Automatic neural detection of anomalies in electrocardiogram (ECG) signals

A two-stage architecture is proposed for recognition of five types of ill complexes in ECG signals. First, a classical signal processing method allows detection of relevant portions of the signal (QRS complexes), and reduction of the information needed for classification. Second, a neural network architecture is used for classification, implying a Kohonen map and a perceptron, with the cardiologist as a supervisor. Two types of troubles are perfectly recognized, while the three others remain hard to detect as such. >