Discrimination between healthy and diseased hearts by spectra decomposition of their left ventricular three-dimensional geometry.

Abstract

A quantitative procedure for 3-D shape analysis of human LVs (left ventricles,) studied in vivo by 3-D computed tomography is introduced. The procedure utilizes the geometrical cardiogram signal for characterizing the 3-D shape of the studied LV, and uses spectral decomposition to obtain the corresponding feature vectors needed for automatic classification. It was found that the different pathological states of the LV were associated with characteristic changes in the geometrical spectrum domain. Representing each heart by a feature vector in the spectral domain and applying unsupervised fuzzy clustering of the obtained 27 feature vectors, an overall success of 85% in classification was obtained. These results indicate that an operator-independent shape-based diagnosis is potentially feasible for the four different pathological categories studied.