Key result
Deep learning on CCTA predicts LVEF <50% with ~94% accuracy versus echocardiography.
Why the study?
Accurate LV segmentation is essential for cardiac function assessment, prompting the development and evaluation of a deep learning method using an 8-layer residual U-Net to calculate LVEF from CCTA compared to conventional echocardiography.
Does a deep learning method (AccuLV) on CCTA images accurately calculate LVEF compared to conventional 2D echocardiography in patients undergoing both tests?
Cross-Sectional (n=180)
No
Does a deep learning method (AccuLV) on CCTA images accurately calculate LVEF compared to conventional 2D echocardiography in patients undergoing both tests?
Effect estimate: r=0.82
Absolute Event Rate: 73% vs 64%
p-value: p=<0.001
A fully automated deep learning method (AccuLV) applied to CCTA images provides accurate quantification of LVEF that correlates well with conventional 2D echocardiography.
May support automated CCTA-based LVEF assessment when echocardiography is limited; leaves open prospective validation before routine use.
Background: Accurate segmentation of the left ventricle (LV) is an important step in assessing cardiac function. Cardiac CT angiography (CCTA) has become an important means of clinical diagnosis of cardiovascular diseases (CVDs) because of its advantages of non-invasive, short examination time and low cost. In order to obtain the segmentation of LV in CCTA scans, we propose a deep learning method based on 8-layer residual U-Net with deep supervision. In this study we compared the left ventricular ejection fraction (LVEF) calculated by deep learning (DL) method (AccuLV) from CCTA to LVEF by conventional two-dimensional echocardiography (EC). Methods: This was a retrospective cross-sectional study, and consecutive patients who had undergone CCTA and EC in our hospital from February 2021 to May 2021 were recruited. The current study included 180 patients who had undergone CCTA and EC. To obtain LVEF, we used an 8-layer residual U-Net with deep supervision to segment LV contours from CCTA images and compute LVEF (DL-LVEF). The EC and DL-LVEF measurements were compared. A 50% EC-LVEF cut-off value was used as a reference standard to assess the diagnostic performance of AccuLV in assessing LV function. Results: The overall mean EC-LVEF and DL-LVEF values were 64.0% (52.3%, 69.0%) and 73.0% (52.3%, 77.0%), respectively. Three patient groups were studied: (I) hypertensive patients, (II) postmenopausal women, and (III) diabetes. EC-LVEF and DL-LVEF were found to be positively correlated for all of the included patients (r=0.82, P<0.001), with the detailed results for the three groups as follows: hypertensive patients (r=0.77, P<0.001), postmenopausal women (r=0.92, P<0.001) and diabetes (r=0.88, P<0.001). The diagnostic accuracy, sensitivity, and specificity of the DL method in predicting EC-LVEF <50% for all patients were 93.9%, 92.3%, and 94.3%, and for hypertensive patients were 95.4%, 93.8%, and 95.8%, for postmenopausal women were 87.0%, 100%, and 84.2%, for diabetes were 97.4%, 100%, and 96.6%. Conclusions: In comparison to echocardiography, which is commonly used in clinical setting, AccuLV may be a promising, fully automated tool for rapid and accurate quantification of LV function and thus for making reliable clinical decisions.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2023) conducted a cross-sectional in Patients undergoing CCTA and echocardiography (n=180). Deep learning method (AccuLV) from CCTA vs. Conventional two-dimensional echocardiography (EC) was evaluated on Left ventricular ejection fraction (LVEF) (r=0.82, p=<0.001). Deep learning-based LVEF calculation from CCTA strongly correlated with echocardiography (r=0.82, P<0.001) and predicted LVEF <50% with 93.9% accuracy, 92.3% sensitivity, and 94.3% specificity.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: