Image‐enhanced surgical navigation for endoscopic sinus surgery: evaluating calibration, registration and tracking

Abstract

Endoscopic sinus surgery (ESS) is generally applied to treat sinusitis when medication is not effective in eliminating the symptoms. Images captured by the endoscope are viewed on a monitor placed near the surgeon. Due to the separation of the handling of the endoscope from the viewing of the image, ESS requires surgeons to have well-trained hand-eye coordination. Unlike the use of the stereo surgical microscope in ENT, the endoscope does not provide the stereo cue for depth perception, hence a surgeon can only perceive depth through motion and shading, which may affect the accuracy of tool placement. Whilst the skill and experience of the surgeon are important factors to the success of ESS, the assistance of image-enhanced surgical navigation (IESN) can further reassure the surgeon's judgement and enhance surgical performance. We developed and validated an IESN system (ARView) for a rigid zero-degree endoscope, typically used for ESS. We present the interface, and calibration and registration (pre-operative and intra-operative) methods of the system. We then quantitatively assess the performance of each of the steps needed to generate the overlay of a real endoscope image with its 'virtual' counterpart, obtained from computed tomography (CT) image data of a real skull. These steps include calibration, registration, motion tracking and final overlay. Calibration results using a planar calibration object displayed optimized object space errors of 0.025 +/- 0.013 mm, whilst a non-planar calibration object displayed errors of 0.12 +/- 0.08 mm. Target registration errors (TREs) near the region of interest (ROI), using our pre-operative registration method with the calibration object located near the mouth of the patient (skull), were 2.3 +/- 0.4 mm. The proposed photo-consistency method for intra-operative registration has not yet yielded satisfactory results for ESS-based IESN. (RMS) values for tracking accuracy were found to be around 1.2 mm in a typical workspace of 400 x 400 mm. Object space overlay errors in a small measurement volume of 10 x 10 x 10 mm were found to be around 0.4 +/- 0.02 mm. We conclude that, in agreement with individual experiments, the current overall overlay accuracy is of the order of 2-3 mm in the x-y plane, which is in line with current conventional SN systems. The method which is most in need of improvement is registration, hence we wish to investigate the application of the proposed photo-consistency method further.