MBAS2024: A large-scale benchmark for multi-class bi-atrial segmentation in multi-center contrast-enhanced MRIs

Area of Science:

• Medical Imaging
• Computational Anatomy
• Cardiac Electrophysiology

Background:

• Atrial fibrillation (AF) is a common arrhythmia affecting millions, necessitating improved treatment strategies.
• Current benchmarks for cardiac imaging analysis primarily focus on the left atrium, neglecting the complexities of bi-atrial anatomy crucial for AF ablation.
• The thin atrial walls are critical for substrate-guided ablation planning but pose significant segmentation challenges.

Purpose of the Study:

• To introduce the first large-scale, multi-class benchmark (MBAS2024) for simultaneous segmentation of left atrial (LA) and right atrial (RA) cavities and bi-atrial walls.
• To systematically evaluate state-of-the-art segmentation methods on a large, curated bi-atrial dataset derived from late gadolinium-enhanced (LGE) MRI.
• To provide a comprehensive assessment of current segmentation capabilities and limitations for improving AF treatment.

Main Methods:

• Development of the Multi-class Bi-Atrial Segmentation 2024 Challenge (MBAS2024) dataset, featuring 175 3D multi-center LGE-MRI scans with expert-validated annotations.
• Systematic evaluation of 13 advanced segmentation methods on the MBAS2024 dataset.
• Analysis of segmentation performance concerning image quality, acquisition protocols, model architecture (U-Net, state-space models), slice position, and labeling strategies.

Main Results:

• Segmentation of LA and RA cavities demonstrates robustness to image quality, while atrial wall delineation is highly sensitive to image degradation.
• Performance variability across different centers indicates limited generalization of atrial wall segmentation.
• U-Net-based models and emerging state-space models (e.g., UMambaBot) exhibit superior accuracy and efficiency, with hybrid labeling strategies enhancing performance.

Conclusions:

• The MBAS2024 challenge establishes a foundational benchmark for bi-atrial segmentation, crucial for understanding AF pathophysiology.
• Atrial wall segmentation requires further development to overcome sensitivity to image quality and improve cross-center generalizability.
• Validated baselines and insights from MBAS2024 will guide the creation of clinically relevant, efficient, and anatomically aware segmentation algorithms for targeted AF ablation.