PhD Thesis Defense: Ryan Duke

ZOOM LINK
Meeting ID: 931 9017 9828
Passcode: 306610

"Novel Methods for Image Guided Neurosurgery"

Abstract

Three-dimensional image guidance has become commonplace within neurosurgical applications where sensitive anatomy is occluded from the view of surgeons. All image guidance systems require accurate three-dimensional imaging of the dynamic surgical scene. The current state of the art requires intraoperative CT or MRI. Mobile systems are being developed such as O-Arm (Medtronic, Dublin, Ireland). These systems are incredibly expensive, require additional safety requirements and disrupt surgical work flow. This thesis covers my work in novel imaging systems, and novel animal models for image guided neurosurgery. First, my development of a Hand-Held Stereovision (HHS) system and calibration methods are defined. Two iterations of HHS, a semi-automated calibration method, and a method for video data acquisition of intraoperative data are covered. Calibration is completed with an automated script using a tracked checkerboard to a reprojection error of ~0.30mm, and video data can be collected with accuracy comparable to a step and shoot method (<2mm). Second, a live animal double blind pedicle screw implant study using our HHS and deformable registration was completed. HHS navigated pedicle screws only had one case of grade 3 (>4mm) breach of 5.5mm compared to preoperative CT imaging which had 3 instance of grade 3 breach 18mm, 9.5mm, and 10.9mm. A retrospective re-calibration of HHS data was performed and the data were re-calibrated to ~0.8mm with an acceptable TRE of ~2.0mm. Third, an analysis of our deformable registration model was performed. Live animal registration error was reduced from 3.34 ± 1.72mm to 2.59 ± 1.64mm. An interpretable method for manual masking of the bony surface of the spine was developed. Fourth, a simulated pedicle screw placement study was completed to test our system under ideal conditions. Fifth, a novel, in vivo , porcine, genetic model of glioblastoma  was developed. All 6 animals grew tumors which fluoresce and are pathologically similar to human, high-grade glioblastoma tumors. These tumors are much more realistic that xenograft models due to the infiltrative nature of the spontaneously growing tumors. This work further serves as a valuable source of data, and analysis of our level-wise registration model, additionally the genetic model of glioblastoma is currently in use for valuable fluorescent guided and image guided studies.

Thesis Committee

• Keith Paulsen (Co-Chair)
• Xiaoyao Fan (Co-Chair)
• Sohail Mirza
• Linton Evans
• Matthieu Chabanas (Univ. Grenoble Alpes)

Contact

For more information, contact Thayer Registrar at thayer.registrar@dartmouth.edu .