Wavelet based compression of Holter ECG signals

Presents a wavelet based algorithm for compression of long ECG data records, typically associated with a Holter. The algorithm takes special use of the time-domain morphology of the signal as well as its clinical importance. As the Holter is processed, one builds a growing dictionary of heartbeats. Each time a new heartbeat is detected, it is compared to the heartbeats already stored in the dictionary. If one finds a dictionary entry close enough to the new heartbeat, the heartbeat is coded as a pointer to that particular entry. If no entry is similar to the new heartbeat, one codes it by finding the wavelet that maximizes the energy of its projection and keeping the corresponding wavelet and projection coefficients. While the final performance of the algorithm is data dependent, typical compression ratios obtained are larger than 30:1. This number compares very favorably to the typical 10:1 reported in the literature