Endovascular Tool Segmentation with Multi-lateral Branched Network during Robot-assisted Catheterization.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Robotics

Background:

• Robot-assisted catheterization is crucial for cardiovascular disease intervention.
• Effective endovascular tool navigation relies on visualization and tracking.
• Current methods lack real-time motion analytics and struggle with poor fluoroscopy illumination.

Purpose of the Study:

• To propose a novel Multi-Lateral Branched Network (MLB-Net) for improved tool segmentation in cardiovascular angiograms.
• To address limitations in current tool segmentation methods caused by poor illumination and lack of real-time analytics.

Main Methods:

• Developed a Multi-Lateral Branched Network (MLB-Net) featuring an encoder with multi-lateral separable convolutions and a pyramid decoder.
• Trained and validated the model on 1320 angiograms from robot-assisted catheterization in rabbits.
• Evaluated performance using F1-score and mean intersection-over-union metrics.

Main Results:

• Achieved an F1-score of 89.01% and a mean intersection-over-union of 90.05% on 330 validation frames.
• Demonstrated superior performance compared to state-of-the-art models like U-Net, U-Net++, and DeepLabV3.
• Confirmed the model's robustness for guidewire segmentation in angiograms.

Conclusions:

• The MLB-Net provides robust and accurate guidewire segmentation in cardiovascular angiograms.
• This model can serve as a foundation for advanced endovascular tool tracking and surgical scene analytics.
• Improved segmentation accuracy can enhance the safety and efficacy of cardiovascular interventions.