OR WAIT null SECS
A new study indicates a novel, 2-step deep learning algorithm developed using echocardiogram videos was able to identify left ventricular hypertrophy and the cause with greater accuracy than trained clinicians.
A new artificial intelligence (AI) tool can help improve diagnosis and management for a pair of cardiovascular conditions, according to the results of a new study.
A cohort study of more than 20,000 led by a team of physician-scientists at the Smidt Heart Institute at Cedars-Sinai, results of the study demonstrate use of a deep learning algorithm for measuring left ventricular dimensions developed using cardiac ultrasound videos could improve the diagnosis and management of hypertrophic cardiomyopathy and cardiac amyloidosis.
“These two heart conditions are challenging for even expert cardiologists to accurately identify, and so patients often go on for years to decades before receiving a correct diagnosis,” said lead investigator David Ouyang, MD, cardiologist at the Smidt Heart Institute, in a statement. “Our AI algorithm can pinpoint disease patterns that can’t be seen by the naked eye, and then use these patterns to predict the right diagnosis.”
With advances in technology, the utility of artificial intelligence in diagnosis and risk stratification has become a focus of research across specialties and subspecialties. The current study was created by Ouyang and colleagues at the Smidt Heart Institute to evaluate whether a deep learning algorithm assessing measurements of left ventricular dimension could identify patients who might benefit from additional screening for hypertrophic cardiomyopathy and cardiac amyloidosis.
The novel algorithm had a 2-step design and was trained on a data set of 17,702 echocardiogram videos from Stanford Health Care and then evaluated on held-out test cohorts from Stanford Health Care, CSMC, and Unity Imaging Collaborative. The first used a neural network trained on annotations to provide measurement predictions for every frame of the entire video to allow for beat-to-beat estimations of ventricular wall thickness and dimensions. After detection of left ventricular hypertrophy, the algorithm identifies the specific cause of the disease based on training that allowed for classification of videos based on probability of hypertension, aortic stenosis, hypertrophic cardiomyopathy, or cardiac amyloidosis as causes of ventricular hypertrophy.
Using patient data from individuals who received treatment at the Stanford Amyloid Center, Cedars-Sinai Medical Center (CSMC) Advanced Heart Disease Clinic, Stanford Center for Inherited Cardiovascular Disease, and the CSMC Hypertrophic Cardiomyopathy Clinic from 2008-2020, investigators identified a cohort of 23,745 patients with either parasternal long-axis videos or apical 4-chamber videos for inclusion in their analyses. Parasternal long-axis videos were available from 12,001 individuals treated at Stanford Health Care and 1309 treated at CSMC. Apical 4-chamber videos were available from 8084 individuals treated at Stanford Health Care and 2351 treated at CSMC.
Upon analysis, results indicate the algorithm accurately measure intraventricular wall thickness (mean absolute error [MAE], 1.2 mm; 95% CI, 1.1-1.3), left ventricular diameter (MAE, 2.4 mm; 95% CI, 2.2-2.6), and posterior wall thickness (MAE, 1.4 mm; 95% CI, 1.2-1.5). Additionally, results indicate the algorithm accurately classified cardiac amyloidosis (AUC, 0.83) and hypertrophic cardiomyopathy (AUC, 0.98) separately from other causes of left ventricular hypertrophy.
Further analysis using external data sets suggested the algorithm accurately quantified ventricular parameters (domestic: R2, 0.96; international: R2, 0.90). Investigators noted the MAEs for the domestic data set were 1.77 mm (95% CI, 1.6-1.8) for intraventricular septum thickness, 3.8 mm (95% CI, 3.5-4.0) for left ventricular internal dimension, and 1.8 mm (95% CI,1.7-2.0) for left ventricular posterior wall thickness. In the international data set the MAEs were 1.7 mm (95% CI, 1.5-2.0) for intraventricular septum thickness, 2.9 mm (95% CI, 2.4-3.3) for left ventricular internal dimension, and 2.3 mm (95% CI, 1.9-2.7) for left ventricular posterior wall thickness. Investigators also noted the algorithm accurately detected cardiac amyloidosis (AUC, 0.79) and hypertrophic cardiomyopathy (AUC, 0.89) in the domestic external validation site.
“The algorithm identified high-risk patients with more accuracy than the well-trained eye of a clinical expert,” Ouyang added. “This is because the algorithm picks up subtle cues on ultrasound videos that distinguish between heart conditions that can often look very similar to more benign conditions, as well as to each other, on initial review.”
This study, “High-Throughput Precision Phenotyping of Left Ventricular Hypertrophy With Cardiovascular Deep Learning,” was published in JAMA Cardiology.