Deep Learning Model Automates Severe Heart Valve Disease Detection
A novel deep learning approach can detect aortic stenosis using ultrasound scans of the heart, which could enable early detection and intervention.
Source: Getty Images
By - Researchers at Yale School of Medicine have developed a deep learning (DL) approach that can accurately detect aortic stenosis by analyzing heart ultrasound scans, according to a study published last week in the European Heart Journal.
The research team indicated that aortic stenosis, a common heart disease caused by the narrowing of the aortic valve, is a significant driver of mortality and morbidity. Early detection of the disease is critical to prevent these outcomes, but it requires specialized ultrasound imaging of the heart, known as doppler echocardiography.
Doppler echocardiography is the main test leveraged to detect aortic stenosis, but the specialized nature of the imaging makes it inefficient and inaccessible for use in early detection efforts.
“Our challenge is that precise evaluation of [aortic stenosis] is crucial for patient management and risk reduction. While specialized testing remains the gold standard, reliance on those who make it to our echocardiographic laboratories likely misses people early in their disease state,” said senior author Rohan Khera, MD, MS, assistant professor of cardiovascular medicine and health informatics at Yale and director of its Cardiovascular Data Science (CarDS) Lab, in a press release detailing the study.
The research team’s goal was to develop a model that could enable point-of-care ultrasound screening to facilitate early detection of the disease.
To do this, the researchers developed the deep learning model using 5,257 studies from transthoracic echocardiography (TTE) exams that included 17,570 videos between 2016 and 2020 at Yale New Haven Hospital.
The tool was then externally validated using 2,040 consecutive studies from Yale New Haven Hospital, in addition to two geographically distinct cohorts of 4,226 and 3,072 studies from California and other hospitals in New England.
The model achieved high performance across cohorts, reaching an area under the receiver operating characteristic curve of 0.978 in the test set. Further, the DL achieved an area under the receiver operating characteristic curve of 0.952 in the California cohort and 0.942 in the New England cohort.
These findings led the researchers to conclude that the model has potential utility for early detection of aortic stenosis.
“Our work can allow broader community screening for [aortic stenosis] as handheld ultrasounds can increasingly be used without the need for more specialized equipment. They are already being used frequently in emergency departments, and many other care settings,” Khera stated.
However, more research is needed before the tool can be deployed in clinical settings.
Previous research has also aimed to improve aortic stenosis detection using artificial intelligence (AI).
In 2021, Kaiser Permanente researchers demonstrated that natural language processing (NLP) could assist clinicians with identifying aortic stenosis.
The model was trained to sift through echocardiogram reports and EMR data to flag abbreviations, words, and phrases that are associated with the condition.
The tool then rapidly identified almost 54,000 patients who met the criteria for aortic stenosis, a process that the research team noted may have taken years if undertaken manually.
