Surgeons performing endoscopic sinus surgery may soon be able to watch a patient’s CT scan update itself in real time on the operating table, without extra radiation, without bulky tracking hardware, and without pausing the operation. A research team led by Nicole M. Gunderson, Graham J. Harris, Jeremy S. Ruthberg, Pengcheng Chen, Di Mao, Randall A. Bly, Waleed M. Abuzeid, and Eric J. Seibel has introduced a technique called Virtual Intraoperative CT, or viCT, which transforms ordinary monocular endoscopic video into sequentially updated CT-format images of the evolving surgical cavity. In a cadaveric feasibility study published in the International Journal of Computer Assisted Radiology and Surgery, the method achieved submillimeter agreement with true intraoperative CT scans, suggesting that the operating endoscope itself could one day double as a radiation-free surrogate for one of the most accurate—but least practical—intraoperative imaging tools in medicine.
The clinical problem the technology addresses is a stubborn one. Chronic rhinosinusitis, a persistent inflammatory condition of the nasal cavity and paranasal sinuses, affects roughly one in eight adults in the United States and severely degrades quality of life, with health utility scores comparable to other major chronic illnesses. When medical therapy fails, endoscopic sinus surgery is the primary treatment, and about 28.3 percent of patients ultimately undergo the procedure, with roughly three-quarters achieving sustained symptom relief. Yet revision surgery is required in 15 to 20 percent of cases, and among patients who have had multiple operations, 29 percent undergo a third procedure within 36 months. Revision surgery carries lower success rates and independently predicts poorer outcomes, making the quality of the first operation critical.
Incomplete dissection is a leading culprit behind these failures. Postoperative CT evaluations have identified residual bony partitions and persistent inflammation in 64 to 96 percent of cases. Strikingly, when intraoperative CT has been performed at the conclusion of sinus surgery in prospective studies, it altered the surgical plan in 30 percent of patients by prompting additional tissue removal. That statistic reveals both the problem and a paradox: intraoperative CT is the most reliable way to verify the extent of resection, yet it is rarely used because it interrupts the workflow for 30 to 40 minutes, prolongs anesthesia, increases cost, and exposes the patient to additional ionizing radiation. Existing image-guided surgery platforms from manufacturers such as Medtronic, Stryker, Brainlab, GE, and Acclarent overlay instruments on a static preoperative CT scan, leaving surgeons to mentally reconcile where the tissue used to be with where it has already been removed.
The viCT framework, by contrast, never requires the patient to be wheeled into a scanner. Instead, at defined points during the operation, the surgeon performs a simple three-to-five-second sweep of the resection cavity with a standard 4-millimeter rigid 30-degree endoscope recording 1080p video at 30 frames per second. After automatically discarding frames degraded by blur, instrument occlusion, irrigation fluid, or severe specular reflection, the remaining images feed into a depth-supervised Neural Radiance Field, or NeRF, framework. NeRF technology, which has transformed computer vision since its introduction, learns a continuous three-dimensional scene representation by training a neural network to reproduce the appearance of the scene from any viewpoint. In this implementation, spatial positions are encoded using multi-resolution hash encoding and viewing directions using spherical harmonics, while camera intrinsics and poses are estimated with the structure-from-motion pipeline COLMAP. The network is trained by minimizing the photometric reconstruction error between rendered and actual images, and crucially, the depth supervision combined with virtual stereo synthesis allows the reconstruction to be metrically scaled rather than merely proportionally correct—a distinction that matters enormously when the surgical field spans only a few centimeters.
Once a metric 3D reconstruction of the cavity exists, the harder challenge is fusing it with the preoperative CT. The reconstructed geometry is exported as an STL mesh, registered to the preoperative CT coordinate frame through rigid, landmark-based registration performed in the open-source platform 3D Slicer, and then voxelized into the native CT grid. Because STL meshes can be hollow shells, the team developed a ray-based voxelization scheme: rays are cast from a fiducial point defining the reconstructed camera origin through every mesh vertex, and sampled points along these rays are mapped to voxel indices, forming a sparse occupancy scaffold that is then refined with binary dilation, morphological closing, and hole filling into a watertight volumetric mask. The actual CT update follows an elegant Boolean logic. Voxels belonging to the preoperative CT mask that overlap the reconstructed free-space mask are classified as resected tissue; voxels outside the free space are classified as preserved anatomy. Resected voxels have their Hounsfield units reassigned to an air-equivalent value of −1000 HU, while unchanged anatomy retains its original intensities and DICOM metadata. The result is a genuine CT-format volume, viewable in standard axial, coronal, and sagittal planes in any clinical viewer, that reflects the anatomy as it exists at that moment in surgery. Because the procedure can be repeated as new video sweeps are acquired, the CT effectively evolves step by step alongside the operation.
To validate the approach, the researchers imaged four cadaveric heads at four timepoints—preoperatively, after maxillary antrostomy, after partial anterior ethmoidectomy, and postoperatively after posterior ethmoidectomy and sphenoidotomy—with true CT scans at each stage serving as ground truth. An otolaryngology surgeon captured endoscopic video at anatomically relevant angles throughout each surgical interval, and viCT volumes were generated and compared against the interval CT segmentations. The quantitative results were remarkably strong. The Dice Similarity Coefficient, a standard measure of volumetric overlap, averaged 0.88 ± 0.05, and the Jaccard Index averaged 0.79 ± 0.07, indicating that the virtual updates closely matched the true resection cavities in three dimensions.
Surface-level agreement was even more impressive. The 95th-percentile Hausdorff distance, a strict measure of worst-case surface deviation, averaged just 0.69 ± 0.28 millimeters, while the symmetric Chamfer distance averaged 0.09 ± 0.05 millimeters. Mean surface distance came in at 0.11 ± 0.05 millimeters and root-mean-square surface distance at 0.32 ± 0.10 millimeters—all comfortably below the submillimeter threshold that clinicians would demand of an intraoperative imaging surrogate. Because global overlap metrics can sometimes conceal localized reconstruction failures, the team also performed an anatomy-specific analysis of 104 labeled 3D reconstructions grouped by region, including the anterior ethmoid, maxillary sinus, middle turbinate, nasal floor, posterior ethmoid, and sphenoid sinus. Change-specific Dice coefficients ranged from 0.74 to 0.80 and average symmetric surface distances from 0.22 to 0.27 millimeters across the major anatomical groups, with no evidence of a pronounced region-specific failure pattern. Qualitative side-by-side comparisons of viCT slices against ground-truth interval CT showed close visual correspondence across all three imaging planes.
What makes viCT genuinely distinctive among surgical navigation technologies is that it updates the image itself rather than merely annotating a stale one. Conventional tracked-instrument systems display the position of a probe on orthogonal preoperative CT planes, while newer platforms such as Acclarent TruDi’s Fast Anatomic Mapping accumulate the spatial path traversed by a tracked instrument so surgeons can infer which regions have been explored. Both approaches, however, leave the underlying CT voxels untouched and provide only indirect surrogates for updated anatomy—forcing the surgeon to interpret instrument location in the context of tissue that may no longer exist. Prior vision-based approaches have used monocular depth estimation and TSDF fusion to update surface models, but viCT goes further by generating a full volumetric, Hounsfield-unit-preserving CT update from metric-scale NeRF geometry, all without external tracking hardware, fiducial markers, or additional intraoperative imaging.
The implications extend beyond sinus surgery. The core pipeline—video sweep, NeRF reconstruction, CT-grid registration, ray-based occupancy comparison—could in principle be adapted to any endoscopic procedure where surgeons navigate relative to preoperative cross-sectional imaging, from skull base surgery to neurosurgical cavity monitoring. For the approximately one in eight Americans living with chronic rhinosinusitis, the prospect of reducing the 15 to 20 percent revision rate by giving surgeons a continuously refreshed, radiation-free view of what tissue remains is significant. The method essentially approximates the anatomical fidelity of intraoperative CT while preserving operative efficiency, sidestepping the workflow interruptions of 30 to 40 minutes that have kept true intraoperative CT out of routine practice.
Important caveats remain before the technology reaches the operating room. All processing in this feasibility study was performed retrospectively rather than as a continuous real-time stream, and the workflow included manual landmark-based registration between reconstruction and CT updating. The end-to-end latency from video acquisition to completed viCT was not measured, though the team reports that computation ran on a workstation with an Intel Core i7-6850K CPU, 64 gigabytes of RAM, and two NVIDIA GeForce RTX 5080 GPUs, with each video sweep contributing roughly 90 to 150 raw frames before quality filtering. The minimum number of retained views and the angular coverage required to maintain reconstruction accuracy were not independently varied, leaving open questions about how robust the method will be under the messier conditions of live surgery, where bleeding, anatomical distortion, and time pressure prevail.
The researchers are explicit about their roadmap. Future work will focus on fully automating the registration step, expanding validation into live clinical cases, and optimizing runtime for genuine intraoperative deployment. If those challenges are met, the humble endoscope—already the workhorse of minimally invasive sinus surgery—may evolve into something far more powerful: a handheld device that not only removes diseased tissue but continuously rewrites the patient’s CT scan as it does so, ensuring that no bony partition is left behind and fewer patients ever need a second operation.
Subject of Research: Virtual Intraoperative CT (viCT) for sequential, CT-format anatomic updating during endoscopic sinus surgery using NeRF-based 3D reconstruction from monocular endoscopic video
Subject of Research: Medicine
Article Title: Virtual intraoperative CT (viCT): sequential anatomic updates for modeling tissue resection throughout endoscopic sinus surgery
Article References: Gunderson, N. M., Harris, G. J., Ruthberg, J. S., Chen, P., Mao, D., Bly, R. A., Abuzeid, W. M., & Seibel, E. J. (2026). Virtual intraoperative CT (viCT): sequential anatomic updates for modeling tissue resection throughout endoscopic sinus surgery. International Journal of Computer Assisted Radiology and Surgery. https://doi.org/10.1007/s11548-026-03783-0
Image Credits: AI Generated
DOI: 10.1007/s11548-026-03783-0
Keywords: virtual intraoperative CT, viCT, endoscopic sinus surgery, chronic rhinosinusitis, Neural Radiance Field, NeRF, intraoperative imaging, image-guided surgery, 3D reconstruction, surgical navigation
Cite Scienmag News
APA MLA Chicago
Ophelia Keating. (September 4, 2026). Virtual CT updates tissue resection modeling during endoscopic sinus surgery. Scienmag. https://scienmag.com/virtual-ct-updates-tissue-resection-modeling-during-endoscopic-sinus-surgery/
Ophelia Keating. “Virtual CT updates tissue resection modeling during endoscopic sinus surgery.” Scienmag, 4 September 2026, https://scienmag.com/virtual-ct-updates-tissue-resection-modeling-during-endoscopic-sinus-surgery/. Accessed 4 September 2026.
Ophelia Keating. “Virtual CT updates tissue resection modeling during endoscopic sinus surgery.” Scienmag. September 4, 2026. https://scienmag.com/virtual-ct-updates-tissue-resection-modeling-during-endoscopic-sinus-surgery/
Copy citation Download RIS
Tags: cadaveric feasibility studies in ENTcadaveric feasibility studies in sinus surgerychronic rhinosinusitis surgical planningcomputer-assisted radiology and surgerycomputer-assisted sinus surgeryendoscope-based tissue resection modelingendoscopic sinus surgeryendoscopic sinus surgery imagingendoscopic visualization advancementsintraoperative imaging accuracyintraoperative imaging technologyminimally invasive sinus surgery technologymonocular endoscopic video processingradiation-free surgical imagingradiation-free surgical navigationreal-time CT scan updatesreal-time CT update during sinus surgerysubmillimeter accuracy in intraoperative imagingsubmillimeter imaging precisionsurgical cavity modelingsurgical cavity monitoringtissue resection modelingVirtual intraoperative CTvirtual intraoperative CT imaging
