Pattern detection in vital signals daily collected by means of a tele-monitoring application

Abstract

This work deals with methods for similarity measuring and indexing in physiological time series. A similarity measure is introduced based on the natural set of features generated by the Haar wavelet transform, which reflect the fundamental dynamics of a time series. Basically, a time series transformation procedure is carried out by means of a set of orthogonal wavelet basis functions, which is then reduced to an optimal subset through the application of the Karhunen Loève transform. As result, a physiological signal is efficiently described by a linear combination of a reduced set of wavelet basis with the corresponding coefficients reflecting its main dynamic behavior. These coefficients are then used to efficiently evaluate the similarity between biosignals, allowing a significant reduction of the computational complexity of the method. The validation of the proposed similarity is assessed by the comparison with other common measures, when several variations in a baseline are introduced. For this purpose, blood pressure and heart rate daily collected by means of a tele-monitoring application (TEN-HMS) are employed. The obtained results suggest that the proposed similarity is particularly appropriate to deal with noise, trends and signals that are not perfectly aligned in time.