以單導程心電圖訊號進行心房顫動偵測之研究

Atrial fibrillation detection based on single lead ECG

高東韋(Dong-Wei Gao)

指導教授 ： 曹恆偉

國立臺灣大學/電機資訊學院/電子工程學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101035

摘要

科技日新月異，隨著時代的進步，智慧型裝置運用在疾病診斷愈來愈受到重視，就以近幾年的資訊科技來說，舉凡人工智慧、大數據、雲端存取及遠端運算都已經大量運用在醫療領域。這些應用可用在病患的生理監測和監控、行動照護甚至在一些突發的緊急事件上做出預警，以期能告知病患盡早就醫達到預防醫學的目標。

心房顫動（Atrial fibrillation）是心律不整的一種，其特點是心臟會呈現快速並且不規則之心律，心房顫動起初發生的持續時間非常短暫並且具有偶發性，但隨著時間推移發作時間會愈來愈長到不易緩解的程度。為了能達到提早預防，本論文使用近幾年快速崛起的深度學習及機器學習技術，提出一個系統化並且客觀的方法，在既有之大量資料的情況下，利用單導程心電訊號將病人心房顫動狀態辨識出來，提供醫師參考及輔助診斷病患的疾病。

本論文第一章為緒論，簡介研究主題的原由並提出改良方法及貢獻，第二章以及第三章為心房顫動及機器學習的背景知識介紹，從第四章開始為本論文的主要貢獻，其中包含減少特徵值和運算量並把實際患病的人數估算出來並搭配統計數據(靈敏度、特異度以及F1 Score)來衡量判斷的正確率，最後採用不同的分類法，提出不同的系統架構設計，包含心電圖訊號的前處理、特徵擷取、深度殘差網路運算、十折交叉驗證及分類器的運用，最後把判別結果輸出。接下來在第五章探討模擬結果並配合統計圖表及前人所做的數據交叉比對來驗證此系統的效能，最後一章則是結語和未來展望，提出電極擷取訊號及使用者穿戴便利性的討論，並提出未來的改良建議像是建立病況嚴重程度等級等。

關鍵字

單導程 ； 心電圖 ； 心房顫動 ； 機器學習 ； 深度學習

並列摘要

Technology is changing with each passing day. With the advancement of the times, smart diagnosis has received more and more attention. In recent years, in the case of information technology, artificial intelligence, big data, cloud access, and remote computing have been widely used in the medical field. These applications can be used in the physiological monitoring and monitoring of patients, mobile care, and even early warning of emergencies, so as to inform patients that they can only seek medical treatment early to achieve the goal of preventive medicine.

Atrial fibrillation is a type of arrhythmia, which is characterized by a rapid and irregular heart rhythm. Atrial fibrillation has a very short duration and is sporadic at first, but the attack time will get more and more time. It grows to the point that it is not easy to relieve. In order to achieve early prevention, this thesis uses the deep learning and machine learning technologies that have emerged rapidly in recent years to propose a systematic and objective method to use single-lead ECG signals to detect patients. The state of atrial fibrillation can be distinguished, which can provide reference for cardiologists and assist in diagnosing the patient's disease.

The first chapter of this thesis is an introduction, introducing the reason for the research topic and proposing improved methods and contributions. Chapters 2 and 3 are introductions to the background knowledge of atrial fibrillation and machine learning. Starting from Chapter 4, the main contribution of this thesis. It includes reducing the feature and the amount of calculation, and estimating the actual number of people who are sick, and using statistical data (sensitivity, specificity, and F1 Score) to measure the correctness of the judgment. Finally, different classification methods are used to propose different system architecture designs, including the pre-processing of the electrocardiogram signal, feature extraction, deep residual network calculation, ten-fold cross-validation and the application of the classifier, and finally output the discrimination result. Next, in Chapter 5, we will discuss the simulation results and verify the performance of this system with statistical charts and data cross-comparisons done by predecessors. The last chapter is the conclusion and future prospects, proposing electrode acquisition signals and user convenience discussions between them, and put forward suggestions for future improvements, such as establishing a severity level of the condition.

並列關鍵字

ECG ； atrial fibrillation ； machine learning ； deep learning ； single lead

參考文獻

[1] Anumonwo, JM, Kalifa, Risk Factors and Genetics of Atrial Fibrillation Cardiology clinics, 32 (4): 485–494, November 2014.

[2] Munger, TM, Wu, LQ, Shen, WK, Atrial fibrillation Journal of biomedical research, 28 (1): 1–17, January 2014.

[3] Mischke, Knackstedt, Marx, Vollmann, Insights into atrial fibrillation, Minerva medica, 104 (2): 119–30, April 2013.

[4] Ferguson C, Inglis SC, Newton PJ, Middleton S, Macdonald PS, Davidson PM, Newton, Middleton, MacDonald, Davidson, Atrial fibrillation: stroke prevention in focus, 00 (2): 92–8, 2014.

[5] Hampton, John R, The ECG Made Easy 8th. Edinburgh: Churchill Livingstone, 4 2013.