solution. What AI is doing? Utilising a deep learning segmentation model from 3D ultrasound data, Liang et al. pinpointed the follicle volume thresholds that more accurately predict oocyte maturity and optimising HCG timing.
The use of the model increases yield by two to three more oocytes. The timing is guiding and we are getting at the end two to three more mature oocytes. Another AI model examining 19,000 patients across UK and Poland in their first IVF cycle, they found out the follicle size between 12 to 20 contributes most of the number of oocytes retrieved. And follicle size between 13 to 18 contributes most of the number of mature oocytes. And when we bring it to the lab, there we found out that follicle between 13 to 18 mm contributes most to the number of 2PN, which is fertilised oocytes. And follicle size between 14 to 20 were most important for high-quality blastocyst.
It's guiding that we should trigger much earlier. Of course, we'll follow AI only later on. Now the retrieval has been done. The oocyte has come to the lab. Now we ask the husband to provide a sample. When the sample comes, the question would be the sperm quality. The research application of AI system to sperm evaluation, sorting, and selection has generally been more advanced than that of oocytes, with many using computer-assisted sperm because we have been using already CASA, Computerised Assisted Sperm Analysis. Moreover, unlike oocytes, visually identifiable parameters such as sperm motility and morphology have been shown to reflect DNA integrity. AI can tell whether it has a good DNA integrity by looking at the motility pattern.
As a result, AI system have been developed to automate this evaluation to save time and avoid subjectivity and variability inherent in manual assessment. Sperm Selection, Goodson et al., they have developed a support vector machine model that reported motility characteristics with 90% accuracy. Means that motility actually is linked with the DNA integrity.
Mendizabal Rui et al. used a proprietary computer vision system to select individual spermatozoa for XE based on kinetic linear velocity and lateral movement of the head, and that correlates with the fertilisation and elastosis formation. Sato et al. developed a CNN-based system to simultaneously perform morphological assessment and tracking of sperm in real time to assist in sperm injection. Means, during XE, we look at the microscope that has a 400 magnification, but we cannot actually make out how good is the sperm quality-wise. So for that, then again, we started doing MC with a 6,500 magnification. That is very expensive. Sometimes even an embryologist needs half an hour to select one sperm. But here, real time, it will guide us, this is the sperm you catch and inject.
Now, those sites, normally after pickup, we keep it in an incubator. They will take some rest. They are tired. Then we take them out for denuding because we can't really say anything about the quality unless the cumulus cells are removed. So we just checked the site quality by looking at the zona and the polar body and the cytoplasm and the perivitelline space.
But AI algorithm with advanced image analysis can assess oocyte's homogeneity, plasm's homogeneity, zona's integrity, and the presence of cytoplasmic inclusion of vacuoles, all of which are important indicators of oocyte quality, which are not discernible by manual evaluation. The embryologist can't say much about that. So oocyte selected, sperm selected.
Now we get the embryo. Now, how we see the quality of the embryo. In clinical embryology, we don't yet know the feature or the set of features that is most predictive of IVF success. It is possible that the most important variable for a successful IVF cycle could still be unknown to science. That's why when an IVF cycle fails, we really can't say anything to the patient why this has happened. But in principle, maybe uncovered by AI system, this may be referred to as a computational embryology.
That means in future, there won't be clinical embryology. It will become computational embryology. AI model analysing time-lapse imaging can significantly improve embryo implantation potential as it can detect subtle morphological changes and dynamic behaviour, which is difficult for human observer to discern.
Studies highlight AI potential to standardise embryo selection, particularly in centres, with limited access to experienced embryologists and underscored reliability. I'll take just three minutes. VitroLite's IDS score evaluates embryo virality with a neural network analysing time-lapse videos and Fertility's CHLA embryo assistant, which uses both embryo and patient data to predict the probability of embryo implantation.
AI is seen as an opportunity to increase the positive stagnant success rate by 15% to 17%. So we'll have more implantation by at least 15% to 17% potential because we'll be doing biopsy to find out whether it's euploid or chromosomally competent or not. But AI will tend non-invasive, just by looking at that blastocyst with image enhancement. So it will do that. And how much? AI's ability to analyse embryologic images and clinical data presents a quicker, most cost-effective non-invasive solution. New AI model accurately predicts embryo ploidy starts from optical microscopy image, increasing precision from 65% to 77%.
You can see how they do the enhancement of the images. Clearly, inertial mass and top ectoderm is seen. Now, this was with a husband and couple, aside from the wife and sperm from the husband. But suppose we need an AI donor. It's difficult. They put a laundry list on the couple, this kind of donor they need.
So the use of this process actually is very labour-intensive, manually searching for clinics in order to secure a good match. And AI models trained on large database of facial images have been used in a variety of settings to identify suitable individuals. The intended result is a child that is physically resembling its intended parents so that they can pass on as genetically related.
In Spain, they are doing a very good business, and the clinicians are very much under pressure to find a suitable donor. AI can even tell about the lactobacillus dominance in the uterine cavity. And AI, now we know about the embryo, now we come to the guest house, I mean, how about the endometrium before transfer? There also we use AI.
There's an endoclassifying AI software that will tell us good endometrium and bad endometrium. Good endometrium has a 74% success, bad endometrium has 90% negative pregnancy result. So now endometrial, using a combination of AI on embryos and endometrium in the same patient, we could achieve a high success rate in IVF.
Now when the embryo is transferred, we have to think, does this embryo belong to this lady? So, AI safe tracking, tracking of embryos also is done by artificial intelligence. And I finish. I think AI, it's not a plug-and-play device. The tools in the market as under development have the potential to transition care to the next level, but require consider appraisal before they're incorporated in IVF practise. AI has some challenges, and especially data privacy. Anyway, that has to be encrypted.
Another business effort of Padma Shri Prof. Dr. Kamini A. Rao is Medline Academics, which focuses on educating people in the field of reproductive medicine. The
Fellowship in Reproductive Medicine is this institution's most sought-after course. This institution's hybrid training approach, in which all theory is taught online and students participate in the simulations in person in our Bangalore location, is its most distinctive characteristic. Everyone can benefit from the hybrid learning style. For all active clinicians, Medline Academics provides a hybrid learning environment.
Dr. Kamini Rao Hospitals, the
top IVF treatment centre in Bangalore, has been a centre of excellence for many years and is carrying on the tradition. In addition to assisting couples in starting their own small children, this institution exposes students enrolled in Indian embryology fellowship programs to a variety of ART cases. From menarche to menopause and beyond, we also provide comprehensive healthcare for women. Under the direction of Padma Shri Prof. Dr. Kamini A. Rao, students enrolled in prestigious fellowship programs are finishing their clinical attachment at Dr. Kamini Rao Hospitals.
