Redundancy reduction for improved display and analysis of body surface potential maps. II. Temporal compression

Abstract

This paper describes the use of the Karhunen-Loeve expansion to identify and reduce temporal redundancy in electrocardiographic body surface potential maps (192 body surface leads recorded simultaneously at 1 kHz/channel for approximately 600 msec). Temporal data compression of about 20 to 1 was obtained with accurate representation of the original data. Use of separate sets of orthonormal basis functions for QRS and ST-T provided a more accurate representation than the basis derived from QRST. Combined with the spatial compression described in the preceeding paper, overall map data compression of about 320 to 1 was obtained without significant loss of accuracy of representation or map appearance. With both spatial and temporal compression the 100,000 numbers which typically comprise a single cardiac complex were accurately represented by 216 coefficients. Using basis functions derived from a training set of 221 maps, the estimated average rms error of representation was 60 μV during the QRS and 25 μV during the ST-T. For 34 test maps which were not part of the training set, measured average errors were 64 μv during the QRS and 23 μV during ST-T. This technique provides a basis for quantification of the diagnostic content of maps and automated classification of maps.