Automatic classifiers for the interpretation of electrocardiograms

The morphological diagnosis of ECGs is a pattern recognition procedure. The way the clinician does this is not clearly elucidated. Nevertheless, several models aimed at achieving identical results by automatic means are empleyed. While in the doctor's case this is not exactly so, the computer task for ECG interpretation comprises two distinct and sequential phases: feature extraction and classification. A set of signal measurements containing information for the characterization of the waveform is first obtained. These waveform descriptors are then used to allocate the ECG to one or more diagnostic classes in the classification phase. The classifier can embody rules-of-thumb used by the clinician to decide between conflicting ECG diagnosis and formal or fuzzy logic as a reasoning tool (heuristic classifiers). On the other hand, it can use complex and even abstract signal features as waveform descriptors and different discriminant function models for class allocation (statistical classifiers). More recently, artificial neural network techniques have also been used for signal classification. The authors review feature selection techniques and classification strategies, problems and methods of performance evaluation and results obtained by different classification approaches. A brief discussion of the relative merits of the two main types of ECG classifiers, logical and statistical, is included.