Registration of real and CT-derived virtual bronchoscopic images to assist transbronchial biopsy.

This paper describes research work motivated by an innovative medical application: computer-assisted transbronchial biopsy. This project involves the registration, with no external localization device, of a preoperative three-dimensional (3-D) computed tomography (CT) scan of the thoracic cavity (showing a tumor that requires a needle biopsy), and an intraoperative endoscopic two-dimensional (2-D) image sequence, in order to provide assistance in transbronchial puncture of the tumor. Because of the specific difficulties resulting from the data being processed, a multilevel strategy was introduced. For each analysis level, the relevant information to process and the corresponding algorithms were defined. This multilevel strategy, thus, provides the best possible accuracy. Original image processing methods were elaborated, dealing with segmentation, registration and 3-D reconstruction of the bronchoscopic images. In particular, these methods involve adapted mathematical morphology tools, a "daemon-based" registration algorithm, and a model-based shape-from-shading algorithm. This pilot study presents the application of these algorithms to recorded bronchoscopic video sequences for five patients. The preliminary results presented here demonstrate that it is possible to precisely localize the endoscopic camera within the CT data coordinate system. The computer can thus synthesize in near real-time the CT-derived virtual view that corresponds to the actual endoscopic view.