Coherent pattern identification in non-stationary periodic data and

Abstract

A method and apparatus for automatically recognizing meaningful patterns in non-stationary periodic data is provided. The invention may be used in making measurements of physiological phenomena, such as arterial blood pressure, wherein data of interest is undergoing small changes in timing and magnitude in succeeding cycles, and wherein noise artifacts are detected along with the data of interest. Data points of interest are identified by grouping all received data points into family groups of multi-dimensional data points wherein each data point in the family group has dimensional values similar to at least one other data point in the family. The present invention may preferably be employed for the determination of a patient's systolic and diastolic blood pressure levels. For this application, the data points include a time dimension value corresponding to the delay time between an ECG signal and the detection of a potential blood pressure sound by a microphone placed in an inflatable cuff on the arm of a patient, and a magnitude dimension value corresponding to the cuff pressure at the time of the potential blood pressure sound detection. Data points are collected for a series of cuff pressures ranging from above the systolic blood pressure level of the patient to below the diastolic blood pressure level of the patient. The highest and lowest pressures of data points in the family containing the largest number of data points are used to derive the systolic and diastolic blood pressure levels of the patient, respectively.