三導程推衍十二導程心電圖之非線性演算法

利用較少的導程數推衍出準確的標準12導程心電圖應該會是未來的趨勢，但是以推衍為基礎的心電訊號如受到干擾可能會對推衍出的多個導程心電圖波形產生較大影響等問題；且針對運用最佳化的固定係數獲得的衍生12導程心電圖可能在大多數患者中非常接近標準12導程心電圖，但對於某些人卻可能出現很大的差異。故對於衍生心電圖來說，如何設計出準確的推衍算法並降低個體化差異即成為相當重要的課題。本研究主要目的在於利用遺傳演算法的全域搜索特點，對倒傳遞類神經網路進行權值優化，以此作為三導程推衍十二導程心電圖的非線性推衍方法。優化完成後的類神經網路以PTB資料庫中全部249位病患（包括：心肌梗塞、心肌病、心力衰竭、束支傳導阻滯、心律不整、心肌肥大、瓣膜性心臟病、心肌炎等病例）共549組十五導程心電訊號進行推衍與驗證。其結果並與多元線性迴歸及委員會機器方法所推衍之結果做比較。由推演結果顯示，優化完成後的類神經網路無論在均方根誤差或相關係數兩項效能指標上，皆領先多元線性迴歸方法（遺傳演算法優化完成後的類神經網路之均方根誤差：0.073±0.04，相關係數：0.898±0.043；多元線性迴歸方法之均方根誤差：0.083±0.05，相關係數：0.858±0.066）。同時與類神經網路常用於改善泛化能力的委員會機器方法做比較，兩者結果接近（委員會機器之均方根誤差：0.073±0.039，相關係數：0.895±0.047）。但運算時間卻明顯縮短（當委員成員數越多時，兩者運算時間差距更明顯），表示本研究所提出之方法，確實可由三導程推衍出令人滿意之標準十二導程心電圖，未來更可作為居家照護即時心電量測系統之設計應用。

關鍵字

類神經網路；遺傳演算法；十二導程心電圖

並列摘要

Inferring accurately the standard 12-lead ECG with less leads should be the future trend, but ECG based on inferring, if interfered, may have a greater impact on inferring a multiple lead ECG waveform. For most patients, the derived 12-lead ECG from the use of fixed coefficients may be very close to the standard 12-lead ECG, but for others, the two results may be greatly different. Therefore, in terms of the derived ECG, how to design a precise algorithm and reduce the individual differences is a very important issue. The main purpose of this study was to use the global search characteristics of genetic algorithm to process weighting optimization of the neural network, as a method of three lead ECG inferring 12-lead ECG nonlinear. After optimizing the neural network, the database from PTB 249 patients (including: myocardial infarction, cardiomyopathy, heart failure, bundle branch block, arrhythmia, cardiac hypertrophy, valvular heart disease, myocarditis and so on ), which were divided into a total of 549 groups of fifteen-lead ECG, were inferred and verified. The results were compared with those inferred by the multiple linear regression and commission machine. The results showed that the neural network after optimization surpassed the multiple linear regression method in terms of the two indicators: root mean square error and correlation coefficient (root mean square error of neural network after optimizing genetic algorithm: 0.073 ± 0.04, correlation coefficient: 0.898 ± 0.043; root mean square error of multiple linear regression: 0.083 ± 0.05, correlation coefficient: 0.858 ± 0.066). Also, when compared with the neural network commonly used to improve the generalization ability, the results were close (root mean square error of commission machine : 0.073 ± 0.039, correlation coefficient: 0.895 ± 0.047), but the computation time was significantly shorter (when members of the commission were increasing, the gap between the computing time was more obvious). This indicated that the method proposed by the study indeed could infer a satisfactory standard 12-lead ECG by using the three lead. The method could be applied to the future design of immediate cardiograph systems for home nursing.

並列關鍵字

Artificial Neural Network ； Genetic Algorithms ； 12-lead ECG

參考文獻

42. 蕭俊祥，呂理仲，"12導程心電圖快速擷取技術研發 Development of Rapid Capturing Technique for 12-Lead ECG"，臺北科技大學學報，第四十一之一期 (2008):1-9

2. Wilson FN, Johnston FD, Macleod AG, Barker PS, "Electrocardiograms that represent the potential variations of a single electrode.", Am Heart J., 9 (1934):447-58.

3. Goldberger E, "The aVL, aVR, and aVF leads; A simplification of standard lead electrocardiography.", Am. Heart J., 24 (1942):378-96.

4. Mason RE, Likar I, "A new system of multiple-lead exercise electrocardiography.", Am Heart J., 71(2) (1966):196-205.

7. Wilson FN, Johnston FD, Macleod AG, Barker PS, "Electrocardiograms that represent the potential variations of a single electrode.", Am Heart J., 9 (1934):447-58.

國際替代計量

三導程推衍十二導程心電圖之非線性演算法

全文下載

主題瀏覽