Automation of four nephrometry scores using 3D models in renal cell carcinoma.

OBJECTIVES

Nephrometry scores quantify renal tumor complexity to standardize reporting, however, manual calculation is time-consuming and prone to interobserver variability. Furthermore, 2-dimensional preoperative imaging limits anatomical interpretation. Therefore, we developed a fully automated method using 3-dimensional (3D) models to calculate 4 established scoring systems: RENAL, PADUA, C-index and CSA, and benchmarked them against manual calculations for validation.

METHODS

Preoperative CT scans and manually segmented 3D models of 80 patients undergoing robot-assisted partial nephrectomy at Ghent University Hospital were analyzed. 3 observers (student, resident, expert) manually scored each CT.Automated 3D scores were generated using our Python-based algorithm. Discrepant resident-expert scores were re-evaluated to reach consensus for comparison with automatic scores. Interobserver variability was assessed using intraclass correlation coefficients (ICC). Scoring durations were compared using paired t-tests. Pearson correlation coefficients were calculated between automated/manual scores and surgical complexity metrics: estimated blood loss (EBL), operative time (OT), and warm ischemic time (WIT).

RESULTS

Manual interobserver agreement was moderate to good (ICC(2,1) = 0.72-0.90). Re-evaluation was needed in 67 cases (83.75%). Automated scores demonstrated moderate to good agreement with manual consensus (ICC(3,1) = 0.68-0.89) and substantially reduced scoring time (manual = 5.9 minutes vs. automated = 0.2 minutes; p < 0.001). Expert re-evaluation confirmed automated accuracy in 92.9% of reviewed cases. Correlations with EBL, OT, and WIT did not differ significantly between manual and automated methods (p > 0.1).

CONCLUSIONS

Our fully automated 3D model-based calculations of RENAL, PADUA, C-index, and CSA offer improved reproducibility, accuracy and efficiency while maintaining clinical validity.