Optimization of CT-video registration for image-guided bronchoscopy

Global registration has been shown to be a potential reality for a bronchoscopy guidance system. Global registration involves establishing the bronchoscope position by comparing a given real bronchoscopic (RB) video view to target virtual bronchoscopic (VB) views derived from a patient's three-dimensional (3D) multi-detector computed tomography (MDCT) chest scan. Registration performance depends significantly on the quality of the computer-generated VB views and the error metric used to compare the VB and RB views. In particular, the quality of the extracted endoluminal surfaces and the lighting model used during rendering has especial importance in determining VB view quality. Registration performance is also affected by the positioning of the bronchoscope during acquisition of the RB frame and the error metric used. We present a study considering the impact of these factors on global registration performance. Results show that using a direct-lighting-based model gives slightly better results than a global illumination model. However, the VB views generated by the global illumination model more closely resemble the RB views when using the weighted normalized sum-of-square error (WNSSE) metric. Results also show that the best global registration results are obtained by using a computergenerated bronchoscope-positioning target with a WNSSE metric and a direct-lighting model. We also identify the best airway surface-extraction method for global registration.