Hopfield network applied to blood vessel detection in angiograms

A neural network classifier for detecting vascular structures in angiograms is developed. The classifier consists of a Hopfield network applied to a square window in which the centre pixel is classified from binary information within the window. Tests are performed using a binary test image corrupted by inverting a percentage of the image pixels. The resulting noisy images simulate the output of a detector using a simple threshold derived from local image statistics. The factors affecting the size of window and the choice of stored patterns are discussed. The results are compared with those obtained from a multi-layer perceptron using a similar approach. The Hopfield network is found to be effective at rejecting the high levels of noise that would result from low-contrast source imagery. Another important feature is that the processed image retains an accurate representation of blood vessel diameter. A neural network classifier for detecting vascular structures in angiograms is developed. The classifier consists of a Hopfield network applied to a square window in which the centre pixel is classified from binary information within the window. Tests are performed using a binary test image corrupted by inverting a percentage of the image pixels. The resulting noisy images simulate the output of a detector using a simple threshold derived from local image statistics. The factors affecting the size of window and the choice of stored patterns are discussed. The results are compared with those obtained from a multi-layer perceptron using a similar approach. The Hopfield network is found to be effective at rejecting the high levels of noise that would result from low-contrast source imagery. Another important feature is that the processed image retains an accurate representation of blood vessel diameter.