Development of a Free-Viewpoint Bladder Observation System for Cystoscopy

Patients with bladder cancer typically require both a transurethral examination and surgery, and accurate recording of information observed during a transurethral examination is necessary for surgery. However, observational recordings are made manually and depend on the physician's ability, and the digital information gathered during the examination is not fully utilized in surgery. To address these issues, this study aimed to propose a free-viewpoint bladder observation system for cystoscopy based on multi-view image-based three-dimensional (3D) reconstruction. We used neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS) to develop the proposed system as 3D reconstruction methods. Although these methods are state-of-the-art in the field of 3D reconstruction, adapting them to a small dome-shaped bladder is challenging; there needs to be an assumption that the inputs of these methods are the sets of images of the same region captured from multiple viewpoints. To overcome this problem, by classifying the images into bladder dome and neck areas and reconstructing each location, we obtained the entire 3D bladder model. This approach leads to the generation of free-viewpoint images of the whole bladder. We present the results of a comparative study concerning the possibility of preserving examination records. Our findings indicated that NeRF outperformed 3DGS in terms of texture reproducibility, and that reconstruction through dividing area of the bladder was successful in generating free-viewpoint inner images.