ECG classification algorithm using shape context

Abstract

ECG classification algorithm based on machine learning is often required to obtain a classification model by analyzing and studying a large number of sample data. Whether the distribution of sample data is uniform or whether the coverage of the sample is comprehensive determines the accuracy of the classification model finally obtained. The method proposed in this paper to classify ECG waveforms using shape context features combined with labeling information, zoom matrix mechanism, sliding comparison mechanism, standard deviation distance and Hamming distance, requires only a few typical samples. Experiments have shown that this method can obtain fairly accurate classification results for typical common heart diseases. Compared with the traditional diagnosis method based on the discriminant tree, this method has less dependence on prior knowledge, it does not need to measure the width and amplitude of each wave, and it has good robustness, and will not cause jumps in the classification results due to wave group boundary measurement errors. Compared with the machine learning method, this method does not need a large number of training sets and does not need to train the model. It is also relatively easy to extend the new disease classification type.