【作者】 周群一；

【导师】 段会龙；

【作者基本信息】 浙江大学 ， 生物医学工程， 2004， 博士

【摘要】 心拍的自动分类对于心律失常的临床诊断有着非常重要的意义。本文在总结前人工作的基础上，针对正常心拍与室性早搏的识别分类做了进一步的深入研究。 首先，本文对国内外近年内发展起来的心律失常识别的各种方法进行了分类与总结，并且针对不同方法的特点进行了详细的比较，深入探讨了各种方法存在的问题，并指出心律失常识别技术的难点。 第二，本文对QRS复合波群形态、RR间期变化和长时间心电信号的主导节律进行了详细的分析。其结果对RR间期在心律失常分类中应用具有一定的指导价值。 第三，本文提出了QRS复合波群的镜像高斯参数模型(MGM，Mirror Gauss Model)。QRS复合波群的形态变化是心律失常临床诊断的重要参考标准，因而准确地提取QRS复合波群的波形特征是心律失常自动化诊断中面临的关键问题。本文通过对QRS复合波群形态的研究，提出使用高斯函数拟合QRS复合波群的方法，并通过镜像手段，提高了拟合的准确度，从而比较准确、方便地提取了QRS复合波群的宽度特征。 第四，本文引入了基于相关性匹配与模板队列相结合的聚类模板技术。将QRS复合波群与模板队列中的模板集进行相关性匹配，如果匹配均不成功，则建立新的镜像高斯模型。通过聚类模板技术，减少了高斯拟合的次数，并且在保证检测精度的前提下，大大提高了算法的运行效率。同时，聚类模板还可以有效地识别出多种室性早搏心拍类型，为临床多形性室性早搏诊断提供了有效的方法。 最后，本文在QRS复合波群镜像高斯模型和聚类模板技术的基础上，结合数学形态学滤波、模糊推理系统等技术，实现了一个完整的正常心拍与室性早搏的识别分类算法。 通过MIT-BIH心律失常标准数据库全部48条记录对本文算法进行性能评估，正常心拍和室性早搏总体检测灵敏度分别达到93.01％和94.25％，结果优于使用同样测试集的其它算法。本文的算法在分析速度上也达到令人满意的效果。实验证明，镜像高斯模型(MGM)还可以有效地刻画其它种类的常见心律失常的QRS复合波群形态。更多还原

【Abstract】 ECG (electrocardiogram) beat classification is very important for clinical automated detection of arrhythmia. Based on previous work, this paper focuses on the further research on recognition and classification of ventricular premature beats and normal beats.Firstly, various existed methods for arrhythmia recognition are categorized and summarized. Comparing the methods in detail, the problems are discussed in depth and the main handicaps in arrhythmia analysis are given.Secondly, the shape of QRS complex wave, RR duration variation and predominant rhythm are analyzed in detail. The results are valuable to arrhythmia classification.Thirdly, the mirror Gauss model(MGM) of QRS complex wave is presented. As the variation of QRS complex wave is the primary reference criteria on clinical diagnosis of arrhythmia, exact extraction of the features of QRS complex wave is the key problem. Based on study in shape, the method of fitting QRS complex by Gauss function curve is presented to extract the width of QRS complex easily. By employing mirror operation, more high precision is achieved.Fourthly, cluster template which combine template queue with correlation match is introduced. If the QRS complex wave fails to match each template in the queue, then mirror Gauss model will be practiced. By the application of the cluster template, the frequency of Gauss fitting is greatly reduced, and even high efficiency is reached in the assurance of detection precision. On the other hand, as cluster template can distinguish multi-class of ventricular premature beats effectively, it may instruct the clinical diagnosis of multiform ventricular premature.Finally, an integrated algorithm for recognition and classification of normal and ventricular premature beats is implemented, which utilizes other technologies, such as mathematical morphology filtering and fuzzy reasoning, based on the mirror Gauss model and cluster template.The complete set of MIT-BIH arrhythmia database is used to evaluate this algorithm, the gross sensitivity of normal beats and ventricular premature beats are 93.01% and 94.25% respectively. With the result better than other algorithm which used the same test set, the analysis speed is also satisfying. It is proved by experiment that mirror Gauss model is able to effectively describe the QRS complex wave shape of other common arrhythmias.更多还原