Automatic detection of heart disease from twelve channel electrocardiogram waveforms

Abstract

A statistical means to automatically detect heart disease from twelve channel ECG waveforms is reported. Patient ECG records from the PTB Diagnostic ECG Database are assigned to one or more of 18 heart disease groups. A single heart beat cycle near the beginning of the record is used to measure eight temporal intervals (e.g. QT, ST, PR, QS) for each of the twelve channels, resulting in 96 parameters for each patient record. The average and standard deviation of the 96 parameters for 63 non-disease patients are used as a control reference. The number of parameters that exceed two times the standard deviation of the control reference is used as a disease indicator. Using ten or more out-of-bound parameters as the disease criteria, 72% of disease patient records are correctly identified while 8% of control patient records are falsely identified as having a heart disease.