| This study designs and implements a distributed electrocardiogram (ECG) analyzer (“ECGOnline”) for academic use. The analyzer consists of the client and server programs. The client, with a graphic user interface, performs I/O operations, filters ECG signals, extracts ECG features, and displays ECG diagrams, features, and classification results. Four groups of ECG features—amplitudes, durations, slopes, and areas—are extracted. The server runs an adaptive weighted k nearest neighbors (AWKNN) algorithm, which was modified from the k nearest neighbors (KNN) algorithm by this study, to classify an unknown ECG by comparing its features to those of the stored training samples, and reports the results to the client. The analyzer is designed to read the II-lead ECG signals and classify them as normal or abnormal. It is implemented using the Java programming language. The test results show that the AWKNN classifier improves the traditional KNN classifier for ECG interpretation from 72.7% to 76.5% in accuracy. The addition of the slope and area features to the classic amplitude and duration features improves the performance of the analyzer from 72.5% to 76.5%. It was found that the normal ECG's are more clustered than the abnormal ECG's in the 34D feature space defined in this study. |
