基于多特征融合的J波分类模型

J wave classification model based on multi-feature fusion

J波是心电图上出现的一种异常变异。应用计算机实现J波自动分类对J波疾病的临床诊断有着重要意义。基于时频域和相空间两个分析角度,一方面使用调Q小波变换(Tunable Q Wavelet Transform,TQWT)和高阶累积量挖掘信号时频域的细节特性;另一方面应用递归图(Recurrence plot,RP)评估心脏系统递归点的发生状态。两类特征降维后并行融合于改进的AdaBoost分类器实现正常、良性J波和恶性J波分类。结果显示,设计的J波多分类算法平均准确度约达到79%,可以用于J波良、恶性辅助诊断。

J wave is an abnormal variation in the electrocardiogram(ECG).It is of great significance to realize J wave automatic classification for the clinical diagnosis of J wave related disease.In this paper,features are extracted based on time-frequency domain and phase space.On the one hand,tunable-Q wavelet transform(TQWT)and high-order cumulant are used to mine the detail characteristics.On the other hand,recurrence plot(RP)is adopted to show the occurrence of recursive points in the cardiac system.After dimensionality reduction,the two types of features,which are fused in parallel,are input to the improved AdaBoost classifier to realize the classification of normal beats,benign J wave beats and malignant J wave beats.The experimental results show that the average accuracy of the classification is about 79%,which is helpful for the diagnosis of benign and malignant J wave.