Deep Learning-based Workflow for Automatic Extraction of Atria and Epicardial Adipose Tissue on cardiac Computed Tomography in Atrial Fibrillation

BackgroundPreoperative measurements of left atrium (LA) and epicardial adipose tissue (EAT) volumes in computed tomography (CT) images have been reported to be associated with an increased risk of atrial fibrillation (AF) recurrence. We aimed to design a deep learning-based workflow to provide a reliable automatic segmentation of atria, pericardium and EAT, which can facilitate future applications in AF. MethodsA total of 157 patients with AF who underwent radiofrequency catheter ablation were enrolled in this study. The 3D U-Net models of LA, right atrium (RA) and pericardium were used to develop the pipeline of total, LA-and RA-EAT automatic segmentation. We defined the attenuation range between -190 to -30 HU as fat within the pericardium to obtain total EAT. Regions between the dilated endocardial boundaries and endocardial walls of LA or RA within the pericardium were used to detect the voxels attributed to fat, resulting in LA-EAT and RA-EAT. ResultsThe LA, RA and pericardium segmentation models achieved Dice coefficients of 0.960 {+/-} 0.010, 0.945 {+/-} 0.013 and 0.967 {+/-} 0.006, respectively. The 3D segmentation models correlated well with ground truth for LA, RA and pericardium (r=0.99 and p < 0.001 for all). For the results of EAT, LA-EAT and RA-EAT segmentation, Dice coefficients of our proposed method were 0.870 {+/-} 0.027, 0.846 {+/-} 0.057 and 0.841 {+/-} 0.071, respectively. ConclusionsOur proposed workflow for automatic LA/RA and EAT segmentation applying 3D U-Nets on CT images was reliable in patients with AF.