Care Fertility and BJSS Use AWS and Machine Learning to Improve Embryo Assessment for IVF Success

Care Fertility operates 15 IVF clinics in the UK and Ireland, three in Spain, and one in the USA. It wanted to use machine learning (ML) to improve the way its embryologists assess and select embryos most likely to lead to successful pregnancies and births. After evaluating various off-the-shelf solutions, it worked with AWS Partner BJSS to build a system to analyze images in near-real time. This improves accuracy and facilitates decision making around which embryos to transfer to patients, leading to better results. It also frees up highly skilled embryologists from repetitive manual work.

The UK’s largest network of fertility clinics, Care Fertility, wanted to explore ways to improve and accelerate the process of embryo assessment and selection during in vitro fertilization (IVF) using ML and artificial intelligence (AI). A World Health Organization research found 17.5 percent of the adult population are affected by infertility, or 1 in 6 people. The fertility clinic group aimed to improve both accuracy and reproducibility.

The assessment and selection procedure is time-intensive and requires knowledge, training, and quality assurance because it is such a key part of the IVF process.

After trying three off-the-shelf solutions with mixed results, Care Fertility decided to create its own system and algorithms. When it needed help, it chose AWS Partner BJSS for its experience and expertise with AI, ML, and Amazon Web Services (AWS).

“We were really pleased by the level of enthusiasm and passion we got from the guys at BJSS,” says Dr. Alison Campbell, chief scientific officer at Care Fertility. “It was in the middle of the COVID-19 pandemic so we didn’t meet in person but it was really exciting work and we could quickly see that we would build something brilliant.”

Care Fertility uses time-lapse video to monitor embryo development and help embryologists make the right decisions.

To capture images of the embryos, Care Fertility uses the EmbryoScope time lapse incubator, made by Sweden’s Vitrolife. The device has an integrated camera system that takes images in up to 11 focal planes, every 10 minutes, from fertilization to the time of embryo transfer, around 5 days later.

Care Fertility and BJSS created an abstraction layer to extract images from the embryologists’ desktop machines. To reduce impact on clinic networks, the ML system then compresses the images by a factor of 30 before they are transferred to Amazon Simple Storage Service (Amazon S3), an object storage service offering industry-leading scalability, data availability, security, and performance.

These multi-focal plane images are processed and resampled before a focal stacking algorithm merges the frames into a single image. The images are then fed into a two-stage ML system. The first model is a Convolutional Neural Network classifier to extract features and classify the stage of each embryo image.

A Temporal Convolutional Network then interprets a time-series of image features, producing annotations from fertilization through to embryo development, over several days. Finally, the system provides automatic time stamps for each embryo developmental event and a confidence score—if confidence is low, it requests embryologists’ input.

The ML models are all run using AWS Step Functions for orchestration, which creates visual workflows for distributed applications. This solution helped developers design a scalable, serverless system that boots up quickly but still controls costs.

Using AWS, the compute resources automatically scale to manage two different performance profiles—one for processing individual images and one when the second model runs looking at images over 90 hours, then hourly. The system can process 800 concurrent embryos. The algorithm is run on demand by the embryologist with the click of a button and returns a result in 1-2 seconds.

This tool drastically reduces expert time spent on embryo assessment, freeing up approximately 6 months of embryologist time per year—half a full-time equivalent. BJSS created a user interface for embryologists to view images through a web interface.

The solution uses caching, compression, parallel streaming, and lazy loading technology to reduce lag and remove the need to download thousands of images at a time. The team ran usability workshops with its scientists to make sure the system was as easy to use as possible.

Care Fertility’s embryologists have welcomed the new tools and are excited about their potential. After running old and new systems in parallel, its laboratories are now live—and user feedback has been excellent.

Over the last decade, Care Fertility has collected close to half a billion images of embryos—all of which had been meticulously and manually annotated. This provided the unique dataset that the ML models were trained on. No other fertility group is thought to have this amount of high-quality data.

Care Fertility is research-focused and works to define and refine best practice to replicate at every clinic. “The BJSS team approached this complex project thoroughly and confidently. They communicated regularly and clearly. The project hit a few challenges which were tackled swiftly and competently by the talented, multidisciplinary BJSS team,” says Campbell. “The resulting product is world-class and something BJSS and Care Fertility can all be proud of. Importantly, the ML tool, developed by BJSS to help Care Fertility assess human embryos, will help improve the chance of patients reaching their goal of having a baby.”

The new AI and ML system has proved so successful after being rolled out to 15 laboratories that Care Fertility is looking into whether it could be applicable to clinics outside the group. “There is also potential for us to offer this as a service to other clinics and it will certainly ease the process of onboarding as we grow and our industry continues to consolidate,” says Campbell.

But the benefits go beyond efficiency; crucially, this tool has direct impacts on patient treatment. By increasing the reproducibility and reliability of the embryo selection process and the accuracy of predictions, Care Fertility can ensure they give patients their best chance of having a child. 

In May 2023 the project won the National Technology Award 2023 for healthcare tech of the year.