Jigsaw training-based background reverse attention transformer network for guidewire segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Guidewire segmentation is vital for percutaneous coronary intervention.
• X-ray sequences present challenges like low signal-to-noise ratio and class imbalance.
• Existing methods often focus solely on single guidewire segmentation.

Purpose of the Study:

• To develop a method for segmenting both single and dual guidewires in X-ray fluoroscopy sequences.
• To address the limitations of current guidewire segmentation techniques.

Main Methods:

• A jigsaw training strategy was employed to train the guidewire segmentation network.
• A background reverse attention (BRA) transformer network was proposed.
• Robust principal component analysis was used to generate background maps for BRA computation.

Main Results:

• The proposed method successfully segmented single and dual guidewires in X-ray fluoroscopy sequences.
• The approach achieved higher F1 scores and precision compared to state-of-the-art methods.
• The method was validated on data from three hospitals.

Conclusions:

• The jigsaw training strategy reduces the need for dual guidewire data and boosts network performance.
• The BRA module effectively minimizes background interference and enhances guidewire distinction.
• The proposed methods demonstrate superior performance over existing state-of-the-art guidewire segmentation techniques.