Computer-aided signal analysis of measurement data supplied by medical sensor systems to extract diagnostic information, e.g. from cardiological measurements, increasingly gains in significance. Apart from the classical evaluation methods which are based on signal measurement, also methods requiring greater computational efforts can be used for large quantities of data.
With most of the methods presently used for computer-aided analysis and interpretation of an ECG, more or less complicated individual attributes and/or sets of attributes are singled out from the signal waveforms measured. These sets of attributes are then further processed by means of decision trees, neuronal networks or statistical methods so that in the end, diagnostic hints can be given to the doctor. The extent to which the ECG findings obtained by this method are correct and complete strongly depends on the sets of attributes selected and on the accuracy of parameter determination. Due to the fact that usually a combination of some hundred sets of attributes based on the actual state of knowledge is examined, the natural variability of the ECG signals of different patients is represented insufficiently only and the number of possible findings is limited.
By using signal processing algorithms which require, however, greater computational efforts, the breakdown of the ECG signals into a great number of individual parameters can be avoided and it is possible to use the ECG signal pattern as a whole for the analysis. By linking the algorithms with large ECG signal pattern libraries, the classical ECG analysis can be improved and, what is more, completely new methods of analysis can be created allowing, for example, a trend statement to be made.