Skip to content

Improving Infertility Outcomes Through Artificial Intelligence

In This Article

  • Rapid advances in artificial intelligence (AI) are powering improvements on multiple fronts in infertility research
  • Infertility laboratories have historically relied on highly detailed physical, microscopic morphological grading
  • Massachusetts General Hospital researchers are piloting AI-assisted approaches and obtaining results that match or exceed human assessment
  • Key AI applications: sperm morphology assessment, normally fertilized human embryo evaluation and selection, blastocyst development prediction system, embryo classification, specimen tracking and embryo selection for transfer based on AI image analysis

As artificial intelligence (AI) methodologies, such as machine learning, become more robust and more widely available, clinicians across specialties are looking for ways they can be applied to medicine. The power of AI comes from its ability to conduct robust analysis of layered biological data to identify and interpret revealing patterns about health and disease.

Researchers within the Fertility Center at Massachusetts General Hospital are pioneering innovative AI-based methods to improve the precision of sperm morphology assessments and embryo selection, which traditionally have relied upon analysis of specimens through visual inspection, human discernment and individual discretion.

This work is being led by Charles L. Bormann, PhD, director of embryology for the Mass General Fertility Center, in collaboration with Hadi Shafiee, PhD, and bioengineers at Brigham and Women's Hospital.

"Currently, the most widely used method for embryo selection is strictly based on the morphology of the cells. We take embryos out of their culture incubator and look at each one individually and grade them," Dr. Bormann says.

Inspecting Cells

In the embryology lab, the demanding precision of the visual assessment includes such parameters as degree of blastomere symmetry and percent of cytoplasmic fragmentation. It culminates in a rank ordering of an embryo based on an objective score assigned after microscopic visual analysis, requiring high expertise and collaboration.

According to Dr. Bormann, one drawback of the traditional manual embryo grading system is its inability to consistently identify the implantable embryo from the patient's cohort. This limitation can lead to the selection of multiple embryos for transfer, increasing the likelihood of giving birth to twins—an outcome that is not always preferable for prospective parents.

Analyzing Images

The researchers believe they can increase the precision of embryo transfer and reduce multiple births using AI techniques based on past studies.

Dr. Bormann and colleagues recall a randomized controlled trial they were involved in several years ago that tested the clinical efficacy of using time-lapse imaging to assess embryos and select candidates for transfer.

"When I downloaded the individual frames and looked at all 12 dishmates, I found it was easy to identify the cleavage stage embryo with the top morphologic score," say Dr. Bormann. "I could also see more clearly which one had the most potential for developing into a blastocyst."

Realizing the Potential of AI

The team's next thought was: "How could we get a system to perform this task so we can consistently pick the very best quality embryo?" They turned to AI at just the right time when early improving results were widely publicized in the field of facial recognition.

"All of sudden AI was producing results where the computer could not only identify that an image was of an individual, but that it could be used to correctly determine the identity of the person," Dr. Bormann says.

From there, it seemed plausible that the same AI technology for facial recognition could be applied to embryo recognition.

Expanding the Application of AI to Infertility

AI has the ability to make fine discernments in images, so the team used the Fertility Center's treasure trove of images to train the AI algorithm to make embryo selections based on set criteria. The results were strong—typically outperforming human visual inspection.

That success inspired the team to look for other applications for AI-assisted infertility testing. While still experimental, the team now has a number of promising research projects underway to explore many possible additional applications of AI, including:

  • Human sperm morphology analysis using smartphone microscopy and deep learning
  • Deep learning-enabled prediction of fertilization based on oocyte morphological quality
  • Deep convolutional neural network for assessment and selection of normally fertilized human zygotes
  • Deep learning-enabled blastocyst prediction system for cleavage-stage embryo selection
  • Deep learning to improve day 5 embryo scoring and decision making in an embryology laboratory
  • An inexpensive, automated AI system for human embryo morphology evaluation and transfer selection
  • Development and evaluation of inexpensive, automated deep learning-based imaging systems for embryology
  • Witnessing system to track and identify patient specimens based on unique oocyte and embryonic morphologic signatures
  • AI-enabled system for identifying the correct location to safely inject sperm into oocytes during intracytoplasmic sperm injection (ICSI)
  • AI-enabled system for identifying the correct location to safety perform laser-assisted zona-pellucida hatching on cleavage stage embryos
  • AI-enabled quality assurance system for monitoring the performance of the embryo culture system

Dr. Bormann expects to see greater efficiencies and reduced costs, while maintaining excellent results with the application of AI-assisted methods.

"There is so much at stake for our patients with each IVF cycle," he says. "Embryologists make dozens of critical decisions that impact the success of a patient cycle. With assistance from our AI system, embryologists will be able to select the embryo that will result in a successful pregnancy better than ever before."

Learn more about the Fertility Center

Explore research in the Fertility Center

Related topics


Black non-Hispanic and Hispanic women are less likely than white women to receive medication for opioid use disorder consistently or at all.


Ilona Goldfarb, MD, MPH, of the Massachusetts General Hospital Department of Obstetrics & Gynecology, and colleagues have documented disparities between Latina and non-Latina pregnant women in the prevalence of SARS-CoV-2 infection and the severity of COVID-19.