PURPOSE: Bladder tumor delineation and localization during treatment are challenging problems in radiotherapy for bladder cancer. The purpose of this study is to investigate improvement of tumor delineation by the fusion of cystoscopy images with the planning CT-scan using lipiodol markers injected around the visible tumor during cystoscopy. METHODS: A registration method was developed for the fusion of cystoscopy images with a planning CT-scan and was tested on a phantom and retrospectively on the imaging data of four bladder cancer patients. For the patients, small deposits of lipiodol were injected at the visible margin of the tumor or previous transurethral resection site during cystoscopy. These deposits were clearly visible on the planning CT-scan and served as markers for both tumor delineation and image guidance of the radiotherapy treatment. Here, the markers were used for the registration of cystoscopy images with the planning CT-scan. The registration procedure works as follows: First, coarse registrations were made to orient the cystoscopy image correctly, using the center of gravity of the markers, the center of the CT bladder, and one of N markers as fiducial points in a point matching procedure. Starting from these N orientations, full registrations are performed taking lens deformation into account. Since a cystoscopy image is 2D, each pixel corresponds to a line-of-sight. The distances between the CT markers and the lines-of-sight of the cystoscopy markers were minimized. The final cost function (the root mean square distance between corresponding CT markers and lines-of-sight) was used to quantify the quality of the registration. The registration with the lowest final cost was considered to represent the correct orientation. The CT-based tumor delineation was finally backprojected onto the cystoscopy image. RESULTS: The fusion of cystoscopy images with a planning CT-scan succeeded for the phantom and three out of four patients. The fiducial registration error (FRE) for the phantom image registration based on five markers was 1.1 mm, while the target registration error was 1.2-1.7 mm. The FREs for the patient images were 0.1-3.6 mm. The registration procedure failed for one patient, since it was not possible to indicate unambiguously the corresponding lipiodol marker locations in the cystoscopy image and the planning CT-scan. The difference between the CT and cystoscopy defined tumor outlines clearly exceeded the registration accuracy. CONCLUSIONS: Registration of cystoscopy images and planning CT-scan is feasible and allows for improvement of tumor delineation. However, the lipiodol injection protocol needs to be improved to facilitate identification of markers on both cystoscopy images and planning CT-scans.