Fully automated 2D-3D registration and verification.

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Image Registration

Background:

• Clinical use of 2D-3D registration is limited by manual initialization and verification.
• Automated solutions are needed to enhance efficiency and accessibility.

Purpose of the Study:

• To develop and validate novel techniques for fully automated 2D-3D registration.
• To reduce human interaction in pose estimation and result verification for clinical applications.

Main Methods:

• Preoperative 2D template generation for automated initial pose estimation.
• Generalised Hough Transform and Gradient Difference Similarity Measure for 3D pose selection.
• Similarity value and pose agreement algorithms for automated registration verification.

Main Results:

• The fully automated system achieved a 95.73% correct registration rate for CT to fluoroscopy.
• Zero incorrect registration rate was observed, with 4.27% of cases producing no result.
• The system successfully detected images outside its operating range.

Conclusions:

• The proposed automated 2D-3D registration techniques significantly reduce manual intervention.
• This automation facilitates wider clinical adoption of 2D-3D registration, particularly for vertebra-based systems.
• The system demonstrates robust performance on low-dose, high-noise fluoroscopy images.