GoRG: Towards a GPU-Accelerated Multiview Hyperspectral Depth Estimation Tool for Medical Applications

Area of Science:

• Medical Imaging
• Computer Vision
• Neurosurgery

Background:

• HyperSpectral (HS) imaging aids brain tumor boundary detection.
• Integrating HS with MRI and IOUS creates unified 3D models for neurosurgery.
• Real-time 3D model generation is crucial for surgical guidance.

Purpose of the Study:

• To develop a real-time, GPU-accelerated tool for hyperspectral image depth estimation.
• To enable the creation of accurate 3D immersive models for neurosurgical applications.
• To assess the performance of the tool in operating room conditions.

Main Methods:

• Introduction of Graph cuts Reference depth estimation in GPU (GoRG).
• GPU-accelerated multiview depth estimation for HS and YUV images.
• Real-time processing capability (under 5.5s for YUV, 850ms/frame for HS).

Main Results:

• GoRG YUV shows minimal quality loss compared to MPEG DERS.
• GoRG achieves an average RMSE of 7.5 cm for HS images.
• Processing times demonstrate feasibility for intraoperative use.

Conclusions:

• GoRG enables real-time 3D model generation from HS images for neurosurgery.
• The tool effectively integrates multimodal imaging data for enhanced surgical planning.
• Feasibility demonstrated for intraoperative application in tumor resection.