Quantification of Global Tortuosity in Retinal Blood Vessels

Abstract

Tortuosity is a parameter that indicates the tendency of a blood vessel segment to contain multiple twists and turns. Chronic hemodynamic changes in the body due to diabetes and hypertension will manifest as increased retinal vascular tortuosity, rendering tortuosity as a suitable indicator for diabetic and hypertensive retinopathy. Retinal tortuosity may be evaluated locally on a single segment or globally in the complete vascular network. Global tortuosity quantification consists of automated segmentation and partition of retinal vessel network, local tortuosity measurement, and global tortuosity index derivation from weighted combination of local tortuosity values. This paper proposes several weighting schemes and evaluates their performance when combined with different local tortuosity indexes. We perform rank correlation analysis to find the global tortuosity quantification that is most consistent with the ophthalmologists. Our results show that local tortuosity indexes that are robust to variations in scale and number of sampling points provide the best performance. Furthermore, weighting scheme based on chord length yields better results than the one based on arc length. The combination of Tortuosity Density (TD) local index and Tortuosity Density Global (TDG) weighting scheme provides the highest consistency with ophthalmologists, with the average rank correlation coefficient of 0.98 (p-value < 0.03).