Abstract 11962: Applying a Deep-Learning Approach for Automated Quantification of Epicardial Adipose Tissue on Coronary Computed Tomography Angiography in Challenging Clinical Populations

Abstract

Introduction: Epicardial adipose tissue (EAT) is a visceral fat deposit within the pericardial sac. The automated quantification of EAT volume is possible from routine CCTA scans via a deep-learning approach. Purpose: To apply a deep-learning approach for automated segmentation of EAT from CCTA scans in challenging clinical populations to assess the real-world viability of automated segmentation. Methods: A deep-learning automated EAT segmentation tool using a 3D Residual-U-Net neural network architecture for 3D volumetric segmentation of CCTA data was created and trained on over 2500 consecutive CCTA scans performed as part of clinical care in patients with stable chest pain from 2015 onwards within the European arm of the Oxford Risk Factors And Non Invasive Imaging (ORFAN) Study. External validation was performed in 817 scans from USA ORFAN sites (Figure 1A demonstrates human vs machine segmented EAT volume for a single case). The network was then applied to sets of unseen CCTAs from the AdipoRedOx Study (UK, n=253) and the SCOTHEART trial (UK, n=1558) to test its ability to perform EAT segmentations in challenging but common patient populations: 1) recent cardiac surgery (<6 weeks post operation), 2) BMI ≥ 40, 3) CAC ≥ 400, 4) significant metallic artefact within the pericardium, and 5) a combined group of recent open-heart surgery, BMI ≥ 30 & CAC ≥ 400. Results: There was excellent correlation between machine and human expert in the within-ORFAN validation cohort (Lin's concordance correlation coefficient (CCC) = 0.972, for both whole pericardial volume and EAT volume). CCC for machine vs human experts for all five challenging but common patient groups were also excellent (range 0.955-0.962) and are presented in Figure 1B-F. Conclusions: A new deep learning network achieves excellent automated segmentation of EAT and can be used to provide this information in a highly accurate and reproducible way, even in challenging common clinical populations.