根據衛生署所公布的十大死因統計,心臟疾病中以心肌梗塞(Myocardial Infarction)為高死亡率的病症,且需要快速且準確的診斷以爭取時間急救。心肌梗塞的發生是因為心臟所需血液受阻,而醫師在診斷心肌梗塞病症時,主是依據12導程心電圖的ST段變化來診斷,因此本研究希望從12導程心電圖萃取重要特徵,輔助醫師進一步診斷,避免醫師因過勞而誤診。本研究資料以PTB資料庫的12導程心電圖進行實驗分析,首先利用低通濾波器(Low-Pass Filter)解決高頻雜訊問題,並利用經驗模態分解(Empirical Mode Decomposition)與中值濾波器(Median Filter)去除基線飄移,再結合多項式近似法(Polynomial Approximation)和主成份分析(Principle Component Analysis)的優點進行特徵擷取,多項式近似法可描述ST段波形並轉換成係數,而主成份分析可縮減維度,減少資料複雜度提高分類效果。實驗結果顯示本研究所提出的多項式近似法結合主成份分析進行特徵擷取,無論是在支持向量機(Support Vector Machine)或類神經網路(Neural Network)分類器上皆優於ST段為基礎的特徵,分類準確度最高達到98.28%。有此可知,有效的特徵擷取可提升分類效果,故本研究在心肌梗塞的辨識上有顯著且穩定的效果,並有效降低醫師誤診的情況,提高醫療品質。
Myocardial infarction is high mortality which needs rapid and accurate diagnosis for saving the time of treatment. Myocardial infarction occurs when the blood supply to the part of heart is interrupted. The clinical diagnosis in myocardial infarctions is bases on ST-segment in 12 lead ECG. The purpose of this study is to extract significant features from the ST-segment. Furthermore, this research can support physician for further diagnosis and avoid the misdiagnosis. In this thesis, firstly, the low-pass filter is adopted to depress the high-frequency noise, and EMD and median filter are used to remove the baseline drift. Secondly, this study combines the advantages of polynomial approximation and principal component analysis. Polynomial approximation can describe the morphology of ST segment change by using the coefficients and principal component analysis can reduce dimensions and generate the significant features. Experimental results show our feature extraction is better than other ST-based, and the accuracy can achieve 98.28%. In conclusion, that effective approach for feature extraction can improve the classification performance, and the proposed approach is also stable. According to the analysis the outcome for this study can be used to improve the quality of healthcare.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。