Real time extraction of ECG fiducial points using shape based detection

The electrocardiograph (ECG) is a common clinical and biomedical research tool used for both diagnostic and prognostic purposes. In recent years computer aided analysis of the ECG has enabled cardiographic patterns to be found which were hitherto not apparent. Many of these analyses rely upon the segmentation of the ECG into separate time delimited waveforms. The instants delimiting these segments are called the "fiducial points". Rapid automatic identification of the fiducial points is a task which the biomedical engineer needs to perform as a prerequisite to further analysis. To the researcher, post-processing of pre-recorded results is an acceptable mode of analysis. However clinical staff require an immediate real-time diagnosis. This thesis is concerned with the detection of fiducial points using methods which lend themselves to real-time analysis, suitable for use in the development of systems intended for environments where immediate results are called for. By way of introduction, an examination of contemporary methods of fiducial point detection is presented, followed by a discussion of the conventional methods of assessing their performance. It is found that these assessment methods, whilst widely used, are not only mathematically imprecise but are formed in such a way as to be useful to the biomedical engineer, but not to the clinician. An alternative assessment measure which overcomes these problems is proposed, and examples presented to demonstrate how the proposed new measure can be used. Secondly, a novel method of ECG peak detection is presented. Since real-time detection is of interest, the method is developed with emphasis on optimising its performance without the use of filtering or other pre-processing stages. The method relies upon examination of the shape of the signal's peaks rather than its spectral analysis. This approach brings benefits in terms of noise immunity, particularly where the spectral response of the feature is similar to that of the noise signal. Initially, the method is applied to the task of detecting R peaks, the most prominent feature of the ECG. Results are presented which demonstrate how the method is not only effective at discriminating between these peaks, and other maxima in the ECG, but also how the speed of execution suggests that the method lends itself to real-time applications in a clinical setting. Later in the thesis, the method is extended and a hybrid approach developed which uses the advantages of the new method in conjunction with conventional linear signal processing techniques to detect another two of the most important features, namely the P and T waves. Finally, since the basic method encapsulates the geometry of peaks in the signal, the author presents a discussion of the opportunities that the method holds for further uses. In particular it is proposed that the detection of onset and offset detected waveforms can be extracted without significant extra time penalty. Overall, the thesis presents an alternative basis for ECG analysis and an alternative methodology of assessing the performance of such systems. It is hoped that the reader finds these methods more intuitive than their existing counterparts, and will inspire further research.