Abstract
Abstract: :
Purpose: To develop and evaluate a new fast computerized method for measuring caliber and length of retinal vessels on fundus images. Methods: Computer analysis of digitized color fundus photographs using active iterative contour modelling to provide a fast estimate of retinal trunk vessel caliber and length. The fit is initiated after visual identification and manual marking of the start and end points of the vascular segment of interest. Arterio-venular crossings or bifurcations define start and endpoints. The point is set exactly where the medians of the two vessels are crossing or fusing. Reproducibility was tested on a subset of photographs by repeat analysis of identical vascular segments in different photographs from the same eye. The analysis included intra- and inter-photograph (frame-to-frame) reproducibility, as well as intra- and inter-observer reproducibility and day-to-day reproducibility. Results: Retinal arteriolar and venular segments of 5±0.5 mm length were contour modelled within 10 s with an intra-photograph coefficients of variability (CV) of 0,003% to 0,044% for length and 0,15% to 0,34% for caliber on an Intel Pentium III 800 Mhz processor personal computer. These fluctuations reflect the iterative nature of the fitting procedure. The intra-observer CV was 0,25% for the measurement of length and 0,69% for caliber. Inter-user CV was 0,68% for length and 1,1% for caliber. The frame-to-frame CV was 0,65% for length and 1,38% for caliber measurements. Conclusion: The speed and precision of our active contour modelling algorithm was faster and had a better reproducibility than previously described methods for estimation of quantitative trunk vessel characteristics as recorded on fundus photographs. The results demonstrate that changes as small as 1.5% can reliably and consistently be detected by this method, indicating the the method has a potential for application in the study of diabetic retinopathy, retinal venous occlusions, and arterial hypertension where changes in length of up to 10% and in caliber up to 25% have been found to accompany the progression to sight-threatening disease.
Keywords: 430 imaging/image analysis: clinical • 364 computational modeling • 388 diabetic retinopathy