Characterizing Low-cost Registration for Photographic Images to Computed Tomography

Area of Science:

• Medical Imaging
• Computer Vision
• 3D Reconstruction

Background:

• Accurate mapping of photographic images to volumetric medical scans (e.g., CT) is vital for image-guided surgery and linking physical spaces.
• Current mapping methods are often computationally intensive or require specialized hardware, limiting their accessibility for general 3D mapping of bulk specimens.

Purpose of the Study:

• To evaluate a cost-effective solution for 3D mapping of bulk specimens in histological processing.
• To compare the integration of a commercial 3D camera and smartphone imaging within a surface registration pipeline for medical imaging applications.

Main Methods:

• Utilized surgical implants and a phantom (chuck-eye steak) for testing.
• Acquired 3D CT reconstructions and photographic images from a Canfield H1 camera and an iPhone 14 Pro.
• Performed surface reconstruction using commercial tools and Neural Radiance Fields (NeRF), followed by Iterative Closest Point (ICP) registration.

Main Results:

• Registration of Canfield camera surfaces yielded landmark distance errors of 1.747, 3.932, and 1.692 mm.
• Registration of iPhone 14 Pro surfaces resulted in errors of 1.222, 2.061, and 5.155 mm.
• Demonstrated that photographic imaging before tissue sectioning offers a low-cost method for correlating histological samples with 3D anatomical data.

Conclusions:

• Both commercial 3D cameras and smartphones can be effectively integrated into surface registration pipelines for 3D mapping.
• Smartphone imaging presents a viable, low-cost alternative for establishing correspondence between histological samples and 3D anatomical data, enhancing accessibility in research and clinical settings.