Today the standard for the diagnosis and monitoring of bladder tumors is bladder endoscopy. The performance of this exam is not perfect. With this work, based on artificial intelligence, the investigators wish to combine endoscopy with a complementary diagnostic tool in order to improve patient care. The main objective will be to reduce diagnostic errors / wanderings in patients treated or followed for bladder tumors, by imposing a new standard of diagnostic bladder mapping (high PPV and VPN, high precision)(primary purpose diagnostic). The secondary objective will be to homogenize and systematize the descriptive part of the lesions, and to use AI to better characterize tumor aggressiveness. The final objective being to validate a new precision tool (diagnostic companion) essential for developing and standardizing the therapeutic management of bladder tumors (correcting inter-observer heterogeneity).
In this project, video frame will be first extracted from our dataset of cystoscopy videos hosted in in the Next Cloud Recherche. Selected medical image will be segmented and analyzed using our pre-trained CNN model with a feature detection algorithm to obtain features.
Data will be analyzed on both patient and lesion levels. The study will assess the Bladder-PAD accuracy on the detection of bladder tumors, and its ability to predict tumor risk of recurrence and progression.
