A robust sensor fusion method for heart rate estimation.

Problem. Physiologic data measured in the clinical environment isfrequently corrupted causing erroneous data to be displayed, periods ofmissing information or nuisance alarms to be triggered. To date, thepossibility of combining sensors with similar information to improve thequality of the extracted data has not been developed. The objective of thiswork is to develop a method for combining heart rate measurements frommultiple sensors to obtain: (i) an estimate of heart rate that is free ofartifact; (ii) a confidence value associated with every heart rate estimatewhich indicates the likelihood that an estimate is correct; (iii) a moreaccurate estimate of heart rate than is available from any individualsensor. Solution. The essence of the method is to discriminate between goodand bad sensor measurements and combine only the good readings to derive anoptimal heart rate estimate. Past estimates of heart rate are used to derivea predicted value for the current heart rate that is also fused along withthe sensor measurements. Consensus between sensor measurements, thepredicted value and physiologic credibility of the readings are used todistinguish between good and bad readings. Three sensor measurements and thepredicted value are evaluated yielding 16 possible hypotheses for thecurrent state of the available data. A Kalman filter uses the most likelyhypothesis to derive the fused estimate. Statistical measures of the sensorerror and rate of change of heart rate are adaptively estimated when dataare sufficiently reliable and used to enhance the hypothesis selectionprocess. Discussion. The method of sensor fusion presented has beendocumented to perform well using clinical data. Limitations of the techniqueand the assumptions employed are discussed as well as directions for futureresearch.