Vertebrae segmentation in reduced radiation CT imaging for augmented reality applications

Area of Science:

• Medical Imaging
• Surgical Navigation
• Radiology

Background:

• Augmented reality (AR) navigation shows promise in spinal surgery for enhancing accuracy and outcomes.
• Current AR applications often rely on intraoperative CT scans, but pre-operative scans may offer advantages.
• Accurate auto-segmentation of preoperative CT images is crucial for AR-guided spinal surgery.

Purpose of the Study:

• To investigate the feasibility of using reduced radiation dose preoperative computed tomography (pCT) images for AR auto-segmentation in spinal surgery.
• To determine the minimum image quality threshold of pCT scans required for successful AR auto-segmentation.
• To evaluate the impact of dose reduction and denoising algorithms on segmentation quality.

Main Methods:

• Systematically reduced pCT radiation dose by adding Gaussian noise, simulating lower doses with a scalar multiplier (L).
• Determined the image quality threshold for auto-segmentation by varying noise levels.
• Applied denoising algorithms to assess their effect on segmentation performance and image quality.

Main Results:

• Successful auto-segmentation for AR applications was achieved with pCT images at a simulated dose of L=0.06 (144 mAs), below current clinical thresholds.
• Segmentation failed at a lower simulated dose of L=0.09 (136 mAs).
• Denoising algorithms increased image artifacts and reduced bone density, negatively impacting segmentation.

Conclusions:

• The image quality required for AR auto-segmentation in spinal surgery is achievable with significantly reduced pCT radiation doses.
• A reduced radiation dose protocol of approximately 140 mAs is recommended for pCT scans used in AR spinal surgery.
• Further research is needed to establish precise, clinically relevant radiation dose thresholds for AR-guided surgical navigation.