C73-25 The Impact of Ventilation Defect Analysis Pipelines on Test-retest Variability of Hyperpolarized Xenon-129 (129Xe) Magnetic Resonance Imaging (MRI) in Young People With Cystic Fibrosis (CF)

Rationale

The ventilation defect percentage (VDP) is the most commonly reported quantitative outcome derived from hyperpolarized Xenon-129 (129Xe) magnetic resonance imaging (MRI). While high test-retest repeatability of VDP has been reported, multiple computational pipelines have been described to define VDP. This study sought to determine the impact of four common VDP quantification pipelines on the same-day and 28-day variability of VDP in children and young adults with cystic fibrosis (CF) and healthy controls (HC).

Methods

Twenty-two participants with CF (median age 15.3 years) and 11 HCs (median age 14.8 years) were prospectively recruited. Participants underwent two repeated MRI scans on the same day (∼20-30 minutes apart), followed by a single scan at a follow-up visit ∼28 days later (median 29.0 days). Scans were analyzed using the newly developed XIPLine toolkit across four pipelines: thresholding, linear binning (LB), k-means, and adaptive k-means (ak-means). Repeatability was assessed via intraclass correlation coefficients (ICC) and Bland-Altman (BA) analysis. To address observed heteroscedasticity in the BA analysis, where variability increased with VDP magnitude, a log-log transformation was applied to the data, followed by back-transformation to derive robust BA limits of agreement (LoA).

Results

VDP values were significantly higher in the CF group compared to HCs across all methods (p < 0.001). A consistent trend in VDP magnitude was observed across the pipelines (thresholding > LB > ak-means > k-means). In the CF cohort, same-day ICC values ranged from 0.94 (k-means) to 0.98 (thresholding) with tight confidence intervals. The 28-day ICCs ranged from 0.79 (k-means) to 0.96 (thresholding), with notably wider confidence intervals. Same-day bias in the standard BA analysis was near zero across all methods. Standard BA analysis revealed significant heteroscedasticity, where the difference between repeat scans increased with VDP magnitude. The log-log transformation successfully accounted for this behavior, allowing for the derivation of robust LoA. Comparison of the back-transformed LoA revealed similar bias trends and LoA across all pipelines.

Conclusion

The test-retest repeatability of VDP in children and young people with CF is excellent across same-day and 28-day timepoints, further solidifying its utility as a stable biomarker for longitudinal monitoring of CF. While all methods performed well, thresholding and LB techniques demonstrated slightly superior reliability based on ICCs. Furthermore, BA analysis indicated relatively similar LoA across all pipelines.

This abstract is funded by: None