Cyclic signals classification by codegrams characterizing the dynamics of cycles shape changing

Abstract

Research in engineering, biology, economics and other areas is often associatedwith the analysis of observed processes that are repetitive in time. One of thepromising approaches to solving the problem of analyzing and interpreting suchsignals is based on converting the original cyclic signal into a sequence of symbols of some alphabet, for which methods of mathematical linguistics can beused. The linguistic approach to the processing of cyclic signals involves theconstruction of a codogram that characterizes the dynamics of changes in theshape of successive cycles. To construct codograms, it is proposed to use twovalued and three-valued indicator variables. A procedure is proposed for determining the optimal value of the threshold of insensitivity to changes in signalparameters, which provides a minimum of intra-class distances and a maximumof inter-class distances. The construction of standards for recognizable classes isbased on the Levenshtein matrix of paired distances between the codograms ofthe training sample of each of the classes and the definition of a codogram that isat the minimum total distance from the rest of the codograms of the class underconsideration. Computational procedures are proposed that allow determiningthe dominant patterns of classes in the form of three-character patterns of codograms. Decision rules have been developed to classify processed cyclic signalsaccording to both codegram standards and dominant patterns. The effectivenessof the proposed approach has been demonstrated using examples of processingelectrocardiograms. It has been established that the constructed decision ruleprovides sensitivity and specificity in the classification of electrocardiograms ofpatients with coronary heart disease and healthy volunteers, even in the absenceof generally accepted diagnostic signs of myocardial ischemia on the ECG. It isadvisable to continue research aimed at studying the possibility of further improving the efficiency of the proposed approach, in particular, based on the processing of codograms using sequence alignment algorithms that are activelyused in bioinformatics.