Centerline depth world for left atrial appendage orifice localization using reinforcement learning.

Area of Science:

• Medical Imaging
• Cardiovascular Interventions
• Artificial Intelligence

Background:

• Left atrial appendage (LAA) occlusion (LAAO) is crucial for stroke prevention in non-valvular atrial fibrillation.
• Accurate localization of the LAA orifice in preoperative CT angiography is vital for successful LAAO implant selection and C-arm angulation.
• Existing computational methods for LAA orifice localization have limitations due to anatomical variations and unclear orifice orientation.

Purpose of the Study:

• To develop an effective and efficient computational method for LAA orifice localization.
• To address the challenges posed by LAA anatomical variations and the small orifice size within large CT volumes.
• To improve the accuracy and speed of LAA orifice localization for preprocedural planning.

Main Methods:

• Proposed a centerline depth-based reinforcement learning (RL) approach for LAA orifice localization.
• An RL agent navigates the LAA centerline using centerline-to-surface distance, reducing the search space.
• The method focuses on learning an effective localization model despite the small orifice structure within CT volumes.

Main Results:

• Achieved high localization accuracy comparable to expert annotations.
• Demonstrated significant efficiency, with localization taking approximately 7.3 seconds, 18 times faster than existing methods.
• The RL agent effectively navigated the reduced search space for precise orifice identification.

Conclusions:

• The proposed RL-based method offers a highly accurate and efficient solution for LAA orifice localization.
• This approach can significantly aid physicians in preprocedural planning for LAAO, improving patient outcomes.
• The method overcomes limitations of previous techniques by effectively handling anatomical variations and optimizing the search space.