Attention Gate Based Dual-Pathway Network for Vertebra Segmentation of X-Ray Spine Images.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate segmentation of spine and vertebrae from X-ray images is crucial for medical diagnosis.
• Existing methods face challenges in precise localization and boundary identification.

Purpose of the Study:

• To develop a robust two-stage automatic segmentation framework for spine and vertebrae in X-ray images.
• To enhance the accuracy of spinal image analysis for clinical applications.

Main Methods:

• A novel Attention Gate based dual-pathway Network (AGNet) with context and edge pathways was designed.
• A multi-scale supervision mechanism and Edge aware Fusion Mechanism (EFM) were employed.
• Techniques like centralized backbone clipping and convex hull detection were utilized for refinement.

Main Results:

• The proposed AGNet demonstrated superior performance compared to state-of-the-art methods on spine and vertebrae datasets.
• The coarse-to-fine framework effectively segmented spine regions and identified individual vertebrae with clear boundaries.
• Experimental validation confirmed the method's efficacy.

Conclusions:

• The developed AGNet-based framework offers a significant advancement in automatic spine and vertebra segmentation.
• The system shows potential for integration into real-world spinal diagnosis systems.
• This approach aids in more accurate and efficient spinal image analysis.