BEA-CACE: branch-endpoint-aware double-DQN for coronary artery centerline extraction in CT angiography images

Yuyang Zhang1, Gongning Luo1, Wei Wang2

  • 1Faculty of Computing, Harbin Institute of Technology, Harbin, 150001, China.

Insights

This study introduces an automated coronary centerline extraction method using deep reinforcement learning and a 3D CNN. The approach successfully tracks branches, navigates plaques, and accurately detects endpoints with minimal user input.

Area of Science:

  • Medical Imaging
  • Computational Cardiology
  • Artificial Intelligence in Medicine

Background:

  • Automating coronary tree centerline extraction is crucial for cardiovascular analysis.
  • Existing methods face challenges in automatic branch tracking, plaque traversal, and endpoint detection.

Purpose of the Study:

  • To develop an automated method for coronary centerline extraction.
  • To address challenges in tracking branches, navigating plaques, and detecting endpoints.

Main Methods:

  • A novel framework combining a deep reinforcement learning-based tracker and a 3D dilated Convolutional Neural Network (CNN) detector.
  • The tracker predicts agent actions for centerline tracking.
  • The detector identifies bifurcations and endpoints, aiding tracking and estimating vessel radius.

Main Results:

  • Achieved state-of-the-art performance in coronary centerline extraction and radius estimation.
  • Demonstrated minimal user interaction, outperforming other interactive methods.
  • Successfully tracked branches, navigated plaques, and accurately detected endpoints.

Conclusions:

  • The proposed method effectively automates coronary centerline extraction.
  • Requires only a single seed for complete tree extraction, unlike multi-seed interactive methods.
  • Offers a robust solution for accurate and efficient coronary tree analysis.
Abstract