Clinically validated annotated dataset of cystoscopy videos with bladder cancer

High-quality annotated medical video datasets are essential for the development and validation of artificial intelligence (AI) systems in endoscopy. However, creating such datasets for complex and variable procedures like cystoscopic diagnosis of bladder cancer (BCa) is challenging. We present a dataset of 30 clinical cases, comprising cystoscopy video recordings, corresponding histological diagnosis, and frame-by-frame annotations of pathological areas. Pathological areas included both high-grade and low-grade tumors, carcinoma in situ (CIS), as well as inflammatory changes. Annotation accuracy has been clinically validated by oncourologists. This dataset represents clinically verified comprehensive resource for training neural network models. The resource enables population studies in oncourology, facilitates the development and testing of computer vision algorithms, and contributes to the creation of clinical decision support systems and training simulators, both virtual and physical.