小波包分析和支持向量机用于直肠感知功能重建

Rectal Perception Function Rebuilding Based on Wavelet Packet Analysis and Support Vector Machine

针对临床上肛门失禁导致的直肠感知功能丧失,提出一种基于小波包分析和支持向量机(support vector machine,SVM)重建患者直肠感知功能的方法.分析人体直肠生理特征,将典型直肠压力收缩波形中的巨大移行性收缩作为产生便意的主要依据.利用小波包分析对直肠压力信号进行特征提取,以分解层结点的L2范数和标准差作为特征向量.通过提取的直肠压力信号特征向量对基于SVM的直肠感知预测模型进行训练,对SVM的惩罚因子和核函数宽度进行交叉验证优化,并利用训练后的模型进行便意预测,同时对比分析了基于前馈神经网络和基于不同核函数的SVM便意预测的准确率.实验结果表明,所提出的方法能帮助患者重建直肠感知功能.

To solve the problem of rectal perception loss caused by anal incontinence, a rectal perception function rebuilding method is proposed based on wavelet packet analysis and support vector machine （SVM）. By analyzing the characteristics of human rectum, high-amplitude propagated contractions （HAPC） in rectal contractions are used to indicate an urge to defecate. Feature extraction of rectal pressure is done using wavelet packet analysis, and take L2 norm and standard deviation of decomposition nodes as eigenvector. A rectal perception prediction model is trained using SVM. By extracting eigenvector from rectal pressure signal, penalty factors and slack variables are cross validated and optimized. Then the trained model is used to predict the urge to defecate. Prediction accuracy of the feed-forward neural network and SVM with different kernel functions is compared. Experiment results show that the proposed method is effective to rebuild patients＇ rectal perception function.