Analysis of vectorcardiograms using fourier descriptors

Abstract

This thesis presents the results of analysis of three important aspects of vectorcardiograms: rhythm monitoring, morphological studies, and data compression. For this purpose, the method of Fourier Descriptors (FDs)-a technique used for the analysis of closed contours-is examined. Arrhythmic beats such as supraventricular premature complexes (SVPCs) and premature ventricular complexes (PVCs) are detected using the FDs of the QRS vector loops. QRS events are detected using either a two level bidirectional search on spatial or planar velocity or a one level unidirectional search on planar velocity. A representation of vector loops that is both insensitive to QRS delineation errors and computationally less expensive is achieved with the unidirectional search procedure. An unsupervised adaptive sequential clustering algorithm is adopted for detecting arrhythmic QRS complexes. Analysis of nearly 60 normal and abnormal VCG records reveals that the method not only detects PVCs and SVPCs efficiently but also identifies multifocal ectopics. These results demonstrate higher classification efficiency than existing methods. ________________________________________ For the purpose of data compression, the ECG cycle is divided into two segments: the active QRS complex and the relatively calm SQ interval. FDs of these intervals are considered separately. Differential compression ratios of three for the QRS segment and ten for the SQ interval are employed without compromising the fidelity of the reconstructed signal. The technique is evaluated by computing the percent RMS difference between the original and reconstructed data. It is shown that even noisy records can be compressed effectively using a built in filtering mechanism. Results from compression of more than 50 records of different clinical categories indicate that the FD method performs better than existing techniques, yielding an overall compression ratio greater than 7, with strong potential for real time applications. ________________________________________ The FDs of QRS vectorcardiograms are analysed to demonstrate the feasibility of the method for morphological analysis. Significant diagnostic information contained in segments of the vector loop is extracted in the form of descriptors of the initial, middle, and terminal segments, in addition to the complete QRS loop. A method for computing the representative beat-to eliminate the effects of noise and artefacts-is described, along with the extraction of FD features. It is shown that qualitative shape features of high diagnostic value used in clinical vectorcardiography can be derived from the FDs, while normalized descriptors serve as potential features for diagnostic classification. The discrete attributes of these qualitative features hold promise for mass screening of morphological abnormalities. Analysis of several typical abnormal cases-including myocardial infarction, ventricular hypertrophy, hemiblocks, and bundle branch blocks-yields encouraging results. ________________________________________ Additionally, an outline of clinical vectorcardiography, its theoretical basis, and methods of recording and display is presented, along with typical VCG examples. Equipment developed for data acquisition and VCG display is described. Pre processing for detection and delineation of the QRS complex in the ECG cycle using spatial and planar velocities is discussed. The method of Fourier Descriptors is outlined, and typical applications in pattern recognition are mentioned.