Angiograms using video-based deep neural networks may predict LVEF: JAMA
USA: Left ventricular ejection fraction (LVEF) can be estimated from commonly obtained coronary angiograms using video-based deep neural networks (DNNs), researchers state in a recent study published in JAMA Cardiology. They, however, warrant caution in interpreting DNN estimates at LVEF extremes.
In the cross-sectional study of 4042 adult angiograms matched with corresponding transthoracic echocardiograms (TTEs) from 3679 patients, a DNN was trained to estimate reduced LVEF (≤40%) and continuous LVEF. The DNN demonstrated good reduced left ventricular ejection fraction discrimination using angiogram videos compared with a transthoracic echocardiogram.
Previous studies have shown that understanding left ventricular ejection fraction during coronary angiography can assist in disease management. Therefore, Robert Avram, University of California, San Francisco, Cardiology, San Francisco, and colleagues aimed to develop an automated approach to predict LVEF from left coronary angiograms.
They determined if deep neural networks predict LVEF from angiograms compared with a transthoracic echocardiogram–measured LVEF.
The study with external validation used patient data from 2012 to 2019 from the University of California, San Francisco (UCSF). Data were randomly split into development, training, and test data sets. External validation data were acquired from the University of Ottawa Heart Institute.
The researchers included patients 18 years and above who received a coronary angiogram and transthoracic echocardiogram within three months before or one month after the angiogram.
CathEF, a video-based DNN, was used to discriminate reduced LVEF and to predict (continuous) LVEF percentage from standard angiogram videos of the left coronary artery. Pixels were visualized in angiograms using GradCAM (guided class-discriminative gradient class activation mapping), which contributed most to the prediction of DNN LVEF.
In the analysis, a total of 4042 adult angiograms with corresponding TTE LVEF from 3679 UCSF patients were included. The mean patient age was 64.3 years, and 65% of patients were male.
The study revealed the following findings:
- In the UCSF test data set (n = 813), the video-based deep neural networks discriminated (binary) reduced LVEF (≤40%) with an area under the receiver operating characteristic curve (AUROC) of 0.911; the diagnostic odds ratio for reduced LVEF was 22.7.
- DNN-predicted continuous LVEF had a mean absolute error (MAE) of 8.5% compared with TTE LVEF.
- Although DNN-predicted continuous LVEF differed by 5% or less compared with TTE LVEF in 38.0% of test data set studies, differences greater than 15% were observed in 15.2%.
- In external validation (n = 776), video-based DNN discriminated (binary) reduced LVEF (≤40%) with an AUROC of 0.906, and DNN-predicted continuous LVEF had an MAE of 7.0%.
- Video-based DNN tended to overestimate low LVEFs and underestimate high LVEFs.
- Video-based DNN performance was consistent across sex, body mass index, low estimated glomerular filtration rate (≤45), obstructive coronary artery disease, acute coronary syndromes, and left ventricular hypertrophy.
"This cross-sectional study represents an early demonstration of LVEF estimation from standard angiogram videos of the left coronary artery using video-based DNNs," the researchers wrote. "Further research can improve accuracy and reduce DNNs variability to maximize clinical utility."
Reference:
Avram R, Barrios JP, Abreau S, et al. Automated Assessment of Cardiac Systolic Function From Coronary Angiograms With Video-Based Artificial Intelligence Algorithms. JAMA Cardiol. Published online May 10, 2023. doi:10.1001/jamacardio.2023.0968
