Segmentation and classification of the blood flow signals

Abstract

The coronary ischemic disease is characterized by change of blood flow turbulence and, as a result, by appearance of high-frequency sounds that are caused by stenosis of arteries. The most informative part of this signal, which is used for diagnostics of the coronary ischemic disease, is the diastole. In order to extract the diastole from the entire signal, the moments, which correspond to the beginning and the end of a flap of the mitral valve damper, have to be determined. These moments can be detected as change-points of probabilistic characteristics of the signal, especially, as change-points of the spectral density of the signal. A training sample is formed from the extracted stationary fragments of the diastole for the healthy men and men with the coronary ischemic disease. The training sample is used for construction of a decision rule. In the paper the method for segmentation of the blood flow signal based on a spectral test for change-points detection is considered. The proposed decision rule is based on discriminant analysis of parameters of autoregressive models, which describe stationary fragments of the diastole.