Portal de Periódicos da CAPES

TESSLA: Two-Stage Ensemble Scar Segmentation for the Left Atrium

2023; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-031-31778-1_10

1611-3349

Shaheim Ogbomo-Harmitt, Jakub Grzelak, Ahmed Qureshi, Andrew P. King, Oleg Aslanidi,

Advanced MRI Techniques and Applications

Atrial fibrillation (AF) is the most common cardiac arrhythmia worldwide; however, the current success rates for catheter ablation (CA) therapy, the first-line treatment for AF, are suboptimal. Therefore, extensive research has focused on the relationship between scar tissue in the left atrium (LA) and AF, and its application for patient stratification and more effective CA therapy strategies. However, quantifying and segmenting LA scar tissue requires significant data pre-processing from well-trained clinicians. Hence, deep learning (DL) has been proposed to automatically segment the LA fibrotic scar from late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) images. Segmenting LA scar with DL is challenging as fibrosis from LGE-CMR images has a relatively small volume and regions surrounding the scar are also enhanced. Therefore, we propose a two-stage ensemble DL model (TESSLA: two-stage ensemble scar segmentation for the LA) that segments the blood pool of the LA, estimates the LA wall, applies an image intensity ratio with Z-score normalisation and combines a scar segmentation from two independent networks. TESSLA outperformed its constituent models and achieved state-of-art accuracy on the LAScar 2022 challenge evaluation platform for LA scar segmentation with a Dice score of 0.63 ± 0.14 and a Dice score of 0.58 ± 0.11 for the final test phase. Our workflow provides a fully automatic estimation of LA fibrosis from clinical LGE CMR scans.