以心電圖為基礎預測急性冠心症病人預後風險之研究

Abstract

急性冠心症肇因於心臟肌肉缺血或缺氧，是長久以來位居我國十大死因前三的重大疾病。判斷急性冠心症病患預後危險程度的系統是非常重要的研究議題。傳統上，TIMI危險分數採用了病患的病史作為判斷之依據，是目前較廣被接受的作法。近年來，由於心電圖之方便性、快速性以及非侵入性，相當多以心電圖分析系統做為急性冠心症病患預後危險評估的研究出現。 本論文提出一套對急性冠心症病患預後危險評估之心電圖分析系統，蒐集並且使用某醫院所提供的病患資料來設計及驗證本論文所提出的系統。這個系統的心電圖分析系統架構包含資料蒐集、前處理、特徵擷取、以及最後使用人工智慧判斷病患之預後狀況，於系統中每一步驟本論文均有提出改進效能之方法。 對病患危險程度分類而言，傳統醫學上的研究通常使用迴歸分析，而本論文提出使用類神經網路設計較為複雜的系統來對急性冠心症病患預後做危險評估分析，利用人工智慧協助急性冠心症預後病患之危險評估也能達到比人為判斷或用統計模型所得出之預後效果還要好的危險程度判斷效果。傳統上由於使用人為判斷或多變數迴歸分析的方式做預後判斷，特徵必須足夠精確才會有顯著之危險分層效果；使用人工智慧的好處在於，即使特徵是由複數個生理因素所影響，在取用足夠多個特徵的狀況下機器仍能自動計算出特徵內的複數個生理因素，並做出危險判斷。換言之，傳統之分析方法需人工去蕪存菁，而人工智慧之方法則是機器自動做計算。本論文之實驗證實本系統所使用之簡單特徵搭配類神經網路可以得到極佳之危險評估效能。本論文所提出的對急性冠心症病患預後危險評估之心電圖分析系統，準確率可達八成以上。

Acute coronary syndrome (ACS), caused by rupture of an atherosclerotic plaque and partial or complete thrombosis of the infarct-related artery, has been on the first three places of the top ten causes of death in Taiwan for many years. Determine the prognosis of patients with acute coronary syndrome risk level of the system is a very important research topic. Traditionally, TIMI risk score using a patient's medical history as a basis for judgment, is the wide acceptance of the practice. In recent years, because the ECG is convenience, rapid and non-invasive, there are many studies about ECG analysis system as a prognosis of patients with acute coronary syndrome risk stratification. A novel ECG analysis system for post-acute-coronary-syndrome (post-ACS) patients is proposed and applied to clinical data collected from a hospital. The ECG analysis system involves pre-processing, beat detection, feature extraction, and using machine learning to make prediction, is proposed. Each part in this system is enhanced by new algorithms or techniques. Instead of using regression analysis used in traditional biomedical researches, we use artificial neural network to build a more sophisticated model to classify the patients. Machine learning like artificial neural network can find the information hidden in the features, even if the used features are affected by multiple medical factors, while users have to use much more sophisticated features when they make prognosis by inspection or multivariate statistical analysis. The system we proposed gives the accuracy over 80% for stratification using only eight ECG features.