[Rapid 2D-3D medical image registration based on CUDA].

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Image Registration

Background:

• 2D-3D medical image registration is crucial for surgical navigation.
• Existing methods often involve computationally intensive 2D digitally reconstructed radiograph (DRR) generation.
• These traditional approaches struggle with real-time processing demands in surgery.

Purpose of the Study:

• To develop a fast and accurate similarity measure for 2D-3D registration.
• To enable real-time surgical navigation using preoperative CT and intraoperative X-ray images.
• To overcome the computational limitations of conventional DRR-based registration methods.

Main Methods:

• Proposed a new similarity measure function combining geometric features and image density characteristics.
• Utilized a mixed characteristic approach for direct 2D-3D registration without intermediate DRR generation.
• Implemented the algorithm using CUDA hardware acceleration for parallel numerical computation.

Main Results:

• Achieved fast and accurate 2D-3D registration meeting clinical requirements.
• Demonstrated a significantly reduced calculation process compared to traditional methods.
• The algorithm's suitability for highly parallel computation enables real-time surgical applications.

Conclusions:

• The proposed similarity measure offers an efficient and accurate solution for 2D-3D medical image registration.
• This method enhances the feasibility of real-time surgical navigation.
• The approach is computationally efficient and suitable for clinical implementation.