Level-crossing approach to the compression of body surface potential maps

Abstract

Body Surface Potential Mapping (BSPM) is a technique for the high-resolution study and analysis of the electrical events in the heart. The technical complications and the high cost of custom-designed systems have limited the use of BSPM to a small number of laboratories. Almost all efforts to simplify BSPM systems have concentrated so far on the reduction of the number of the electrodes employed with subsequent extrapolation of the measured information to the whole thoracic surface. In this report a new approach for simplifying the BSPM system is suggested. The system uses a different type of electrodes, which do not record the ECG signal, but only detect the time instants when the incoming ECG signal crosses some given threshold level (Level-Crossings (LC)). Different threshold levels may be attributed to different subsets of electrodes. An algorithm for the reconstruction of complete BSPM information from those LC is proposed, based on utilization of the Karhunen-Loeve expansion method. This approach is justified by treating the BSPM signals as samples of a three-dimensional random process. The complete recordings of 200 BSPMs (training set) with different cardiac pathologies is used to derive the set of basis functions for the reconstruction procedure. The reconstructions are then performed on the LC of 81 subjects (test set) with a distribution of pathologies similar to that of the training set. Best performance is found when four threshold levels (± 100, ± 500 μV) are selected, for four intermittent grids of 45 electrodes, thus comprising a whole net of 180 electrodes. The average RMS reconstruction error for a member of the rest set is about 136 μV, whereas the correlation between the original and estimated BSPMs is around 0.91. These results suggest that by implementing an efficient signal compression scheme, even with the limited information provided by the LC, a clinically useful reconstruction of the BSPM is obtained.