Segmenting Bi-Atrial Structures Using ResNext Based Framework

2025-02-28

AI Summary

This work addresses the challenges of automatic segmentation of right and left atrial (RA/LA) walls and cavities in late gadolinium enhancement MRI (LGE-MRI) of atrial fibrillation patients—namely, low segmentation accuracy and poor generalizability of existing methods. We propose a novel two-stage ResNeXt-based segmentation framework that achieves synchronized, high-precision dual-atrial segmentation: Stage I performs coarse localization of atrial regions, while Stage II refines wall–cavity differentiation. To enhance multi-scale feature representation, we integrate a ResNeXt encoder; to mitigate severe class imbalance between wall and cavity tissues, we adopt a cyclical learning rate schedule. Evaluated on multi-center LGE-MRI data, our method significantly outperforms U-Net and other baselines, achieving over 8% improvement in Dice score for atrial wall segmentation. Moreover, it demonstrates superior generalizability and robustness across diverse clinical sites and scanner protocols.

Technology Category

• Computer Vision: Segmentation
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Machine Learning: Multimodal Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Atrial fibrillation (AF) is the most common cardiac arrhythmia, significantly contributing to mortality, particularly in older populations. While pulmonary vein isolation is a standard treatment, its effectiveness is limited in patients with persistent AF. Recent research highlights the importance of targeting additional atrial regions, particularly fibrotic areas identified via late gadolinium-enhanced MRI (LGE-MRI). However, existing manual segmentation methods are time-consuming and prone to variability. Deep learning techniques, particularly convolutional neural networks (CNNs), have shown promise in automating segmentation. However, most studies focus solely on the left atrium (LA) and rely on small datasets, limiting generalizability. In this paper, we propose a novel two-stage framework incorporating ResNeXt encoders and a cyclic learning rate to segment both the right atrium (RA) and LA walls and cavities in LGE-MRIs. Our method aims to improve the segmentation of challenging small structures, such as atrial walls while maintaining high performance in larger regions like the atrial cavities. The results demonstrate that our approach offers superior segmentation accuracy and robustness compared to traditional architectures, particularly for imbalanced class structures.

Problem

• Automates segmentation of bi-atrial structures in LGE-MRIs
• Addresses variability and time constraints in manual methods
• Improves accuracy in segmenting small structures like atrial walls

Innovation

• ResNeXt encoders for bi-atrial segmentation
• Cyclic learning rate enhances model training
• Two-stage framework improves small structure accuracy