Left atrial volumetric assessment using a novel automated framework for 3D echocardiography: a multi-centre analysis

Nuno Almeida, Alexandros Papachristidis, Peter Pearson, Sebastian Imre Sarvari, Jan Engvall, Thor Edvardsen, Mark Monaghan, Olivier Gérard, Eigil Samset, Jan D'hooge

This study aims at validating a software tool for automated segmentation and quantification of the left atrium (LA) from 3D echocardiography. The LA segmentation tool uses a dual-chamber model of the left side of the heart to automatically detect and track the atrio-ventricular plane and the LA endocardium in transthoracic 3D echocardiography. The tool was tested in a dataset of 121 ultrasound images from patients with several cardiovascular pathologies (in a multi-centre setting), and the resulting volumes were compared with those assessed manually by experts in a blinded analysis using conventional contouring. Bland-Altman analysis showed good agreement between the automated method and the manual references, with differences (mean ± 1.96 SD) of 0.5 ± 5.7 mL for LA minimum volume and -1.6 ± 9.7 mL for LA maximum volume (comparable to the inter-observer variability of manual tracings). The automated tool required no user interaction in 93% of the recordings, while 4% required a single click and only 2% required contour adjustments, reducing considerably the amount of time and effort required for LA volumetric analysis. The automated tool was validated in a multi-centre setting, providing quantification of the LA volume over the cardiac cycle with minimal user interaction. The results of the automated analysis were in agreement with those estimated manually by experts. This study shows that such approach has clinical utility for the assessment of the LA morphology and function, automating and facilitating the time-consuming task of analysing 3D echocardiographic recordings.