Automatic recognition of arrhythmia using a CNN-based broad learning system

An accurate electrocardiogram (ECG) beat classification can benefit the diagnosis of the cardiovascular disease. Currently, automatic ECG classification systems based on deep neural network are useful in the field of ECG recognition. However, most of them are time-consuming in training, less robustness to noise, and need to retrain the entire model when new data are added. To address these problems, we propose an ECG classification algorithm by using a CNN-based broad learning system (CNNBLS) for recognition of arrhythmia. We performed two experiments to evaluate the robustness and incremental learning features of the proposed classification system. In noise robustness experiment, we selected five types of original and denoising abnormal ECGs in the MIT-BIH arrhythmia database, and overall accuracy of the five arrhythmia classifications achieved 98.5% and 98%. In incremental learning experiment, we selected 6 types of abnormal ECGs data in the MIT-BIH arrhythmia database. The accuracy and training time before incremental learning were 97.94% and 21.61 s, and the accuracy and training time after incremental learning with additional 12929 new data were 98.45% and 47.23 s. Experimental results show that our model is a practical ECG recognition method with suitable noise robustness and has superiority in training time while the accuracy is guaranteed.