183 A Novel Technique for Neuro-Navigation During Ventricular Endoscopy: Volumetric 3D Reconstruction Based on the 2D Ventricular Endoscopy Video
INTRODUCTION: Ventricular endoscopy is a common procedure in neurosurgery. Traditional neuro-navigation techniques require pre-operative registration and do not account for deformation and shift due to cerebrospinal fluid loss intra-operatively. Artificial intelligence-based interpretation of the video feed could serve as a tool to account for deformation, and provide 3D video reconstruction for adaptive neuro-navigation without conventional tracking. This technology further serves as a foundation for additional capabilities such as augmented reality overlay in the endoscopic scene. METHODS: Thirty patients were enrolled who underwent endoscopic third ventriculostomy and had available brain CT or MR imaging. Segmentation of ventricular and deep brain structures was performed on the preoperative imaging to serve as ground truth. High-resolution intraoperative video was collected, and adapted Neural Radiance Fields (NeRF) methods were applied to the endoscopic video post-operatively. The resulting 3D video reconstruction was registered to preoperative imaging, and segmented structures were overlaid as semi-transparent visual cues in the registered video stream. Geometric accuracy was evaluated in terms of Projected Error (PE). RESULTS: The NeRF-based volumetric 3D reconstructions aligned well with ground truth pre-operative MR (or CT) imaging, with a PE of 0.43 mm (inter-quartile range 0.46 mm), giving an augmented video overlay with submillimeter accuracy. The methods demonstrated a fast, one-time optimization runtime (< 120 s for evaluated scenarios) followed by real-time rendering, which is promising for the initial research prototype and can be accelerated as an online method for eventual clinical use. CONCLUSIONS: This study demonstrates proof-of-concept that real-time ventricular endoscopic video can be used to create accurate 3D reconstructions of ventricular and brain anatomy, allowing for deformation correction, a novel basis for neuro-navigation, and augmented reality overlay.
Authors
- Ryan P. Lee (ORCID: https://orcid.org/0000-0002-2517-9559)
- Prasad Vagdargi (ORCID: https://orcid.org/0000-0002-2740-6604)
- Jeffrey H. Siewerdsen (ORCID: https://orcid.org/0000-0001-6644-1317)
- Mark G. Luciano (ORCID: https://orcid.org/0000-0003-1456-9559)
- Ali Uneri (ORCID: https://orcid.org/0000-0003-3419-1805)
Publication Details
- Journal
- Neurosurgery
- Published
- 2025-03-14
- DOI
- https://doi.org/10.1227/neu.0000000000003360_183
- Citations
- 1
- Primary Topic
- Cerebrospinal fluid and hydrocephalus
- Type
- article
- Field-Weighted Citation Impact
- 0.91