利用二維卷積神經網路進行二維心電圖特徵之病症分類及辨識

Classification and identification of symptoms using two-dimensional convolutional neural networks on two-dimensional electrocardiographic features

李宗翰(Tsung-Han Lee) ； 蔡宗穎(Tsung-Ying Tsai) ； 徐良育(Liang-Yu Shyu)

《先進工程學刊》 15卷3期 (2020/10) Pp. 95-99

摘要

本研究提出一種利用12導程心電圖（Electrocardiogram，ECG）之二維特徵圖同時分類多種病症之自動分類方法，此方法中設計一個3層雙通道卷積神經網絡（Convolutional Neural Networks，CNN），輸入為12導程ECG訊號分別對應第2導程ECG訊號製作的二維特徵圖圖像。為了使CNN能夠更有效的學習ECG病徵，分別將原始ECG圖像分割為包含QRS複合波特徵之圖像及不包含QRS複合波特徵之圖像做為雙通道之輸入，並探討CNN架構中輸入圖像之大小與隱藏層層數對於病症分類準確度及時間之影響。本研究中用來訓練及測試的受測者共182名，並使用10層交叉驗證以驗證CNN分類之能力。在13種病症之分類中，本系統分類平均準確度為84.92%、靈敏度為82.40%、特異性為82.79%，此結果證明本系統可協助專家分析12導程ECG中之特徵，並加以分類其病症。

關鍵字

心電圖 ； 12導程 ； 二維特徵圖 ； 卷積神經網路 ； 病症分類

並列摘要

This study proposes an automatic classification method for simultaneously classifying multiple symptoms using a two-dimensional feature image of 12-lead Electrocardiogram (ECG). In this method, a 3-layer two-channel convolutional neural network (CNN) is designed. The input are two-dimensional feature images that were pairing leadⅡ signal with all the other leads. To enable the CNN to learn ECG symptoms more effectively, the original ECG image was segmented into two images, one containing a QRS complex feature and an image does not including any QRS complex feature. Then the impact of the size of the input image and the number of hidden layers in the CNN architecture on the classification accuracy and time of the disease were explored. In this study, a total of 182 subjects were included for training and testing. A 10-fold cross validation were used to verify the capabilities of the CNN classification. Among the 13 types of symptoms, the average accuracy of the system classification was 84.92%, the sensitivity was 82.40%, and the specificity was 82.79%. This result indicates that the system can assist experts in analyzing the features of the 12-lead ECG and classifying the symptoms.

並列關鍵字

ECG ； 12 lead ； two-dimensional feature images ； convolutional neural network ； classification of symptoms

參考文獻

Addison, P.S.(2005).Wavelet transforms and the ECG: a review.Physiological measurement.26(5),R155.

van Ginneken, B.(2017).Fifty years of computer analysis in chest imaging: rule-based, machine learning, deep learning.Radiological physics and technology.10(1),23-32.

Kallenberg, M.,Petersen, K.,Nielsen, M.,Ng, A.Y.,Diao, P.,Igel, C.,Vachon, C.M.,Holland, K.,Winkel, R.R.,Karssemeijer, N.(2016).Unsupervised deep learning applied to breast density segmentation and mammographic risk scoring.IEEE transactions on medical imaging.35(5),1322-1331.

Andreão, R.V.,Muller, S.M.,Boudy, J.,Dorizzi, B.,Bastos-Filho, T.F.,Sarcinelli-Filho, M.(2008).Incremental HMM training applied to ECG signal analysis.Computers in biology and medicine.38(6),659-667.

Dumont, J.,Hernández, A.I.,Fleureau, J.,Carrault, G.(2008).Modelling temporal evolution of cardiac electrophysiological features using hidden semi-Markov models.2008 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society.(2008 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society).