Automatic detection of LV contours in nuclear medicine using geometrical information and a neural net

Presents a method that makes use of a neural net and geometrical information for the automatic detection of left ventricle (LV) contours in nuclear medicine images. Although the method has been developed for LV contour detection, it can be extended to other classes of structures and images. Learning is carried out by feeding the system with a series of images and their corresponding LV contours drawn by by an operator. The system extracts both pixel-value and geometrical information that is used for training the neural network. Once trained, the network is able to automatically detect LV contours. Apart from presenting errors that are compatible with several other automatic detection techniques the present method has the clear advantage of being able to store geometrical and pixel intensity information that is learned from examples.