基於機器學習技術之心電圖訊號分析研究

Analysis of Electrocardiogram Signal Based on Machine Learning Technology

黃璿彰(Hsuan-Chang Huang)

指導教授 ： 鮑永誠

高雄醫學大學/健康科學院/醫務管理暨醫療資訊學系碩士班/碩士(2021年)

摘要

心血管疾病(Cardiovascular Disease，CVD)是全球死亡的首要原因。其中有部分患者是因為調節穩定心跳的系統異常，而導致心律異常，更嚴重的情況，可能導致心臟瘁死。心電圖(Electrocardiography，ECG)記錄了患者心臟的生理信號。醫生可以透過觀察並以心電圖來判斷給出相應的症狀。對於大量的心電圖數據，單純只依靠醫生的診斷是非常沒有效率的。因此本研究欲使用機器學習的方法，透過數據分析過濾大量的心電訊號，標示出與醫生判斷過的結果相似訊號，藉此篩選出可能的異常訊號，輔助醫生判斷，有效減少醫生的工作量。本論文中，使用卷積神經網路(Convolutional Neural Networks，CNN)技術的心電圖訊號分類方法，該方法由四個部分組成，首先利用移動平均，將特定部分訊號平滑化，以利波形識別。第二，將心電圖數位訊號以斜率來偵測訊號中較具變化的心跳波形擷取出來作為訓練樣本。第三，將擷取出的心電圖訊號樣本，分為訓練樣本及測試樣本，並特過一維卷積神經網路模型訓練樣本。第四，將測試數據透過模型預測出心電圖訊號的分類。我們提出的方法有下列的優點: (1) 從大量的心電圖中自動偵測完整心跳位置，可省去許多人工處理的時間; (2) 利用卷積神經網路的模型預測，判斷出心電圖訊號分類結果。

關鍵字

心電圖 ； 機器學習 ； 卷積神經網絡 ； 分類

並列摘要

Cardiovascular disease (CVD) is the leading cause of death worldwide. Some patients have abnormal heart rhythm due to the system that regulates and stabilize the heartbeat. In more serious cases, the heart may die. Electrocardiogram (ECG) records the physiological signals of the patient's heart. The doctor can judge the corresponding symptoms by observing and using the electrocardiogram. For a large amount of ECG data, it is very inefficient to rely solely on doctor's diagnosis. Therefore, this research intends to use machine learning methods to filter a large number of ECG signals through data analysis, and mark signals similar to those judged by doctors, so as to screen out possible abnormal signals, assist doctors in judgment, and effectively reduce doctors’ workload . In this paper, the ECG signal classification method using Convolutional Neural Networks (CNN) technology is composed of four parts. First, moving average is used to smooth the specific part of the signal to facilitate waveform recognition. Second, extract the heartbeat waveform of the ECG digital signal with the slope to detect the more changes in the signal as a training sample. Third, the extracted ECG signal samples are divided into training samples and test samples, and special training samples for one-dimensional convolutional neural network models are used. Fourth, use the test data to predict the classification of the ECG signal through the model. The method we propose has the following advantages: (1) Automatically detect the complete heartbeat position from a large number of ECGs, which can save a lot of manual processing time; (2) Use the model prediction of the convolutional neural network to determine the result of ECG signal classification.

並列關鍵字

Electrocardiogram ； Machine Learning ； Convolutional Neural Network ； Classification

參考文獻

1. Singh, Y.N. and P. Gupta. ECG to individual identification. in 2008 IEEE Second International Conference on Biometrics: Theory, Applications and Systems. 2008. IEEE.

2. Sufi, F., I. Khalil, and I. Habib, Polynomial distance measurement for ECG based biometric authentication. Security and Communication Networks, 2010. 3(4): p. 303-319.

3. Li, C., C. Zheng, and C. Tai, Detection of ECG characteristic points using wavelet transforms. IEEE Transactions on biomedical Engineering, 1995. 42(1): p. 21-28.

4. Gritzali, F., G. Frangakis, and G. Papakonstantinou, Detection of the P and T waves in an ECG. Computers and Biomedical Research, 1989. 22(1): p. 83-91.

5. Xue, Q., Y.H. Hu, and W.J. Tompkins, Neural-network-based adaptive matched filtering for QRS detection. IEEE Transactions on biomedical Engineering, 1992. 39(4): p. 317-329.