Automatic contour extraction in images using a 2-D hidden Markov model

A new entirely automatic method is proposed to detect the contour of an object in low contrast medical images. The contour extraction is based on a tight cooperation between a multiresolution neural network and a hidden Markov model-enhanced dynamic procedure. Such a modelization allows to introduce relevant high order a priori information at different stages of the extraction process. An application to the automatic detection of the left ventricle in digital X-ray images is proposed.