May 2005
Volume 46, Issue 13
ARVO Annual Meeting Abstract  |   May 2005
Automated Analysis of the Retinal Vascular Tree – Parameterised Data From Healthy Eyes
Author Affiliations & Notes
  • P.C. Knox
    University of Liverpool, Liverpool, United Kingdom
  • F.N. Hatfield
    Medical Imaging,
    University of Liverpool, Liverpool, United Kingdom
  • D.J. Farnell
    University of Liverpool, Liverpool, United Kingdom
  • Footnotes
    Commercial Relationships  P.C. Knox, None; F.N. Hatfield, None; D.J. Farnell, None.
  • Footnotes
    Support  HERG, University of Liverpool
Investigative Ophthalmology & Visual Science May 2005, Vol.46, 4288. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      P.C. Knox, F.N. Hatfield, D.J. Farnell; Automated Analysis of the Retinal Vascular Tree – Parameterised Data From Healthy Eyes . Invest. Ophthalmol. Vis. Sci. 2005;46(13):4288.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Abstract: : Purpose: Digital retinal images are now routinely available, allowing analysis of the retinal vascular tree. We have developed simple automated procedures for analysing retinal vascular patterns to allow later comparison between different patient groups and healthy controls. Methods: A set of 20, 50º red–free filtered images, acquired using a Zeiss FF450 camera from 14 healthy subjects, were output in the form of 8–bit uncompressed TIFF files, 1280 pixels wide (w) by 1024 pixels high (h), with grey scale values from 0 to 255. Analysis algorithms were developed using the IDL software environment. The separation (R) between the optic disk and foveal centre was used to define two circular regions of interest (ROI) of radii R and ½ R centred on the fovea. Each image was rotated to the same alignment so that the centre of the optic disk and fovea were oriented along the horizontal axis. The images were initially enhanced using an unsharp masking routine. Two solid circles (radius 100 pixels) were then superimposed, one at point (1/5 w, 1/2 h), the second at (4/5 w, 1/2 h). From each point, a region–growing algorithm was used to search for all connected pixels with values 50 to 255. Two vessel maps were created from each set of search results and combined using a logical OR operation, and all identified pixels set to 1. Gaps due to image artefacts were filled using a morphological closing operator with a 5x5–structuring element with all values set to 1. A skeletonised image was created and used to identify vessel branching points. Vessel maps were extracted and analysed within the two circular ROIs. Results: Analysis of each image was completed in under 5 seconds when run on a 1.7Ghz Pentium 4 PC. The mean ratio of the total vessel areas within the ROIs (expressed as ½ R/R * 100) was 20%±3% (mean ± sd). The mean percentage of vessel area per unit area was 11%±2% and 14%±2% for small and large ROIs respectively. There were 76±36 and 328±139 branch points detected in small and large ROIs respectively, yielding a mean ratio of 24%±6%. Conclusions: Our results suggest that parameterised information concerning the retinal vascular tree can be quickly extracted from digital images, allowing the rapid analysis of large numbers of images. There was relatively little variability in some ratios (eg mean ratio of vessel area) across our 20 images. It remains to be established how these parameters are affected pathological changes in the retina (eg in the early stages of macular disease) and whether they might be related to disease risk.

Keywords: retina • imaging/image analysis: non-clinical 

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.