May 2007
Volume 48, Issue 13
ARVO Annual Meeting Abstract  |   May 2007
Automated Detection of Hyperfluorescent Leakage in Fluorescein Angiograms
Author Affiliations & Notes
  • C. R. Buchanan
    Electrical, Electronic & Computer Engineering, Heriot-Watt University, Edinburgh, United Kingdom
  • E. Trucco
    Electrical, Electronic & Computer Engineering, Heriot-Watt University, Edinburgh, United Kingdom
  • G. England
    Optos Plc, Dunfermline, United Kingdom
  • D. Cairns
    Optos Plc, Dunfermline, United Kingdom
  • C. Mazo
    Optos Plc, Dunfermline, United Kingdom
  • Footnotes
    Commercial Relationships C.R. Buchanan, Optos plc, F; E. Trucco, Optos plc, F; G. England, Optos plc, E; D. Cairns, Optos plc, E; C. Mazo, Optos plc, E.
  • Footnotes
    Support DTI KTP Grant KTP000684
Investigative Ophthalmology & Visual Science May 2007, Vol.48, 2757. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      C. R. Buchanan, E. Trucco, G. England, D. Cairns, C. Mazo; Automated Detection of Hyperfluorescent Leakage in Fluorescein Angiograms. Invest. Ophthalmol. Vis. Sci. 2007;48(13):2757.

      Download citation file:

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

  • Supplements

Purpose:: To develop image analysis algorithms to detect incompetent vessels causing blood leakages in fluorescein angiographic image sets.

Methods:: Digital angiograms were acquired from the Optos Panoramic 200 SLO in fluorescein angiography mode allowing a 200° field of view of the retina. An image processing algorithm was developed and implemented. Algorithm summary: 1) perform image alignment on multiple late frames; 2) create a composite image from the aligned images; 3) segment candidate leakage regions; 4) classify fluorescein leakage by analysis of time-intensity profiles; 5) track leakage boundaries over time in order to pinpoint source. The key step is the analysis of the temporal change in pixel value across the stack of images. The rationale is that where leakage is present there will be a characteristic evolution of the intensities over time, in the simplest case an increased fluorescein in later frames in comparison to areas of normal perfusion. The specific nature of the leakage-region profiles is determined using an AdaBoost learning algorithm. The resulting classifier is trained using a varied set of ground-truth image sequences with and without leakages.

Results:: Anonymous angiograms of 15 patients were used to train and test the classifier using a leave-N-out procedure so that results are presented on unseen image sequences. Example detections and sensitivity/specificity metrics are presented. Initial results indicate reliable detection of all significant leakage in seen angiograms. Misclassifications of smaller leakages are largely due to inadequate image alignment and insufficient ground-truth data. The results demonstrate some robustness to false detections such as ciliar occlusion and imaging artefacts.

Conclusions:: The software assists the user by pinpointing the source of leakage in time and can overlay angiograms with areas of hyperfluorescent leakage. The method also permits visualisation of different rates of leakage in specific areas. The analysis of intensity changes over time allows for clear discrimination between areas of increasing fluorescence and non-leakage areas where fluorescence is steady or decreasing. This technique is of relevance to conditions such as diabetic retinopathy and BRVO in providing objective detections. Additionally, it is apparent that wide-field imaging is crucial for identification of peripheral pathology.

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

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.