How consistent is the interpretation of renal scarring in pediatric patients using technetium-99m dimercaptosuccinic acid scintigraphy

Abstract

Background: Technetium-99m dimercaptosuccinic acid (DMSA) scintigraphy plays a critical role in pediatric imaging for detecting renal cortical scarring, which is essential for diagnosing and managing kidney damage in children. However, variability in observer interpretation poses challenges, potentially impacting clinical decision-making and outcomes. Objective: This study aims to assess intra- and inter-observer agreement in interpreting DMSA scans for detecting renal cortical scarring in pediatric patients, focusing on the presence, location, and percentage of kidney involvement. Materials and methods: This prospective study analyzed 220 pediatric patients with suspected renal scarring. Four experienced radiologists independently reviewed DMSA scans on two separate occasions, 3-4 weeks apart, using standardized assessment criteria. Intra-observer agreement was measured using Cohen's kappa, while inter-observer agreement was assessed using pairwise Cohen's kappa for categorical evaluations and Kendall's tau-b for the percentage of kidney involvement. Results: Strong intra-observer agreement was observed across all four radiologists, with Cohen's kappa values for renal scarring stages ranging from 0.704 to 0.955. Observer-4 consistently showed the highest agreement across all metrics. Inter-observer agreement varied substantially depending on observer pairs. Pairs excluding Observer-2 demonstrated moderate to substantial agreement (kappa up to 0.8268 and Kendall's tau-b up to 0.7192), while pairs involving Observer-2 showed poor to slight agreement. Variability was particularly notable in assessing scarring severity and defect localization. Conclusion: While intra-observer consistency in interpreting DMSA scans is high, inter-observer variability remains a concern, especially in evaluating the severity and location of renal scarring. These findings underscore the need for standardized protocols and targeted training to enhance diagnostic accuracy. Moreover, the development of validated datasets could support the advancement of machine learning models for automated, precise detection of renal scarring, ultimately improving diagnostic reliability and patient outcomes.