A wavelet scheme for reconstruction of missing sections in time series signals

This work proposes a wavelet scheme to reconstruct missing data in physiologic signals that have been removed from multi-parameter recordings of patients in intensive care units. According to the proposed strategy, the missing data section is estimated based on two other sections. If the signal to be reconstructed is an ECG, the two sections are obtained from the two other ECG derivations available in the record. Otherwise, if the incomplete signal has only one derivation, the two sections are obtained from the signal itself, by means of a pattern matching procedure. In both cases the removed section of the signal is estimated based on a strategy that combines wavelet decomposition with autoregressive models. Applied to all records of set A, set B and Set C, this strategy provided results of (47.75%/56.25%), (46.52%/15430%) and (38.33%/47.33%) for scores 1 and 2, respectively.