Neural networks for optic nerve detection in digital optic fundus images

This paper presents a method used for detection of optic nerve in fundus digital images; for this purpose, initially there is a preprocessing and segmentation of digital images of fundus taken from databases Messidor and Stare in order to stand out the veins and blood vessels of ocular region. The processed image is used in an artificial neural network which has three layers; an input layer with 10000 neurons that belong to the vectorized and processed image, a hidden layer with 100 neurons and an output layer with 42 neurons, which define the pixel deviation between the center of the ROI and the original image. For the training of the Artificial Neural Network, is used an arrangement of fifteen selected images from databases and scaled to a suitable size, so it can fit inside the ROI; this training was performed using backpropagation. At the end of the training, there were two types of experiments for result's validation; the first one was the detection of optic discs in known databases with an error of 0%, the second one was the detection of optic discs in unknown databases, with an error of 5% during the tracking of optic nerve, thus it concludes that the presented method is efficient for the detection of optic disc in fundus digital images.