Background Skull base reconstruction following endonasal transsphenoidal tumor resection is
a focus of active technical development that has seen steady evolution of multiple
approaches to surgical defect closure over the past decade. A primary consideration
during sellar reconstruction is the adequacy of seal around the defect, which may
be implicated in rates of postoperative CSF leak. At our institution, the gasket-seal
method is the preferred closure technique following expanded endonasal approach (EEA)
transsphenoidal tumor resection; however, this technique requires intraoperative preparation
of a semirigid buttress (MedPor or autograft), manually prepared by subjective visual
assessment of the surgical defect on endoscopic images. This work details our experience
establishing an image post-processing pipeline to provide accurate 3D surface reconstructions
of the visible skull base using operative endoscopic imaging, to facilitate design
and fabrication of a custom rigid sellar buttress.
Methods A sellar floor defect was prepared on a cadaveric subject in standard fashion. A
calibration object was placed adjacent to the defect endonasally. High-definition
video was recorded using a standard endonasal telescope (Karl Storz GmbH & Co. KG).
Video clips were separated into individual frames and processed into a 3D surface
mesh using structure-from-motion (SFM) software (VisualSFM). The resulting skull base
mesh was scaled so the calibration object−matched known dimensions. A planar buttress
was designed and modeled based on the mesh. CT images of the prepared cadaveric head
were also acquired, and a standard 3D bone model was generated from CT data for comparison.
Visible skull base surface landmarks were used to coregister the endoscopic 3D surface
mesh to the CT bone model, and subjectively examined for global anisotropy.
Results A 35-second endoscopic video clip yielded a total of 1,063 frames. Still images were
captured at a reduced 12 frames per second to facilitate post-processing, and examined
for exposure consistency and motion. A total of 388 images were processed into a 3D
surface mesh, requiring ∼90 minutes of unmonitored machine processing. Endoscopic
images and screenshots taken at each post-processing step are provided. A 3D printed
model of the skull base and rigid buttress for sellar reconstruction is available
for presentation.
Discussion Subjective assessment of the scaled, coregistered endoscopic 3D surface mesh revealed
dimensional consistency with rigid landmarks identified on the CT bone model, within
expected differences implicit with inclusion of overlying soft tissue. Given substantial
overlap between sequential images as processed, required post-processing time is likely
amenable to additional time optimizations without undue compromise to mesh accuracy.
Conclusion Operative endoscopic image data acquired during endonasal transsphenoidal resection
can be used to directly 3D model visible skull base anatomy using structure-from-motion
software. The resulting surface mesh is of sufficient dimensional accuracy to facilitate
design and fabrication of a custom rigid buttress for gasket-seal closure of a surgical
defect. Subsequent validation studies directly examining the adequacy of the resulting
gasket-seal, post-processing time optimizations and best practices, and quality assurance
metrics are currently underway in advance of a clinical feasibility study.
