Clinical validation of an AI-assisted system for real-time kidney stone detection during flexible ureteroscopic surgery

Flexible ureteroscopy (FURS) is a minimally invasive, standard treatment for kidney stones. This study presents the development and clinical validation of an artificial intelligence system during FURS (AiFURS) for real-time detection, classification, and measurement of stones. Using 6170 annotated ureteroscopy video frames representing 11,870 labeled stones, the AiFURS was trained to identify stone type, size, and number. Ex vivo validation across 191 groups predicted stone counts precisely (r > 0.9) in 300 samples. Size predictions for stones >2 mm (n = 100, r = 0.81) correlated with gold-standard caliper measurements. In vivo and external validation of 100 and 80 cases, respectively, demonstrated diagnostic accuracy (92.2-95.3% and 86.8-92.2%, respectively) for patient-level stone type prediction, outperforming expert surgeons. Logistic regression further identified the proportion of residual fragments (RFs) > 2 mm, measured during the final minutes of FURS, as an independent predictor of reoperation. AiFURS offers a novel solution to enhance surgical accuracy, reduce complications, and improve outcomes in endourology.