Computer Programs in Biomedicine 9 (1979) 293-300 © Elsevier/North-Holland Biomedical Press COMPACT DIGITAL STORAGE OF ECGs Olle PAHLM, Per Ola BORJESSON and Olof WERNER Departments of ClinicaIPhysiology and Teleeommunication Theory, University of Lund, Fack S-221 O1 Lund, Sweden The technique of predictive coding is applied to the problem of reversible compression of digitized electrocardiograms. Integer-based predictors and MMSE predictors are studied as regards performance at varying sampling rates and digital resolutions for both long-term ECGs and ECGs recorded at rest. It is concluded that MMSE predictors are to be preferred only in the case when the ECG is oversampled (i.e., the sampling rate is much higher than twice the cut-off frequency of the presampling filter). In other cases the integer predictor which yields the so-called 2nd differences is superior. The problem of encoding the resulting resi- duals with a variable-length code is studied for long-term ECGs digitized at 100 Hz and using 8 bits digital resolution. The code has a simple structure leading to speed of execution while the efficiency loss is small. 1. Introduction It is often convenient to store ECGs in digital form. One advantage is the more direct computer- compatibility than is achieved by analogue storage. Also, digital storage allows repeated retrieval of exact- ly the same data, while repeated A/D-conversion of a stretch of analogue signal will result in sets of num- bers which vary slightly from one occasion to another. Further, fast random access to the data is achieved by digital storage, e.g., on a computer disk memory. In the analysis of long-term ECGs it is frequently desirable to scan the same stretch of signal repeated- ly, and to compare different incidents of arrhythmia with each other even if they occur in different seg- ments of the recording. Some sort of automatic scan- ning of the signal is also desirable in view of the large amount of data. Therefore, great gains are achieved if the whole long-term ECG is stored in the disk- memory of a computer. This allows automated ECG analysis as well as immediate display to the operator of any desired segment of the signal. If the digital samples are stored on the disk with- out prior attempts at data compression the available space could soon be filled up. For example, 12 h of ECG, sampled at 500 Hz, using 10 bits of digital reso- lution occupy 27 Mbytes. However, by carefully selecting the sampling frequency and the digital reso- lution to comply with the quality demands at hand and by applying data compression methods on the samples, storage demands can be drastically reduced. The data compression can be reversible or non- reversible. Reversability implies that the samples can be reconstructed from the compressed file entirely without error (fig. 1). There are many ways to achieve reversible data compression. It is the intention of this paper to use l--qo, I " 11 f=: I Fig. 1. Block diagram of a fairly general set up for digital storage and retrieval of ECGs. A reversible compression algorithm (sig- naly = signal x) does not add to the signal distortion, which is then determined only by the prefilter, the sampling rate, the num- ber of bits used in the A/D converter and the reconstruction filter. 293