Imaging and AI work together to assess embryo health

Several Illinois researchers, including Bioengineering Department Head Mark Anastasio and Illinois graduate Neha Goswami, have paired label-free optical imaging and AI to create an effective new method of detecting embryo health. Their findings, titled “EVATOM: an optical, label-free, machine learning assisted embryo health assessment tool,” have recently been published in Nature's Communications Biology.

The project began in the late Professor Gabriel (Gabi) Popescu's lab, where the research team first aimed for utilizing the label-free imaging to identify nuclei of individual cells within mouse embryos. The type of imaging was important, as many imaging types require staining or labeling specific cells so they can be easily identified. However, even the mildest forms of labeling have some effect on the cells themselves.

“Embryos are very delicate, as you can imagine,” said Goswami. “We don't want to disturb them. We don't want to introduce any external factors at all. So what we did is use machine learning to identify these nuclei inside each cell from the phase images directly, so we don't need to stain them.”

The researchers trained AI to identify cell nuclei. At first, staining was required to show the AI what to look for, but before long the researchers were able to use the trained deep learning model to predict the nuclei from stain-free imaging. After learning to identify these cell nuclei, the AI can unobtrusively identify healthy and unhealthy nuclei and, by extension, determine whether the embryos themselves are healthy or unhealthy.

Currently, embryologists use standard microscopes to examine developing embryos. This new method of embryo monitoring, paired with the highly specialized imaging equipment used by the research team, can reveal embryo protein content and other health factors that standard methods can’t. 

A project that spans so many disciplines requires a diverse team, and the collaboration was one of Goswami’s favorite aspects of the work.

“The bioengineering department has a vibrant environment where different people with different expertise come together,” said Goswami. “When I started this project, I was a grad student with experience in optics, imaging and machine learning, but I didn't know anything about embryology. Throughout the project, Dr. Nicola Winston and Professor Romana Nowak provided pearls of wisdom about embryology. Later on, during the machine learning process, Professor Anastasio guided me on the feature extraction based model development and supervised the overall machine learning aspect of the project. This collaboration really made the project something different.”

All of the project collaborators mentioned the benefits of interdepartmental research, and how this kind of teamwork benefits scientific knowledge as a whole. 

“This work is a wonderful example of how collaboration between specialists in different fields of science can provide greater insight into biological questions by combining their skills,” said co-author Dr. Nicola Winston, a professor in the University of Illinois Chicago’s department of obstetrics and gynecology. 

Using this technology on human embryos is still far in the future, as it would require retraining the AI on all new samples, but the team is optimistic that it could be used in the future to assist in human IVF, especially with the insight of so many experts in different fields. 

“This project was a collaboration between the Departments of Bioengineering, Animal Sciences and OB/GYN and is an excellent example of what can be achieved through multidisciplinary research,” said paper co-author and professor of animal sciences Romana Nowak. “The seed was planted by our late colleague Gabi Popescu, who enjoyed and pursued many fruitful collaborations. Label free optical imaging combined with AI has exciting long term potential for use in the clinical setting.”  

The complete article in Nature's Communications Biology can be read here.

The authors have dedicated this work to the late Prof. Gabriel Popescu.