Cellular edge detection using a trained neural network explorer
Victor Barrios, J. Torres, G. Montilla, Lovely Rose T. Hernandez, Nahum Rangel, Aldo Reigosa · 2002
A multi-layer perceptron neural network, with a backpropagation training algorithm, was employed for edge detection and tracing in cancer cellular tissue images, obtained with an optical microscope. This network predicts the cellular edges' location, based on information regarding a small known section of it, at every point. We use a metaphorical "worm", based in the neural network, who crawls along the edge and "feeds" from the points it finds along the same. The method was used with a set of 256 gray-level test images, in order to detect edges of known geometric shapes, under controlled noise conditions. It was later applied to mammary tissue images. Additionally, this method provides some learning capabilities, which yield it's application to several image types.