Towards real time guide wire shape extraction in fluoroscopic sequences: A two phase deep learning scheme to extract...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate guide wire localization is crucial for image-guided interventions in cerebral and cardiovascular procedures.
• Traditional methods struggle with guide wire's non-rigid, sparse nature and low signal-to-noise ratio (SNR) fluoroscopic images, leading to inaccuracies and high computational costs.
• Deep learning shows promise but faces challenges with computational expense.

Purpose of the Study:

• To develop a novel, two-phase deep learning scheme for accurate and real-time guide wire shape extraction in fluoroscopic sequences.
• To improve upon the limitations of traditional guide wire tracking methods and existing deep learning approaches.

Main Methods:

• A two-phase deep learning strategy was implemented.
• Phase 1: A guide wire localization network identifies relevant image regions.
• Phase 2: A guide wire shape extraction network precisely marks guide wire pixels within these regions.

Main Results:

• The proposed method achieved 99% accuracy in distinguishing guide wire pixels with near-zero false positives.
• Average offset from ground truth was less than 1 pixel, even in challenging cases.
• Processing time was 78ms for a 512x512 image, meeting real-time requirements.

Conclusions:

• The deep learning method offers superior accuracy and stability compared to traditional filtering techniques.
• It significantly enhances computational efficiency, making it suitable for real-time clinical applications.
• The approach effectively addresses the limitations of existing guide wire tracking technologies.