RESUMO
Most systems for the automatic detection of abnormalities in the ECG require prior knowledge of normal and abnormal ECG morphology from pre-existing databases. An automated system for abnormality detection has been developed based on learning normal ECG morphology directly from the patient. The quantisation error from a self-organising map 'learns' the form of the patient's ECG and detects any change in its morphology. The system does not require prior knowledge of normal and abnormal morphologies. It was tested on 76 records from the European Society of Cardiology database and detected 90.5% of those first abnormalities declared by the database to be ischaemic. The system also responded to abnormalities arising from ECG axis changes and slow baseline drifts and revealed that ischaemic episodes are often followed by long-term changes in ECG morphology.