An ECG Classification Method Based on Feature Fusion Model

Electrocardiogram(ECG) is a significant ancillary tool for cardiovascular disease diagnosis. Deep learning methods have been widely applied on diagnosis tasks using ECGs. However, most methods adopt deep features alone, which have poor interpretability and make it difficult to obtain comprehensive information to better understand ECG signals. Therefore, we designed an ECG classification method based on feature fusion model, which combines deep features extracted by neural networks with shallow features consisting of statistical, time-domain and frequency-domain features. Firstly, we extracted shallow and deep features separately. Secondly, we performed feature fusion by flattening and concatenating them by column. Finally, the concatenated feature was sent to the fully connected and softmax layer for obtaining classification results. Experimental results confirmed that our method outperforms methods that used deep features only.