Researchers have developed cardiology’s first pipeline for automated echocardiogram interpretation—an innovation that could cut healthcare costs while expanding care to underserved communities, according to a study published online this week in Circulation.
Cardiac imaging, including echocardiogram, is vital to catching CVD early, corresponding author Rahul C. Deo, MD, PhD, and colleagues wrote in the journal. It allows physicians to monitor changes in a patient’s heart structure, which can be indicative of conditions like diabetes, hypertension or valvular disease, but it’s not standard practice.
“Although early evidence of these changes is often detectable by imaging and could in principle be tracked longitudinally in a personalized manner, the cost of imaging all individuals with cardiac risk factors would be prohibitive,” Deo, of the Division of Cardiovascular Medicine at Brigham and Women’s Hospital in Boston, and coauthors wrote.
They said automated image interpretation could eliminate that financial strain, especially in rural and underserved communities that might suffer from staff shortages or a lack of qualified cardiologists. An automated handheld device would allow nonexperts to evaluate heart disease by linking a patient’s echocardiogram to a cloud-based interpretation system, which would assess that patient’s cardiac structure and function and compare it to outcomes data from thousands of other echocardiograms for a diagnosis.
Deo and his team developed their pipeline using a decade’s worth of data from 14,035 echocardiograms. Using machine learning, the researchers trained and evaluated convolutional neural network models for multiple tasks, including the automated identification of 23 viewpoints and segmentation of cardiac chambers. Results were evaluated against manual data from 8,666 routine echocardiograms taken during normal clinical hours.
The authors said convolutional neural networks identified views of the echocardiogram well, reaching 96 percent accuracy in identifying the parasternal long axis. The AI was also able to flag obscured heart chambers and enabled the segmentation of individual chambers.
“We achieved our main objective to construct a fully automated pipeline for assessment of cardiac structure, function and disease detection,” Deo et al. wrote, noting the pipeline is fully scalable and represents the first system of its kind. “Because our primary motivation is to enable low-cost serial primary care studies that otherwise would never be performed because of cost, it is important that no step requires an expert sonographer or cardiologist.”
Cardiac structure measurements obtained by the pipeline agreed with manually reported values, according to the study—measurements of left ventricular mass, left ventricular diastolic volume and left atrial volume deviated at most 15 to 17 percent. Median absolute deviation from commercially obtained values for longitudinal strain and automated ejection fraction was 7.5 percent and 9.7 percent, respectively.
Overall, the authors said, they found their automated measurements to be “comparable or superior” to manual measurements across 11 internal consistency metrics. They were also able to train convolutional neural networks to detect hypertrophic cardiomyopathy, cardiac amyloidosis and pulmonary arterial hypertension.
“An automated method to interpret echocardiograms could help democratize echocardiography, shifting evaluation of the heart to the primary care setting and rural areas,” Deo and colleagues wrote. “In addition to clinical use, such a method could also facilitate research and discovery by standardizing and accelerating analysis of the millions of echocardiograms archived within our medical systems.”