Automated vessel segmentation of 35mm colour non-mydriatic images in a community health screening project

Proliferative diabetic retinopathy is a complication of diabetes that can eventually lead to blindness. Early identification of this complication reduces the risk of blindness so that timely treatment can be initiated. In rural and remote regions widespread population screening is practically impossible due to the lack of ophthalmologists and cost associated with rural visits by specialists. Several methods for vessel segmentation have been discussed in the literature but none have used non-mydriatic colour images obtained from community screening initiatives. Rural screening clinics currently use either 35mm or Polaroid photography. In addition the quality of the images is often much lower. Scanning images at 300 dpi provides very low resolution images and therefore combined with the low quality requires a robust algorithm to identify vessels with high accuracy. Visual inspection by an ophthalmologist judged 46 images (88%) to represent an acceptable level of segmentation. Despite the low resolution and quality of images, the Gabor wavelet provided vessel segmentation results that were usable in rural community screening projects and in some cases identified vessels obscured by haemorrhages better than the expert observer.