Abstract
PURPOSE: Intrarenal pressure (IRP) management during endoscopic surgery for urolithiasis is critical to minimize postoperative pain and infectious complications. However, pressure sensors at the ureteroscope tip often register various artifacts when contacting the pelvicalyceal or ureteral wall, leading to significant deviations from true IRP values. This study aimed to develop and validate a machine learning model to detect and remove artifacts from IRP data accurately. MATERIALS AND METHODS: We analyzed 27 retrograde intrarenal surgeries performed using the Boston Scientific(®) LithoVue™ Elite system across academic institutions in Japan and Hong Kong. 32 waveform features were identified and three tree-based machine learning models—Random Forest, XGBoost, and LightGBM—were trained for automated artifact detection. Endpoints included agreement with ground-truth labels and comparison of time efficiency between artificial intelligence (AI)-based and manual annotations. RESULTS: The best-performing model achieved an overall agreement of 93.6% and an area under the receiver operating characteristic curve of 0.95. Each case included 94 artifacts, contributing 258 s per surgery. Artifacts accounted for 31% of the time > 30 mmHg and 72% of the time > 100 mmHg. Without correction, peak IRP was overestimated by 184 mmHg (median, 257 vs. 73 mmHg). False negatives > 60 mmHg had a median 0.0 s per case. AI-based IRP annotation saved 99.95% of the time compared to manual review (1.8 s vs. 56.6 min per case). CONCLUSIONS: The model successfully achieved high-precision artifact removal from IRP data. This system may serve as a standardized process for accurate IRP analysis during endoscopic surgery. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s00345-025-06057-7.