Abstract
Retinal vascular tortuosity is a biomarker associated with various retinal and systemic diseases. However, its clinical application is hindered by the lack of an objective, standardized definition. With the growing number of automated pipelines for the evaluation of tortuosity from retinal images, the impact of image processing, choice of tortuosity index, and retinal camera imaging parameters have been overlooked. This study assesses the robustness of different tortuosity indices to artificial changes in camera imaging parameters and the choice of centerline extraction algorithm to provide a path toward standardization of tortuosity quantification. Five tortuosity indices were implemented and tested on tortuosity-ordered vessel segments using Spearman's Rank Correlation Coefficients for comparison. Comparable results were obtained for all tested indices. Image magnification was found to have minor or no impact on the most tested indices. A key finding from this study is that the same blood vessel imaged in different regions of the field of view of the same imaging system may differ in measured tortuosity. Different centerline extraction algorithms were also found to have a critical impact on tortuosity quantification. Taken as a whole, this study shows the importance of standardizing the analytical procedures in the evaluation of blood vessel tortuosity.