Multi-class Bi-atrial Segmentation from 3D Contrast-Enhanced Magnetic Resonance Imaging

Persistent or long-standing persistent atrial fibrillation (AF) is associated with progressive atrial structural remodelling, including chamber dilatation, fibrosis and atrial wall thickness variation. Left atrial (LA) fibrosis quantified with late gadolinium-enhanced (LGE) magnetic resonance imaging (MRI) or estimated via low voltage map has been used to guide adjunctive ablation beyond pulmonary vein isolation in recent clinical trials with mixed clinical outcomes. Many factors may contribute to this, such as the absence of right atrial (RA) fibrosis as a potential ablation target in these trials. Clinical studies found that a majority of patients with persistent AF have AF drivers in the RA. In addition, the reconstructed bi-atrial chambers can be used to guide ablations and extract key structural biomarkers, such as the LA and RA volume and wall thickness for patient stratification and targeted treatment alongside fibrosis. 

Expanded from our previous LA challenge in 2018 [Xiong et al., Medical Image Analysis, 2021, with 185 citations so far], this new challenge aims to test multi-class machine learning approaches for both LA and RA cavity and atrial wall directly from LGE-MRIs to improve targeted AF ablation. In addition, the proposed methods will be tested for their performances across segmentation and biomarker extraction tasks (such as the LA/RA volume and fibrosis) on cross-center LGEMRI datasets. We will use 200 multi-centre 3D LGE-MRIs for this challenge, the largest one in the field so far. More importantly, the precious data were carefully labelled in consensus with three independent experts for each LGE-MRI scan to obtain one segmentation per scan. 

The developed AI approaches and clinical pipelines may be transferrable to other challenging medical tasks. The proposed challenge and data are a paradigm shift for cardiac structural analysis and may accelerate the search for optimised ablation strategies for patients with persistent and long-standing persistent AF.