Benchmarking commercial depth sensors for intraoperative markerless registration in neurosurgery applications

Area of Science:

• Neurosurgery
• Medical Imaging
• Computer Vision

Background:

• Markerless patient registration is crucial for image-guided neurosurgery.
• Depth information offers a promising avenue for improving registration accuracy.

Purpose of the Study:

• To generalize markerless patient registration using depth information in neurosurgery.
• To evaluate the performance of commercial depth cameras and registration algorithms.

Main Methods:

• Utilized a multimodal experimental setup with five depth cameras in seven configurations.
• Calculated Fiducial Registration Error (FRE) and Target Registration Error (TRE) using Iterative Closest Point (ICP) and Deep Global Registration (DGR) algorithms.
• Employed a phantom head model simulating clinical conditions for camera positioning.

Main Results:

• The D405 and Zed-M+ cameras achieved low Target Registration Error (TRE) values (2.36 ± 0.46 mm and 2.49 ± 0.35 mm, respectively).
• Cameras with texture projectors or enhanced depth refinement performed better.
• The methodology successfully characterized camera suitability for registration tasks.

Conclusions:

• An adaptable framework was validated for evaluating depth cameras in neurosurgical registration.
• D405 and Zed-M+ are identified as reliable options for image-guided neurosurgery.
• Future work will focus on hardware and algorithmic improvements for enhanced depth quality.