london euretina
retina race

Automated analysis of fluorescence gradients in fluorescein angiograms

Session Details

Session Title: Imaging IV

Session Date/Time: Sunday 20/09/2015 | 11:00-13:00

Paper Time: 12:04

Venue: Thalie.

First Author: : S.Haldar UNITED KINGDOM

Co Author(s): :    N. Davies              

Abstract Details

PURPOSE:This study investigates the use of image analysis software to calculate rate of change of fluorescence in the fundus image and automatically delineate abnormal rate of change to identify hyper and hypofluorescent areas. This may allow clinicians a more objective method for examining the results of an angiographic study.

Setting:

Post acquisition analysis of digital fluorescein angiograms

Methods:

Software was written using MATLAB R2014a (Mathworks Inc). Image registration to align angiographic frames was performed using cross-correlation for translational alignment and a Euclidean warping algorithm to correct for rotation and scaling. Median values for each aligned frame were calculated and plotted as a function of time acquired. The frames were separated into dye entry (first) phase and recirculation (second) phases based on the highest median value obtained. Fluorescence gradients (in fluorescence units/sec) were calculated at each pixel for both phases. Finally a gradient classifier identified areas of hyper and hypo fluorescence. Positive gradients in the second phase were used to identify hyperfluorescent pixels and the lowest quartile gradients in both first and second phases identified hypofluorescent pixels.

Results:

70 angiograms were analysed from patients. 3 failed the registration process leaving angiograms from 28 patients with diabetes, 28 with AMD, 4 with vein occlusion and 7 with other retinal conditions. First phase gradients ranged from 1.7 units/sec (SD 3.4) to 12.8 units/sec (SD 8.9) and second phase from -0.3 units/sec (SD 0.8) to +0.3 units/sec (SD 1.0). Hyperfluorescence ranged from 0% to 23% in the registered angiogram area. Areas of hypofluorescence ranged from 0% to 12%.

Conclusions:

The software was able to identify areas of abnormality in each aligned angiogram and the success rate of the alignment algorithm was 96%. The areas of hyperfluorescence were identified as positive gradients in the second angiographic phase and hypofluorescence based on low gradients in both phases. Data relating to gradients at individual pixels can be reviewed, leading to a more standardised analysis of rates of fluorescence. Areas of quantified abnormality can be subsequently colour-coded and overlaid onto an angiographic frame and/or the aligned colour fundus photograph for ease of interpretation.

Back to previous
EURETINA, Temple House, Temple Road, Blackrock, Co Dublin. | Phone: 00353 1 2100092 | Fax: 00353 1 2091112 | Email: euretina@euretina.org

Privacy policyHotel Terms and Conditions Cancellation policy