VDVM: An automatic vertebrae detection and vertebral segment matching framework for C-arm X-ray image identification

Area of Science:

• Medical Imaging
• Spine Surgery Technology
• Artificial Intelligence in Healthcare

Background:

• C-arm fluoroscopy is crucial for precise spine surgery, but identifying surgical locations relies heavily on surgeon experience when comparing C-arm X-ray and digital radiography (DR) images.
• Current methods for vertebrae identification in C-arm fluoroscopy are experience-dependent, highlighting a need for automated solutions.

Purpose of the Study:

• To develop an automated framework for vertebrae detection and vertebral segment matching (VDVM) to enhance the identification of vertebrae in C-arm X-ray images.
• To improve the precision and reduce the reliance on surgeon experience in spine surgery planning and execution.

Main Methods:

• A VDVM framework was designed, comprising vertebra detection and matching.
• Image quality was enhanced using data preprocessing. Vertebrae detection was performed using the YOLOv3 model, followed by region extraction.
• Vertebrae segmentation utilized the Mobile-Unet model, with contour inclination correction and a multi-vertebra strategy for matching based on visual information fidelity.

Main Results:

• The vertebra detection model achieved a mean Average Precision (mAP) of 0.87 on C-arm X-ray test datasets and 0.96 on lumbar DR test datasets.
• The VDVM framework achieved a vertebral segment matching accuracy of 0.733 on 31 C-arm X-ray images.
• The system demonstrated robust performance in both vertebrae detection and segment matching.

Conclusions:

• The proposed VDVM framework effectively automates vertebrae detection and achieves significant accuracy in vertebral segment matching.
• This automated approach has the potential to enhance precision in spine surgery and reduce the dependency on surgeon's experience.