Here’s How AI Is Helping Make Babies By Revolutionizing IVF
One in four couples in developing countries is impacted by infertility. About 48.5 million couples experience infertility worldwide. Today, infertility is rapidly becoming an epidemic.
In vitro fertilization (IVF) is a technique that helps people facing fertility problems have a baby. Despite IVF’s potential, the outcomes are unpredictable. To make matters worse, access to fertility care is abysmal. Even in a developed market such as the United States, just 2% of people suffering from infertility have tapped into IVF.
“IVF has been around for over 40 years,” says Eran Eshed, CEO of Fairtlity. “Despite many innovations on the biotechnology side of things, surprisingly, there has been almost zero usage of data and techniques like artificial intelligence (AI) to influence outcomes.”
While data science can’t solve biological problems, Eshed believes AI will enhance the IVF process at every step where decisions are made.
Today, we’re seeing exciting applications of data science in fertility that could improve embryologists’ capacity cycle by 50% and increase the chances of live birth by 4%.
Where IVF falls short today
IVF is a fertility technique in which an egg is removed from a person’s ovaries and fertilized with sperm in a laboratory. The successfully fertilized egg–an embryo–is then implanted into a uterus to grow.
Clinicians and embryologists make many decisions at several junctures. “These decisions are based on experience, intuition, and a set of very, very rudimentary rules,” laments Eshed.
Today, there are two key challenges with IVF:
1. Poor access to care
“When just 2% of the impacted population can leverage IVF, it’s clear that access to care is a big, big issue,” highlights Eshed. “IVF is currently focused on infertility patients–those not getting pregnant either by timed intercourse or simple treatments such as oral medications,” shares Dr. Gerard Letterie, a reproductive endocrinologist and partner at Seattle Reproductive Medicine. “This is a relatively restricted segment of the population.”
In the future, Dr. Letterie expects the patient segment to include those who are interested in fertility preservation by freezing eggs or creating embryos for future use. “This will markedly expand the number of patients seeking care with assisted reproductive technologies,” he predicts.
2. Uncertain outcomes
How successful is IVF? The chances of conceiving from a single IVF cycle are around 30%. Hence, most patients need to undergo multiple cycles before experiencing a successful live birth.
While the success of IVF is influenced by age, data shows that most IVF cycles fail for even the youngest and healthiest women. IVF outcomes heavily depend on decisions made during the clinical process and on the expertise of the embryologists.
AI has the potential to reinvent fertility treatments
How long is the IVF process and what are the steps involved? IVF starts with a clinician’s assessment of the cause of infertility. “Then, it moves into the stimulation phase where the doctor determines the best protocol for ovarian stimulation,” shares Eshed.
This is commonly followed by the collection of eggs and sperms, fertilization of eggs using sperms to create embryos, embryo culture in the clinic, transfer of embryos to the mother, and a live birth months later.
“As people go through this process, the success rates drop significantly at each stage,” says Eshed. Typically, six to seven strategic decision points determine the effectiveness of each step. “In the business world, we’d call them leverage points where you can make a difference,” he adds.
These points include decisions by clinicians on the stimulation medication protocol or timing of egg retrieval. In the lab, embryologists make several judgments by interpreting images about oocytes (developing eggs), sperms, and blastocytes (fertilized eggs).
“I’m confident that AI can help streamline the decisions to augment clinical decision-making,” claims Dr. Letterie. For example, sophisticated convolutional neural network-based image analytics can aid embryologists’ in interpreting the images to improve outcomes.
The global IVF market is set to reach around $36 billion by 2026, per an industry report. Dr. Letterie anticipates that “there simply won’t be enough skilled embryologists to address this rising demand.” Recently, the fertility space is witnessing multiple technology investments, with several funded, AI-driven startups.
Eshed founded Fairtility in Israel to address the acute challenge of embryo analysis with AI. Recently, his firm raised $15 million in Series A funding. Other startups such as Emrbyonics, Mojo, and ALife have come up with AI-based fertility solutions to analyze embryos, assess sperm quality, and personalize IVF treatment plans.
How Artificial Intelligence is revolutionizing embryo analysis
Today, embryo classification is done by embryologists who manually inspect pictures for a set of visually detectable features. Fairtility utilizes computer vision algorithms to augment this process and predict the likely effectiveness of implantations.
Their AI algorithms are trained from a dataset of over 200,000 embryo videos and over 5 million clinical data points drawn from a diverse patient demographic. This gives the AI models the power to analyze minute features that are often undetectable by even the most experienced embryologists.
Fairtility’s solution, CHLOE, is a cloud-based system that acts as a decision support tool for AI-powered embryo selection. The tool integrates with time-lapse imaging (TLI) systems to provide continuous predictions from fertilization to the blastocyst stage. As the TLI system captures pictures of embryos at different stages of development, they are automatically identified, segmented, and analyzed at the pixel level.
In addition to automating this process, the AI model helps precisely quantify attributes such as size, area, shape, proportion, and symmetry. “That’s not something a human can do, so in a sense, we’re bringing a lot more intelligence in the process,” shares Eshed. Such precise information coupled with implantation probability enables an embryologist to make data-driven decisions for every embryo cultured in the TLI device.
CHLOE’s algorithms can predict blastulation with 96% accuracy, implantation with 71% accuracy, and whether an embryo is genetically healthy at 69% accuracy, per a paper submitted to the ESHRE conference. Such results improve embryologists’ prediction of embryo viability, which is currently around 65%.
Additionally, the AI solution can help embryologists identify anomalies, such as unusual cleavage patterns or severe fragmentation or pronucleate abnormalities that may otherwise be missed. Thus, CHLOE boosts the likelihood of healthy embryo selection.
However, despite improved results in embryo selection and process efficiency, studies have yet to demonstrate concrete improvements in live birth rates, which is considered the gold standard.
“AI cannot and should not replace embryologists and clinicians,” clarifies Eshed. “It is important that every patient gets the same and highest standard of care, irrespective of a practitioner’s experience or workload.” This is where CHLOE levels the playing field.
Fairtility provides its solution in a software as a service (SaaS) model to clinics and fertility centers around the world. Per Eshed, the installation of CHLOE requires no hardware and can be done remotely. With over 25 active installations worldwide, Fairtility has gained CE mark registration (a European safety, health, and environmental certification) and is reportedly in advanced FDA approval stages.
The three pre-requisites to deliver impact with AI
To realize the full potential of AI, several key challenges must be overcome:
1. Tackling data availability issues
“Data is a huge challenge in this space,” says Eshed. Data ranges from notes about treatment history, electronic medical records (EMR), ultrasound images, and videos. Eshed says that while the data exists, it is highly dispersed, and neither curated nor organized well. Even today, several clinics archive records in physical files. The entire process must be digitized to gain an end-to-end perspective from which AI models can learn.
2. Integration into the workflow
“Current practices have not been sophisticated regarding workflow and process development,” shares Dr. Letterie. Such AI solutions can help drive outcomes only when they are integrated into clinical and laboratory workflows. “This will also require education on the part of all stakeholders,” he adds. For example, Dr. Letterie will be launching a 15-course curriculum on AI in fertility using presentations from thought leaders at the upcoming ESHRE conference.
3. Fostering trust and adoption
Even after demonstrating effectiveness, achieving clinical uptake and routine use takes time. “Never underestimate the resistance to change,” cautions Eshed. “A fancy AI solution is not necessarily going to impress anybody.”
Dr. Letterie shares the example of beta-blockers, drugs that prevent cardiovascular disease, as a case in point. These drugs were initially used in patients recovering from myocardial infarction (MI) to prevent the recurrence of a heart attack. Despite studies demonstrating a clear reduction in morbidity and mortality, it took over 7 years to integrate beta blockers into routine clinical care.
“Similarly, we have an uphill battle to convince clinicians and embryologists that using AI tools will improve outcomes,” cautions Dr. Letterie. “Most practitioners aren’t familiar with AI and its applications in the clinical setting; hence, they are extremely hesitant to change practice patterns.” He feels that it is essential to show a clear improvement of outcomes before expecting significant uptake. Meanwhile, we must brace for the time lag in building trust with technology-driven treatments.
A future where fertility care will be commonplace
Dr. Letterie expects IVF to grow in prevalence with better success and fewer cost barriers to care. He foresees the development of early detection tools that warn patients who might be experiencing decreases in fertility, as opposed to today where patients end up with unretrievable fertility potential. With enhanced visibility about their fertility, patients will then be able to take early actions by freezing sperms, oocytes, or embryos.
He concludes that smartphones will be one of the biggest and most significant improvements in the delivery of fertility care.