AI In IVF: Separating The Useful From The So What?

We’re up to ten million IVF babies in forty years — a great start.

Now let’s get to a million babies a month.

The need is there (in reality it’s higher), and many of the tools for expansion of assisted reproductive technology (ART) are already on the shelves, some at work in other areas of cell biology, some as working prototypes in the laboratories and garages of tinkerers and CSO’s.

Now we face the challenge of replacing our mish-mashed global IVF delivery ecosystem with one that leverages the best of what we have now (spectacular science, a core of superbly trained professionals and a well-developed private equity financing system for the mature segments of the industry) to unlock accelerated organic growth of ART, growth not constrained by too few medical or science specialists, or by the artifacts of particular state or country IVF reimbursement policies.

This is an engineering challenge, and one in which artificial intelligence (AI) is indispensable.

Why? Because fertility, like most of women’s health, still lives in the analog, imprecisely diagnosed world of organ and system-based medical care.

The names for the conditions that we treat are the same now as when I was a first year medical student in 1983, unlike oncology and immunology, where the tools that unlock genetics and gene expression greatly amplify the signals of pathology from the noise of physiology, moving the focus of our interventions to the cellular and molecular level and providing a roadmap for innovation and improvement.

A similar mechanism of action (MOA) roadmap will do the same for women’s health (we need a healthy dose of funding too.) And AI is an ideal platform for exploring and better understanding, in a mechanism-agnostic, unbiased way, the data-rich world of IVF.

We have studied artificial intelligence in women’s health in IVF for five years, and met with a couple of dozen startups tackling the challenge of applying enhanced inputs (optical and numerical) with mechanism-agnostic large scale data analysis, in order to define the best practices embedded inside the scientifically brilliant but engineering-challenged world of ART. A lot of this work is shake-your-head-in-amazement brilliant. I spent five days in Dubrovnik last summer, at the first AI Fertility world meeting, and for most of that time I felt I was the dumbest person in the room, as one nuanced application to specific nodes of the IVF decision tree after another flashed across the screen.

Flying home, I reined in my enthusiasm. Our goal is one million IVF babies a month, not the creation of an innovation theme park. How do these tools do against the KPI’s of dollars per baby, time to baby and life disruption to baby?

Benefits can be purely operational, making the cycles easier and cheaper to perform without affecting the pregnancy rate. If the savings are passed on to the patient, this is a dollars to baby benefit. If they result in higher throughout through the facility, this can be a time to baby benefit. Benefits can improve pregnancy rates through better decision making or by identification (and hopefully elimination) of operator performance lag or process inefficiency. And of course the two types of benefit are additive.

What are the candidate nodes in the IVF process that should benefit from AI? Let’s look at an obvious short list of use cases for AI during IVF for infertility, most of which are being developed in multiple sites, some are being trialed and some in the early stages of implementation:

1) more precisely predicting outcomes
2) stimulation protocol selection and making daily medication dosing decisions
3) choosing the day of HCG (or LH) trigger and time for egg retrieval
4) determining whether to use IVF or ICSI for fertilization
4) choosing the specific sperm cell when ICSI (intra-cytoplasmic sperm injection) is performed
4) choosing when to perform the embryo transfer
5) picking the best embryo to transfer

This list is far from comprehensive, and the road to true process optimization has many stops. Biology is complicated.

Take the performance of ICSI, for example. A truly engineered to the last detail IVF performs ICSI robotically, removing operator inconsistency. AI optimizes egg placement on the holding pipette, angle of needle insertion, speed of needle corrected for the elasticity of the zona pellucida, speed of sperm injection and needle removal — an dynamic and ongoing iteration defining best practice. Simultaneously, an upstream AI-determined process optimization of sperm selection makes it easier to isolate and fine tune the benefit of improving ICSI mechanics by removing the noise of sperm variability. Downstream, AI-controlled variance of incubation — responding to metabolic or growth dynamics of the individual embryo, gives the sperm-optimized, ICSI mechanics optimized embryo the best conditions for development, facilitating our ability to isolate factors that correlate with embryo genetics and post-transfer regulatory function.

Over time, AI sands down every rough surface in the construction of each embryo, listens for and provides an optimal, embryo specific environment in which to divide and differentiate, generates an outcome prediction, then crosschecks each optimization decision against the actual outcome when know — a virtuous cycle of unbiased, efficient improvement.


Having defined the ideal future union of AI and IVF, how are we doing in actually getting there?



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
David Sable

bio fund manager, Columbia prof, ex-reproductive endocrinologist, roadie for @PriyaMayadas