Image Registration to Map Endoscopic Video to Computed Tomography for Head and Neck Radiotherapy Patients

The purpose of this work was to explore the feasibility of registering endoscopic video to radiotherapy treatment plans for patients with head and neck cancer without physical tracking of the endoscope during the examination. Endoscopy-CT registration would provide a clinical tool that could be used to enhance the treatment planning process and would allow for new methods to study the incidence of radiation-related toxicity. Endoscopic video frames were registered to CT by optimizing virtual endoscope placement to maximize the similarity between the frame and the virtual image. Virtual endoscopic images were rendered using a polygonal mesh created by segmenting the airways of the head and neck with a density threshold. The optical properties of the virtual endoscope were matched to a calibrated model of the real endoscope. A novel registration algorithm was developed that takes advantage of physical constraints on the endoscope to effectively search the airways of the head and neck for the desired virtual endoscope coordinates. This algorithm was tested on rigid phantoms with embedded point markers and protruding bolus material. In these tests, the median registration accuracy was 3.0 mm for point measurements and 3.5 mm for surface measurements. The algorithm was also tested on four endoscopic examinations of three patients, in which it achieved a median registration accuracy of 9.9 mm. The uncertainties caused by the non-rigid anatomy of the head and neck and differences in patient positioning between endoscopic examinations and CT scans were examined by taking repeated measurements after placing the virtual endoscope in surface meshes created from different CT scans. Non-rigid anatomy introduced errors on the order of 1-3 mm. Patient positioning had a larger impact, introducing errors on the order of 3.5-4.5 mm. Endoscopy-CT registration in the head and neck is possible, but large registration errors were found in patients. The uncertainty analyses suggest a lower limit of 3-5 mm. Further development is required to achieve an accuracy suitable for clinical use