Scholarlink

In patients with Atrial Fibrillation (AF), the Left Atrial Appendage (LAA) is a known site for blood clot formation, which can lead to strokes. The LAA closure procedure is effective in reducing the risk of stroke in patients with AF, and its success hinges on the precise placement of the closure device at the LAA orifice. In this paper, we introduce a novel framework for localizing the LAA orifice, based on Deep Reinforcement Learning (DRL) with guidance from multiple LAA centerlines. Our framework initiates with sophisticated segmentation and extraction of LAA centerlines, followed by the integration of these centerlines with the agent’s estimated location determined by DRL. The proposed method is evaluated against a recent approach for LAA orifice localization and another existing DRL-based method, demonstrating superior performance with an accuracy of 2.76 mm ± 2.01. Our method is robust and can be seamlessly integrated into workstations for diagnosis and procedural planning.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Reinforcement Learning with Multiple Centerline-Guidance for Localization of Left Atrial Appendage Orifice from CT Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jongum Yoon,
  • Sunghee Jung,
  • Byunghwan Jeon

摘要

In patients with Atrial Fibrillation (AF), the Left Atrial Appendage (LAA) is a known site for blood clot formation, which can lead to strokes. The LAA closure procedure is effective in reducing the risk of stroke in patients with AF, and its success hinges on the precise placement of the closure device at the LAA orifice. In this paper, we introduce a novel framework for localizing the LAA orifice, based on Deep Reinforcement Learning (DRL) with guidance from multiple LAA centerlines. Our framework initiates with sophisticated segmentation and extraction of LAA centerlines, followed by the integration of these centerlines with the agent’s estimated location determined by DRL. The proposed method is evaluated against a recent approach for LAA orifice localization and another existing DRL-based method, demonstrating superior performance with an accuracy of 2.76 mm ± 2.01. Our method is robust and can be seamlessly integrated into workstations for diagnosis and procedural planning.