A video-based deep neural network (DNN) may be used to estimate left ventricular (LV) systolic function using left coronary angiogram (LCA) videos, a new analysis suggests. These data were reported by Robert Avram, MD, MSc, of the University of California-San Francisco, and colleagues from the U.S. and Canada in a manuscript published Wednesday online in JAMA Cardiology. Coronary angiography is the standard of care assessment for evaluating those with coronary heart disease, which is the No. 1 cause of death in adults worldwide. LV ejection fraction (EF) is interpreted to determine management of the disease and joint patient decision making. The current modalities of measuring LVEF are not always available, and some patients require more urgent coronary angiography than what is available. This study evaluated DNN as an automated approach to find LVEF from left coronary angiograms. The cross-sectional study used patient data from Dec. 12, 2012, to Dec.31, 2019, from the University of California-San Francisco. Data sets were separated into training, development and tests. The University of Ottawa Heart Institute contributed the external validation data. CathEF, a video-based DNN, was used to determine the (binary) reduced LVEF (≤40% or >40%) and predict (continuous) the percentage of LVEF from standard angiogram videos of the LCA. GradCam (guided class-discriminative gradient class activation mapping) was used to visualize pixels in angiograms contributing to most prediction of DNN LVEF. A total of 3,679 patients (mean age=64.3 years, 65% male) and 4,042 adult angiograms with corresponding transthoracic echocardiogram (TTE) LVEF were included in the study. The test data set (n=813) showed video-based DNN discriminated (binary) reduced LVEF (≤40%) with an area under the receiver operating characteristic curve (AUROC) of 0.911 (95% confidence interval [CI]=0.887-0.934); the diagnostic odds ratio for reduced LVEF was 22.7 (95% CI=14.0-37.0). The mean absolute error (MAE) of DNN-predicted continuous LVEF was 8.5% (95% CI=8.1%-9.0%) compared with TTE LVEF. DNN-predicted continuous LVEF was different in 5% or fewer studies, compared with TTE LVEF, which showed in 38.0% (309 of 813) of test data set studies. In 15.2% of test data set studies, researchers observed differences greater than 15% (124 of 813). The external validation (n=776) showed video-based DNN discriminated (binary) reduced LVEF (≤40%) with an AUROC of 0.906 (95% CI=0.881-0.931). An MAE of 7.0% was seen in DNN-predicted continuous LVEF (95% CI, 6.6%-7.4%). Video-based DNN leaned toward overestimating low LVEFs and underestimating high LVEFs. Video-based DNN performance did not change across sex, body mass index, low estimated glomerular filtration rate, presence of acute coronary syndromes, obstructive coronary artery disease, and left ventricular hypertrophy. The investigators noted a few limitations of this study, including that DNN tended to overestimate LVEF values greater than 65% and underestimate LVEF values less than 30%. Thus, caution should be taken in using DNN in patients who have a very high or very low LVEF. Another important limitation included that this is an early study in using DNN to predict LVEF, and further research is warranted for clinical applications. Overall, DNN can be used to estimate LVEF, but in extremes should be used with caution, the authors concluded Source: 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. 2023 May 10 (Article in press). Image Credit: Sergie – stock.adobe.com