Self-organizing maps and Hermite functions for classification of ECG complexes

The authors present a method for classifying ECG complexes from the MIT-BIH Arrhythmia Database based on Self-Organizing Maps (SOMs). QRS complexes were extracted and RR intervals calculated. The information contained in each beat was represented by the coefficients of the decomposition with the first three Hermite basis functions. These features in combination with measurements of RR intervals were clustered with a SOM. The resulting classifications were compared with the classifications annotated in the database and the classifications performed by a more conventional method based on template cross-correlation. The percentage of misclassifications with the SOM method was 2.4%, which was substantially lower than the 4.4% of the correlation-based method.