A Benchmark Framework for the Right Atrium Cavity Segmentation From LGE-MRIs.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• The right atrium (RA) plays a crucial role in cardiac hemodynamics but is often under-evaluated in clinical diagnostics.
• Accurate segmentation of the RA cavity is essential for quantitative cardiac analysis.

Purpose of the Study:

• To develop and validate a benchmark framework for right atrium (RA) cavity segmentation using late gadolinium-enhanced magnetic resonance imaging (LGE-MRIs).
• To introduce RASnet, a novel 3D deep learning network designed to overcome challenges in RA segmentation, such as class imbalance and anatomical variability.

Main Methods:

• A two-stage strategy was employed, incorporating a novel 3D deep learning network, RASnet.
• RASnet features multi-path input, multi-scale feature fusion, Vision Transformers, context interaction, and deep supervision.
• The framework was evaluated on a large dataset of 354 LGE-MRIs.

Main Results:

• RASnet achieved state-of-the-art performance with a Dice score of 92.19% on the primary dataset.
• The network demonstrated robust generalizability on an independent dataset.
• The proposed framework establishes a new benchmark for RA cavity segmentation.

Conclusions:

• The developed framework and RASnet provide an accurate and efficient method for RA cavity segmentation.
• This advancement facilitates improved analysis in cardiac imaging applications.
• Open-source code and data are provided to promote further research and clinical adoption.