A novel 2D/3D hierarchical registration framework via principal-directional Fourier transform operator

Area of Science:

• Medical Imaging
• Computer Vision
• Image Registration

Background:

• Accurate registration between preoperative 3D computed tomography (CT) and intraoperative 2D X-ray images is vital for image-guided therapy.
• Existing methods may require specific initial values or have limited capture ranges, impacting clinical workflow efficiency.

Purpose of the Study:

• To propose a novel 2D/3D hierarchical registration framework, termed HRF-PDFTO, for enhanced accuracy and efficiency.
• To address the limitations of initial value assignment and expand the capture range in medical image registration.

Main Methods:

• Development of a principal-directional Fourier transform operator (PDFTO) for translation and rotation invariance.
• Implementation of an initial-free template-matching approach leveraging PDFTO to avoid manual initialization.
• Design of a hierarchical registration strategy to reduce the search space dimensions from n^6 to n^2.

Main Results:

• The HRF-PDFTO framework demonstrated high registration accuracy, achieving 0.72 mm.
• The registration process was highly efficient, with a single registration time of 16 seconds.
• The proposed method showed a tenfold improvement in registration efficiency compared to existing approaches.

Conclusions:

• The HRF-PDFTO framework effectively meets the accuracy and efficiency requirements for 2D/3D registration in clinical applications.
• This novel approach offers a promising solution for improving image-guided therapy through precise and rapid image registration.