Intraoperative Registration by Cross-Modal Inverse Neural Rendering

Area of Science:

• Medical Imaging
• Computer Vision
• Neurosurgery

Background:

• Accurate 3D/2D intraoperative registration is crucial for neurosurgical navigation.
• Existing methods face challenges in real-time accuracy and adaptability.

Purpose of the Study:

• To develop a novel cross-modal inverse neural rendering approach for 3D/2D intraoperative registration in neurosurgery.
• To enhance the precision and reliability of surgical navigation systems.

Main Methods:

• Separating implicit neural representation into anatomical structure and appearance components.
• Utilizing a multi-style hypernetwork to control appearance within a Neural Radiance Field.
• Employing a differentiable rendering engine for surgical camera pose estimation.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art registration techniques.
• The approach meets current clinical standards for registration accuracy in neurosurgery.
• Validation was performed on retrospective clinical case data.

Conclusions:

• The novel neural rendering approach offers a significant advancement in 3D/2D intraoperative registration for neurosurgery.
• This method has the potential to improve surgical outcomes through enhanced navigation.
• The developed technique provides a robust and accurate solution for real-time surgical guidance.