A Lightweight Network for Accurate Coronary Artery Segmentation Using X-Ray Angiograms

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Diagnostics

Background:

• Accurate segmentation of coronary arteries in X-ray angiograms is crucial for diagnosing coronary artery disease.
• Current deep learning models often prioritize accuracy over computational efficiency, requiring high-performance hardware.
• This necessitates the development of computationally efficient yet accurate segmentation methods.

Purpose of the Study:

• To develop a lightweight deep learning network for automated coronary artery segmentation.
• To achieve a balance between high segmentation accuracy and reduced computational cost.
• To improve the clinical applicability of automated coronary artery segmentation in resource-constrained environments.

Main Methods:

• Designed a lightweight U-Net architecture using bottleneck residual blocks.
• Incorporated spatial and channel attention modules to capture long-range dependencies.
• Employed Top-hat transforms and contrast-limited adaptive histogram equalization (CLAHE) for image pre-processing.

Main Results:

• Achieved high performance metrics: sensitivity (0.8770), specificity (0.9789), accuracy (0.9729), and AUC (0.9910).
• The proposed network has only 0.75 M parameters, significantly fewer than traditional U-Net (31.04 M).
• Demonstrated strong generalization capabilities on external coronary angiogram datasets.

Conclusions:

• The developed lightweight network provides an effective solution for accurate and computationally efficient coronary artery segmentation.
• The network's low parameter count and strong performance make it suitable for wider clinical adoption.
• The robust generalization indicates its reliability across diverse datasets and imaging conditions.