Abstract: A method of analysing biomedical signals, for example electrocardiograms, by using a Hidden Markov Model for subsections of the signal. In the case of an electrocardiogram two Hidden Markov Models are used to detect respectively the start and end of the QT interval. The relationship between the QT interval and heart rate can be computed and a contemporaneous value for the slope of this relationship can be obtained by calculating the QT/RR relationship for all of the beats in a sliding time window based on the current beat. Portions of electrocardiograms taken on different days can efficiently and accurately be compared by selecting time windows of the ECGs at the same time of day, and looking for similar beats in those time windows.

Abstract: A way of quantifying the shape of an ECG waveform is disclosed by detecting the JT segment using two Hidden Markov Models and calculating the analytic signal of the JT segment. Parameters calculated from the analytic signal are used as shape descriptors for the JT segment. The shape descriptors may be displayed in a dimensionality-reduced mapping. Templates representing characteristic shapes can be produced by finding cluster centers in the shape descriptor space, and the novelty of new waveforms can be quantified by comparing the position in shape descriptor space of new shape descriptors to a predefined normal training set or to previously encountered waveforms. Novel shape descriptors can be used to retrieve the corresponding waveforms, and templates of such novel shapes can be created by averaging such waveforms, using dynamic time warping to allow for variations in heart rate.

Abstract: Improvement in the reliability of segmentation of a signal, such as an ECG signal, is achieved through the use of duration constraints. The signal is analysed using a hidden Markov model. The duration constraints specify minimum allowed durations for specific states of the model. The duration constraints can be incorporated either in the model itself or in a Viterbi algorithm used to compute the most probable state sequence given a conventional model. The derivation of a confidence measure from the model can be used to assess the quality and robustness of the segmentation and to identify any signals for which the segmentation is unreliable, for example due to the presence of noise or abnormality in the signal.