Automated volumetric iRECIST-inspired CT assessment of treatment response to dual immune checkpoint inhibition in metastatic renal cell carcinoma: A feasibility study.

Authors

Abstract

Evaluate the feasibility and interobserver variability of automated volumetric immune Response Evaluation Criteria in Solid Tumours (iRECIST)-inspired scoring of treatment response in metastatic renal cell carcinoma (mRCC). Pretreatment and follow-up CECT or PET/CT scans at 12-week intervals were retrospectively collected from 75 mRCC patients receiving ipilimumab/nivolumab at a single centre (2019-2024). Whole-body lesion segmentation and longitudinal tracking were used to derive volumetric iRECIST-inspired scores. Three radiologists independently assigned iRECIST scores, then re-evaluated 15 test patients using the automated output. Performance was assessed using Dice score, false positives, tracking accuracy, F1, and new lesion detection accuracy (NLDA). Interobserver agreement was compared before and after AI assistance using multi-level κ statistics (p < 0.05). Seventy-five patients (65.6 ± 9.8 years, 74.7 % male) were included. Dice scores post-revision were 0.74 (IQR: 0.71-0.78) and 0.60 (IQR: 0.56-0.64) at baseline and follow-up, with 2.4 (IQR: 2.1-2.7) and 0.6 (IQR: 0.5-0.8) false positives per scan, respectively. Tracking accuracy was 0.98 (IQR: 0.98-1.00), F1 score 0.90 (IQR: 0.80-1.00), and NLDA 0.85 (IQR: 0.80-1.00). Agreement with the consensus yielded κ = 0.60 (95 % CI: 0.50-0.68) and κ = 0.44 (95 % CI: 0.31-0.54) at first and second follow-up. AI assistance significantly improved interobserver agreement from κ = 0.34 (95 % CI 0.11-0.56) to κ = 0.67 (95 % CI 0.40-0.89; p = 0.04). Automated volumetric iRECIST-inspired scoring in mRCC patients is feasible and improves interobserver agreement, supporting an interpretable iRECIST-based response assessment.

Tags

Topics