Semi–automated detection of the foveal avascular zone by a region growing algorithm in diabetic patients

Abstract

Abstract: : Purpose: To test a region growing algorithm to quantify the foveal avascular zone area in fluorescein angiograms of diabetic patients. Methods: Early phase high quality fluorescein angiograms were obtained with a high resolution digital camera (1280x1024 pixels, 50° lens opening), from 44 eyes of 41 patients presenting diabetic retinopathy. Manual measurement of the foveal avascular zone was first performed independently by two experts with the ENVI 3.4 image analysis software after image pre–treatment by contrast stretching. Foveal avascular zone surface was calculated in square pixels. The region growing option of the ENVI 3.4 software was then used on the same images, and on images treated with a 5x5 median filter. Results: Average foveal avascular zone surface area was 3032.59 square pixels after manual measurement and 1951.32 square pixels after semi–automated measurement without median filter and 2226.27 with median filter. Agreement between foveal avascular zone surface areas measured by the two experts was high (R = 0.98). Average manual measurement compared well to semi–automated detection on untreated images (R = 0.89) and slighlty better with median filter treated images (R = 0.91). Conclusions: Semi–automated detection of the foveal avascular zone surface area after image pre–treatment by contrast stretching and median filtering underestimates the manually detected surface in a reliable manner.

Keywords: diabetic retinopathy • imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)