A novel network with enhanced edge information for left atrium segmentation from LGE-MRI

Area of Science:

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

Background:

• Accurate left atrium (LA) segmentation is vital for cardiac interventions like atrial fibrillation ablation.
• Low contrast in medical images and target size imbalance pose significant challenges for precise LA boundary delineation.

Purpose of the Study:

• To develop an advanced deep learning architecture for improved LA segmentation from late gadolinium enhancement magnetic resonance imaging.
• To enhance edge information extraction and address class imbalance issues in LA segmentation.

Main Methods:

• Introduction of an Edge Information Enhancement Module (EIEM) to a foundational network to augment edge details.
• Implementation of a Spatially Weighted Cross-Entropy loss function to prioritize edge pixels and mitigate class imbalance.

Main Results:

• The proposed method achieved an average symmetric surface distance of 0.684 mm and a Dice coefficient of 0.924 on the 2018 Atrial Segmentation Challenge dataset.
• Demonstrated significant improvement over state-of-the-art algorithms, highlighting the effectiveness of edge information enhancement.

Conclusions:

• The developed framework provides a practical solution for precise LA localization and segmentation, enhancing outcomes for irregular and discrete anatomical features.
• The model's robustness was confirmed on the Medical Segmentation Decathlon dataset, indicating generalizability across diverse clinical scenarios.