Researchers Use AI and Deep Learning to Identify Potential Birth Defects

Researchers at the University of Ottawa are pioneering the use of an artificial intelligence-based deep learning model to assist doctors in rapidly reading ultrasound images and locating possible birth defects early in pregnancy.

The goal of the study was to demonstrate the potential of using a new deep learning tool designed to assist doctors in locating and properly identifying cystic hygroma in first-trimester ultrasound scans.

The condition has an overall survival rate of only 10% early in the pregnancy. This is why Cystic Hygroma is considered a serious condition. Because of the early risk to the child, it must be closely followed throughout the pregnancy. Once the fetus reaches 26 weeks, the survival rate rises to over 60%. After which, preconditions must be taken as issues can still arise until delivery.

Cystic hygroma develops due to an error in the development of lymph sacs and vessels as babies gestate. This causes the lymphatic vascular system to develop abnormally. Though rare, it is a potentially life-threatening disorder that leads to fluid swelling around the head and neck. It is documented in approximately 1 in 800 pregnancies and 1 in 8,000 live births.

Dr. Mark Walker, of the University’s Faculty of Medicine, stated in the paper, “What we demonstrated was that in the field of ultrasound we’re able to use the same tools for image classification and identification with a high sensitivity and specificity”.

He continued that this tool could also be used to identify other life-threatening issues at earlier stages, “What we demonstrated was in the field of ultrasound we’re able to use the same tools for image classification and identification with a high sensitivity and specificity.”

The use of ultrasound is a critical tool for observational medicine and development widely used. But, due to the lack of detail given by images, small body structures, and normal fetal movements, the quality of images can be lacking. Which in turn can cause issues for doctors who are seeking to interpret what is shown in the scans as they attempt to locate issues with the growing fetus.

Because of this ongoing issue, the Canadian team wanted to test how an AI-powered pattern recognition tool could assist medical staff in locating early issues such as cystic hygroma.  According to the study, the model was able to achieve an accuracy of over 93% overall, showing excellent promise in opening improved care for pregnant women and their fetuses by preventing birth defects.