About

A brief overview of our project

The human-machine integration for vessel segmentation project has as objective finding efficient and precise methods for the segmentation of blood vessels in retinal fundus images. Through this project we will publish and mantain the current software we have for this and we'd also like to develop a new interface, ideally capable of evolving and learning through analysis of human interaction coupled with machine learning techniques.

Our group (see [People]) has been working on the problem of vessel segmentation since 2000, as well as analysis of the retinal vascular tree structures. A major goal in automatic analysis of retinal vascular structures is the detection of diabetic retinopathy. Proliferative diabetic retinopathy is a complication of diabetes that can eventually lead to blindness. Early identification of this complication reduces the risk of blindness and provides timely treatment. Developing an automated method to identify blood vessels is the first step towards automating recognition of changes associated with diabetic retinopathy. To provide the opportunity for initial assessment to be carried out by community health workers, computer based analysis has been introduced. For more on this motivation, please see our list of [References].

Currently, our vessel segmentaion approach is basically to classify retinal image pixels as vessel or nonvessel, based on each pixel’s feature vector. Feature vectors are composed of the pixel’s intensity and continuous two-dimensional Gabor (or Morlet) wavelet transform responses taken at multiple scales. The Gabor wavelet is capable of tuning to specific frequencies, thus allowing noise filtering and vessel enhancement in a single step. We estimate probability distributions for classification using training sets of labeled pixels obtained from manual segmentations. For evaluation, we are currently using the STARE and DRIVE publicly available databases of retinal images with corresponding manual segmentations. For more on our methods, check out our publications and also our open source softwares.