A fast, accurate, and automatic 2D-3D image registration for image-guided cranial radiosurgery

Area of Science:

• Medical Imaging
• Radiosurgery
• Image Registration

Background:

• Precise patient setup is crucial for effective image-guided cranial radiosurgery.
• Real-time detection and correction of patient movement during treatment are necessary.
• Existing methods may lack the speed or accuracy required for frequent adjustments.

Purpose of the Study:

• To develop a fast and accurate two-dimensional (2D)-three-dimensional (3D) image registration method.
• To enable precise initial patient setup and continuous correction of patient movement during cranial radiosurgery.

Main Methods:

• A multiphase framework registers digitally reconstructed radiographs (DRRs) with acquired x-ray images.
• 3D rigid transformation is decomposed into 2D in-plane and out-of-plane components for independent registration.
• Combines results using 2D-3D geometric backprojection, employing optimized pattern intensity and sum of squared difference similarity measures.

Main Results:

• Registration errors were minimal: 0.33 +/- 0.16 mm (translation) and 0.29 +/- 0.11 degrees (rotation) against fiducial registration.
• Total targeting errors were below 0.51 mm even for targets 10 cm from the rotation center.
• Computation time was under 3 seconds on a standard dual-processor computer.

Conclusions:

• The developed 2D-3D image registration method is fast, accurate, and suitable for image-guided cranial radiosurgery.
• It effectively addresses the need for precise patient positioning and real-time movement correction.
• The method's efficiency and accuracy contribute to improved patient safety and treatment outcomes.