Supervised segmentation of vasculature in retinal images using neural networks

This paper proposes a neural network based supervised segmentation algorithm for retinal vessel delineation. The histogram of each training image patch and its optimal threshold acquired through iteratively comparing the binaryzation result to the manual segmentation are applied to a BP neural network to establish the correspondence between the intensity distribution and optimal segmentation parameter. Finally, each test image can be segmented by using a number of local thresholds that are predicted by the trained the neural network according the histograms of image patches. The propose algorithm has been evaluated on the DRIVE database that contains forty retinal images with manually segmented vessel trees. Our results show that the proposed algorithm can effective segment the vasculature in retinal images.