本文選題：ECG + 異源心電數(shù)據(jù)庫��； 參考：《哈爾濱工業(yè)大學(xué)》2016年碩士論文

【摘要】：心電圖(Electrocardiogram,ECG)從19世紀(jì)被應(yīng)用于臨床醫(yī)學(xué)以來,一直在疾病診斷過程中扮演著重要角色。通過對前人研究的總結(jié)和分析,可以發(fā)現(xiàn)心電信號處理算法非常的豐富,但很少有涉及到針對異源心電數(shù)據(jù)庫問題的研究。然而現(xiàn)實的應(yīng)用往往針對異源數(shù)據(jù)庫,所以異源數(shù)據(jù)庫的研究是非常有必要和價值的。另外,神經(jīng)網(wǎng)絡(luò)被廣泛應(yīng)用于心電識別領(lǐng)域,尤其是BP神經(jīng)網(wǎng)絡(luò)。但通過深入研究和大量實驗發(fā)現(xiàn),當(dāng)樣本容量逐漸增大相應(yīng)測試樣本的容量也隨之增大時,基于BPNN的分類中,測試樣本輸出結(jié)果的區(qū)分度會越來越小。這必然導(dǎo)致很多樣本被錯誤的歸類,所以解決BP神經(jīng)網(wǎng)絡(luò)的這一缺陷也是非常有必要的。針對上面出現(xiàn)的這兩個問題,本文提出了解決異源數(shù)據(jù)庫應(yīng)用障礙的預(yù)處理方法和基于BP神經(jīng)網(wǎng)絡(luò)的多階分類算法。通過對心電圖診斷過程的充分調(diào)研,提出了去噪及采樣頻率轉(zhuǎn)換的預(yù)處理方法。結(jié)合三個異源心電數(shù)據(jù)庫,進(jìn)行了異源數(shù)據(jù)庫處理并通過熵值、標(biāo)準(zhǔn)差、峰度和偏度四個相似度指標(biāo)對處理前后三個數(shù)據(jù)庫信號的相似度做了對比。實驗表明,提出的預(yù)處理方案具有較好的效果。另外,結(jié)合MIT-BIH心率失常數(shù)據(jù)庫和PTB心電數(shù)據(jù)庫,進(jìn)行了異源數(shù)據(jù)庫間的疾病診斷實驗,實驗表明本文提出的預(yù)處理方案可以解決異源數(shù)據(jù)庫之間疾病診斷問題。為了解決BP神經(jīng)網(wǎng)絡(luò)分類區(qū)分度變小的問題,在對BP神經(jīng)網(wǎng)絡(luò)深入研究之后,本文通過增加多個假測試集進(jìn)行多階網(wǎng)絡(luò)訓(xùn)練。得到多個訓(xùn)練網(wǎng)絡(luò),最后應(yīng)用訓(xùn)練得到的多個網(wǎng)絡(luò)對測試集進(jìn)行分類,這樣便解決了區(qū)分度下降的問題。然后針對MIT-BIH心率失常數(shù)據(jù)庫中5類心搏進(jìn)行了分類實驗,結(jié)合波形特征提取方式識別率可以達(dá)到96.8%,比常規(guī)的BP神經(jīng)網(wǎng)絡(luò)算法識別率高出5%左右。另外,本文還利用SVM算法結(jié)合小波特征和波形特征以及調(diào)研中的較好文獻(xiàn)方法,進(jìn)行了分類實驗,實驗對比均顯示本文提出的基于BP神經(jīng)網(wǎng)絡(luò)的多階分類算法效果要好于其他方法。在完成前面研究工作之后,以此作為基礎(chǔ),本課題實現(xiàn)了基于心電圖的心率失常診斷系統(tǒng)。診斷系統(tǒng)以普遍的Android智能手機(jī)作為客戶端載體,將利用便攜式設(shè)備采集到的心電數(shù)據(jù)通過無線網(wǎng)絡(luò)傳送到服務(wù)器,然后進(jìn)行系統(tǒng)化的診斷,并將診斷結(jié)果反饋回手機(jī)供客戶參考。
[Abstract]:Electrocardiogram (ECG) has played an important role in the diagnosis of diseases since it was used in clinical medicine in the 19th century.Through the summary and analysis of previous studies, we can find that ECG processing algorithms are very rich, but there are few researches on the problem of heterogeneous ECG database.However, the actual application is often aimed at heterologous database, so the research of heterologous database is very necessary and valuable.In addition, neural network is widely used in the field of ECG recognition, especially BP neural network.However, through in-depth research and a large number of experiments, it is found that when the sample size increases with the increase of the corresponding test sample capacity, the classification degree of the test sample output will become smaller and smaller in the classification based on BPNN.This will inevitably result in many samples being misclassified, so it is necessary to solve this defect of BP neural network.In view of the above two problems, this paper proposes a preprocessing method to solve the application obstacle of heterogeneous database and a multi-order classification algorithm based on BP neural network.A preprocessing method of denoising and sampling frequency conversion is put forward by investigating the diagnosis process of electrocardiogram (ECG).Combining with three different ECG databases, the paper makes a comparison of the similarity of the three database signals before and after processing through four similarity indexes: entropy, standard deviation, kurtosis and skewness.The experimental results show that the proposed pretreatment scheme has a better effect.In addition, combined with MIT-BIH arrhythmia database and PTB ECG database, the disease diagnosis experiment between heterologous databases is carried out. The experiment shows that the proposed preprocessing scheme can solve the problem of disease diagnosis between heterogeneous databases.In order to solve the problem that the classification degree of BP neural network becomes smaller, after the in-depth study of BP neural network, the multi-order network training is carried out by adding multiple false test sets.Finally, several training networks are used to classify the test set, which solves the problem of the degree of discrimination decreasing.Then, the classification experiments of 5 kinds of cardiac beats in MIT-BIH arrhythmia database are carried out. The recognition rate of the five types of cardiac beats based on waveform feature extraction can reach 96.8%, which is about 5% higher than the recognition rate of the conventional BP neural network algorithm.In addition, the SVM algorithm is used to combine wavelet features and waveform features, as well as the better literature methods in the research, and the classification experiments are carried out.The experimental results show that the proposed multi-order classification algorithm based on BP neural network is better than other methods.On the basis of the above research work, a ECG based diagnosis system for heart rate disorders is implemented in this paper.The diagnosis system takes the universal Android smart phone as the client carrier, transmits the ECG data collected by the portable device to the server through the wireless network, and then carries on the systematic diagnosis.And the results of the diagnosis back to the mobile phone for customer reference.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2016
【分類號】：R540.41;TP311.13

【相似文獻(xiàn)】

相關(guān)會議論文 前1條

1 李輝華;付小愛;鄭勤;祝小靜;熊秀忠;鄭惠伶;張云;劉敬芹;;三大引文庫收刊特色分析[A];華北地區(qū)高校圖協(xié)第二十四屆學(xué)術(shù)年會論文（文章）匯編[C];2010年

相關(guān)重要報紙文章 前6條

1 廣冉邋記者 云全;632家企業(yè)納入重大危險源數(shù)據(jù)庫[N];煙臺日報;2008年

2 姚耀富;江蘇省南京市 建成重大危險源數(shù)據(jù)庫[N];中國安全生產(chǎn)報;2009年

3 記者　余義勇;重大危險源數(shù)據(jù)庫兩年建成[N];四川日報;2006年

4 通訊員 汪翠華　嚴(yán)莉;從“信息孤島”到“共享園地”[N];中國民航報;2009年

5 記者 郭志剛;清除放射源造成的安全隱患[N];吉林日報;2004年

6 侍井林 江蘇省鹽城市濱�？h安監(jiān)局;監(jiān)管重大危險源應(yīng)發(fā)揮兩個責(zé)任主體積極性[N];中國安全生產(chǎn)報;2010年

相關(guān)博士學(xué)位論文 前1條

1 梁平;面向產(chǎn)品生命周期數(shù)據(jù)的XML本源數(shù)據(jù)庫的研究[D];合肥工業(yè)大學(xué);2006年

相關(guān)碩士學(xué)位論文 前4條

1 董海廷;基于心電圖的異源數(shù)據(jù)庫及診斷算法研究[D];哈爾濱工業(yè)大學(xué);2016年

2 王文超;NHX查詢處理及訪問控制的研究與實現(xiàn)[D];南京航空航天大學(xué);2007年

3 夏斌;XML本源數(shù)據(jù)庫及其訪問控制的研究[D];南京航空航天大學(xué);2006年

4 肖建立;基于語義的股評信息的推理與系統(tǒng)架構(gòu)的研究[D];武漢理工大學(xué);2011年

，

本文編號：1772449