N15 - Repeatability of Dynamic Tracer Kinetic Model Analysis for Retinal Vascular Permeability: Impact of Independent Preprocessing

Introduction: : Retinal vascular permeability determination using fluorescein video angiography (FVA) and dynamic tracer kinetic modeling (DTKM) is a promising approach for early detection of diabetic retinal vascular changes. However, the repeatability of DTKM results needs to be evaluated to ensure its reliability as a biomarker tool when preprocessing is performed by different users. Preprocessing steps such as motion correction, eye blinking removal, and manual placement of regions of interest (ROIs) can introduce user variability. The main objective of this study was to determine the repeatability of DTKM outputs between two investigators analyzing the same FVA datasets independently. The goal was to determine the degree of variability in the vascular permeability parameter that may be introduced from independent preprocessing, specifically frame removal and ROI selection.

Materials and

Methods: : FVA datasets were collected from three healthy subjects and three diabetic patients without diabetic retinopathy (DMnoDR). FVA (0.5 ml of 10% fluorescein dye) was performed using a Heidelberg SPECTRALIS® system with a 55° field of view. The videos were exported as uncompressed AVI files. Two users independently performed preprocessing, including removal of poor-quality frames affected by blinking, motion artifacts, or oversaturation, followed by motion correction using the DynamicICG motion correction software. ROIs were manually placed on arteries and neighboring tissue regions. Both users placed ROIs in the same anatomical locations but did so independently. Intensity-time curves were background-subtracted and modeled using a two-step DTKM approach. Initial estimates of blood flow and arrival time were obtained first, followed by five-parameter curve fitting to derive extraction fraction (E), blood flow (F), arrival time (ta), transit time (tt), and adiabatic rate constant (kadb). All statistical analysis was performed using GraphPad Prism. The Shapiro-Wilk test was used to assess normality. Paired t-tests were performed to compare user outputs, with significance defined as p < 0.05. Pearson correlations were also calculated to evaluate the degree of inter- user agreement across all parameters.

Results, Conclusions, and Discussions:: The difference between the parameter values obtained by two users were examined and the paired t-tests showed no statistically significant differences between users for any DTKM parameter. The average of those differences, along with their standard deviations, were as follows: blood flow differed by 17.83 ± 90.22 ml·100g⁻¹·min⁻¹, extraction fraction -0.00167 ± 0.02714, arrival time +0.0400 ± 0.6449 s, transit time -0.1833 ± 0.9847 s, adiabatic rate constant +0.00175 ± 0.00492 min⁻¹. These small average differences and wide standard deviations suggest that while values fluctuated slightly between users, there was no consistent bias in one direction. Pearson correlation analysis demonstrated strong inter-user agreement for extraction fraction r=0.90, p=0.0077, indicating good reproducibility. Transit time also showed good correlation r =0.77, p=0.0358, while arrival time and kadb exhibited weaker correlations across users, suggesting possible sensitivity to both frame selection variability and/or ROI positioning.
These findings support the robustness of DTKM-extracted permeability metrics even when minor manual variations in preprocessing exist. Extraction fraction and transit time showed strong inter-user consistency, reinforcing their reliability as biomarkers for future clinical studies. Standardizing or semi-automating ROI placement and frame inclusion criteria may further enhance repeatability for parameters like arrival time and the adiabatic rate constant. Overall, DTKM analysis of FVA data is a reliable method for vascular permeability quantification, strengthening its potential as a translational biomarker for early retinal vascular dysfunction.

Acknowledgements and/or References (Optional):: Supported by NIH grant RO1 EY032222