Optical imaging of retinal blood flow: Studies in automatic vessel extraction, alignment, and driven changes in vessel oximetry

Access to files

•   FULL-TEXT.PDF (pdf)

Abstract

Recent advances in retinal imaging have made it possible to take measurements of retinal oxygen saturation noninvasively in humans. This allows studying the supply of oxygen in healthy and diseased retinae, thereby advancing our understanding of both the normal functioning of the retina and of retinal pathologies. However, retinal oximetry is still a research tool only and requires further improvement before being used in a clinical setting.Here, a single-wavelength flickering light was used to increase retinal blood flow in healthy subjects. This increase is revealed by both vasodilation and an increase in retinal oxygen saturation. A flickering light stimulus provides the means to assess the sensitivity of any retinal oximetry system, as such systems should be able to pick up this increase in retinal blood flow. In addition, the flickering light allows for com- parison to be made within rather than between subjects and can be used to examine the activation of the eye. This reduces the influence of potential confounding factors between subjects including differences in fundus pigmentation and illumination. The most commonly used method to measure retinal oxygenation is the optical density ra- tio (ODR) approach. The standard approach is to compute the average ODR for each vessel segment by combining the hundreds of individual ODR readings and then to use the mean of these segment averages as a measure of oxygen saturation. Alternatively, it has been suggested that the peak location of Gaussian functions fitted to histograms of individual ODR readings can be used as an measure of retinal oxygenation.In response to a 10Hz flickering light, the venular diameter increased by 3.44% (SEM: ±0.53%) (n=16, p<0.05) and the arteriolar diameter by 1.87% (±0.72 %) (p<0.05). The optical density ratio, measured with the Gaussian fit, decreased in the venules from 0.713 (±0.015) to 0.694 (±0.015) (p<0.05). No changes in arteriolar optical density ratios were measured. The post-flicker measurement was computed as the average of up to four post-flicker datasets obtained at 10s, 20s, 30s and 40s after onset of flickering. These results suggest that the flickering light increased retinal blood flow. The mean absolute percentage error was lower in venules for the Gaussian fit method than for the gold standard method for datasets taken at 30s and 40s after onset of flickering. Thus, the Gaussian fit method was more robust. All measurements were taken with a custom-made retinal oximeter.The pixel intensity of the blood vessel and the intensity on either side of the vessel had to be extracted to compute the individual optical density ratios. This required the automatic extraction of the retinal vasculature. Two such algorithms were developed and applied to two databases of retinal fundus images: the DRIVE and the novel DR HAGIS database. One algorithm was purely based on the pixel intensities, while the other made use of oriented Gabor filters. These two algorithms segmented the images to a similar accuracy (DRIVE: 94.56% and 94.54%, DR HAGIS: 95.83% and 95.71% for the intensity and Gabor filter based algorithm, respectively) and performed as well as a human expert (DRIVE: 94.73%). These algorithms were of sufficient quality to extract individual segments for the oximetry study and to align fundus images.

Keyword(s)

Automatic Vessel Extraction; Blood Flow; Oximetry; Retina; Retinal Oximetry; Retinal Vessel Diameter; Retinal Vessel Tortuosity

Bibliographic metadata

Institutional metadata

Record metadata