Summary: Using datasets of fetal ultrasounds, a new AI algorithm is able to detect cystic hygroma, a rare embryonic developmental disorder, within the first trimester of pregnancy.
Source: University of Ottawa
In a new proof-of-concept study led by Dr. Mark Walker at the uOttawa Faculty of Medicine, researchers are pioneering the use of a unique AI-based deep learning model as an assistive tool for the rapid and accurate reading of ultrasound images.
It’s trailblazing work because although deep learning models have become increasingly popular in interpreting medical images and detecting disorders, figuring out how its application can work in obstetric ultrasonography is in its nascent stages. Few AI-enabled studies have been published in this field.
The goal of the team’s study was to demonstrate the potential for deep-learning architecture to support early and reliable identification of cystic hygroma from first trimester ultrasound scans. Cystic hygroma is an embryonic condition that causes the lymphatic vascular system to develop abnormally. It’s a rare and potentially life-threatening disorder that leads to fluid swelling around the head and neck.
The birth defect can typically be easily diagnosed prenatally during an ultrasound appointment, but Dr. Walker and his research group wanted to test how well AI-driven pattern recognition could do the job. The findings are promising.
Using a dataset of nearly 300 fetal ultrasounds collected retroactively at The Ottawa Hospital, images were analyzed using a DenseNet model to correctly identify cases of cystic hygroma compared to normal controls by calculating sensitivity, specificity, and other measurements. Gradient class activation heat maps—which visualize pixels in images—were also generated to assess model interpretability. Overall model accuracy was 93%.
“The model was outstanding, even with a small number of training images. And so, potentially, 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,” Dr. Walker says.
The findings were recently published in PLOS ONE, a peer-reviewed open access journal.
The uOttawa-led research group has high hopes for where this type of study might lead. With further development, including testing in a large multi-site dataset, the team believes their approach might be applied to other fetal anomalies generally identified by ultrasonography.
Down the line, Dr. Walker says the group will aim to develop an international consortium to upload obstetrical ultrasound images to “the cloud”—the euphemism for the use of remote servers to store data and provide computing services to swaths of users. This could ultimately help physicians in low- and middle-income countries achieve cloud-driven interpretation and diagnosis.
“This particular project is really meant to be the beginning of a large body of work,” says Dr. Walker. “We’ve got several papers to follow this one.”
Dr. Walker is a high-risk obstetrician and clinical epidemiologist who is a Full Professor and the Vice-Dean of Internationalization and Global Health at the uOttawa Faculty of Medicine. He is co-founder of the OMNI Research Group (Obstetrics, Maternal and Newborn Investigations) at The Ottawa Hospital, the largest maternal and newborn research group in Canada.
His collaborators on this publication include top research talent from uOttawa Faculty of Medicine departments and the Ottawa Hospital Research Institute (OHRI), where Dr. Walker is a senior scientist.
Using deep-learning in fetal ultrasound analysis for diagnosis of cystic hygroma in the first trimester
To develop and internally validate a deep-learning algorithm from fetal ultrasound images for the diagnosis of cystic hygromas in the first trimester.
All first trimester ultrasound scans with a diagnosis of a cystic hygroma between 11 and 14 weeks gestation at our tertiary care centre in Ontario, Canada were studied. Ultrasound scans with normal nuchal translucency were used as controls. The dataset was partitioned with 75% of images used for model training and 25% used for model validation. Images were analyzed using a DenseNet model and the accuracy of the trained model to correctly identify cases of cystic hygroma was assessed by calculating sensitivity, specificity, and the area under the receiver-operating characteristic (ROC) curve. Gradient class activation heat maps (Grad-CAM) were generated to assess model interpretability.
The dataset included 289 sagittal fetal ultrasound images;129 cystic hygroma cases and 160 normal NT controls. Overall model accuracy was 93% (95% CI: 88–98%), sensitivity 92% (95% CI: 79–100%), specificity 94% (95% CI: 91–96%), and the area under the ROC curve 0.94 (95% CI: 0.89–1.0). Grad-CAM heat maps demonstrated that the model predictions were driven primarily by the fetal posterior cervical area.
Our findings demonstrate that deep-learning algorithms can achieve high accuracy in diagnostic interpretation of cystic hygroma in the first trimester, validated against expert clinical assessment.