Fig. 6: AiFURS pipeline for kidney stone detection and analysis.

Ureteroscopy video frames are analyzed by the YOLOv11-N Backbone, Neck, and Head to predict stone locations and generate bounding boxes. BoT-SORT then tracks each stone across frames by assigning it a unique ID. Pixel dimensions are converted to millimeters using a fixed ratio, after which RFs are binned into categories of <1 mm, 1–2 mm, and >2 mm for counting and statistical analysis. Grad-CAM heatmaps overlay model attention, offering visual interpretability. Abbreviations: AiFURS artificial intelligence flexible ureteroscopy system, Grad-CAM gradient-weighted class activation mapping, RFs residual fragments, 3 × 3 Conv convolutional layers, C3k2 cross-stage partial with 2 convolution blocks, SPPF spatial pyramid pooling fast, C2PSA cross-stage partial with pyramid squeeze attention.