Waveform recognition with 10,000 ECGs

A method for the comparison and interpretation of 12-lead ECGs without feature extraction is introduced. One beat of the test ECG with unknown diagnosis is compared with ECGs from an ECG database with available diagnoses. With the aid of modified cross correlation methods, reference cases are selected from the ECG database which best match the signal patterns of the unknown EGG. When ECGs with similar waveforms have been found, the results of the corresponding cardiac findings can be used to interpret the unknown EGG. The method is tested using a database with 10,000 ECGs. The present paper briefly demonstrates the method. Investigated attributes of ECG classifications are explained. The method is tested with 8,500 patients or healthy persons. A method of multidimensional clustering with weighted diagnoses is used to show some classification results. The sensitivity and specificity is given for several diagnostic groups. For classifying further attributes, we choose “sex” as a criterion. The results are demonstrated on the basis of 3,819 ECGs stemming from healthy persons