A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic resonance imaging

•A benchmark study of a global segmentation challenge conducted on the largest atrial LGE-MRI dataset.•Performed rigorous subgroup analysis and hyper-parameter tuning experiments.•U-Net achieved better performance compared to others.•2D and 3D CNN methods had comparable accuracies.•The double, sequentially used CNNs achieved superior results.