The study of cardiac motion through CINE MRI is an important non-invasive diagnostic tool for cardiac abnormalities. In this paper, a method for automatic detection of abnormal motion patterns is proposed to be used as a computerized diagnostic tool for pathologic cardiac function. A multi-scale modeling method, based on a Generating-Shrinking neural network and a 4-D surface parametric model were used to extract the deformation of the myocardium from multi slice-multi phase MRI examinations. A feature extraction procedure then calculated myocardial thickening and radial deformation of the left ventricle and produced a set of motion parameters from the surface model. Input patterns consisting of the above features were fed into a feedforward neural network, which was trained to capture the normal cardiac function and to distinguish certain pathologic motion patterns.
展开▼
