Lightweight U-Net for Blood Vessels Segmentation in X-Ray Coronary Angiography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Diagnostics

Background:

• Accurate blood vessel segmentation in X-ray coronary angiography (XCA) is vital for diagnosing cardiovascular diseases.
• Challenges include low signal-to-noise ratio, background interference, and vessel bifurcations, impacting deep learning model accuracy.
• Current deep learning models demand high computational resources, hindering real-time clinical use.

Purpose of the Study:

• To develop a computationally efficient and accurate deep learning model for blood vessel segmentation in XCA.
• To address the limitations of existing models regarding resource requirements and segmentation accuracy.

Main Methods:

• Proposed a lightweight U-Net architecture utilizing structured kernel pruning inspired by the Lottery Ticket Hypothesis.
• Systematically removed entire convolutional filters based on a global reduction factor to create compact subnetworks.
• Evaluated the pruned model on two benchmark datasets for XCA blood vessel segmentation.

Main Results:

• Achieved significant model compression, reducing parameters from 31 million to 1.9 million.
• Demonstrated consistent improvements in segmentation accuracy compared to the standard U-Net model.
• Maintained high segmentation quality despite substantial reduction in model complexity and computational requirements.

Conclusions:

• The proposed lightweight U-Net with structured kernel pruning offers an efficient solution for blood vessel segmentation in XCA.
• This approach enhances practical clinical applicability by reducing computational demands while preserving or improving accuracy.
• The method effectively addresses key challenges in XCA image analysis for cardiovascular diagnostics.