多路脉象信号的特征提取与模式分类
摘要
中医脉诊学是传统医学的瑰宝,公元前就有人们通过脉搏波指导诊断各种疾病的记载、脉诊学有合理的内涵和丰富的经验,脉诊的客观化研究就是要利用先进的科学技术和手段来传统脉诊学和现代技术结合起来。
本文主要研究了基于多点复合式压力传感器的多路脉象信号中特征提取和模式分类的一些技术。这些信号包括寸、关、尺三路主信号以及关部的七路副信号。
在预处理阶段我们通过试验分析了近二十种小波基的性能,最终选择DMeyer小波进行小波包阈值消噪,消噪效果较以往的sym6小波消噪有所改进。我们还提出了LIP(Lowest point In one Period)算法用以改进基于周期起点插值的去基线漂移算法在多路信号应用中对噪声的鲁棒性。由于多路脉象信号具有一定的信息相关性,我们在三路主信号上利用ICA变换后去除噪声信号在反变换的办法恢复了一些被噪声湮没的脉搏波信号。
在特征提取阶段由于基于差分的时域特征提取算法在应用于多路信号时该算法的准确性受噪声影响比较严重,我们提出了基于ILP(Intersection points of Lines and Pulse waveform)的时域特征提取算法,耐噪声能力得到了很大提高。为了定义脉搏波的宽度特征,我们用高斯函数拟合七个副探头的平均幅值然后用拟合得到的高斯函数的delta来代表脉搏波的宽度。最后我们还提取了四个新的面积特征,十维频域特征以及脉搏波的强度特征。
在模式分类阶段我们尝试了支持向量机,神经网络以及基于知识的神经网络算法,我们还对不同种类的特征进行了联合特征的分类实验,结果表明在用神经网络分类时联合特征可以有效的提高识别精度。在基于知识的神经网络应用中由于我们利用了领域知识来改进性能使其络获得了这些分类方法之中最高的识别精度。
Pulse diagnosis is the most distinctive one among the traditional Chinese medicine. Arterial pulse was used as a guide to diagnose and treat different diseases so it is necessary to apply the advantage of modern science and technology to pulse diagnosis.
This thesis is mainly worked on the feature extraction and pattern classification technique of the pulse signal gathered by multiple probes. In the preprocessing a lot of wavelet function was tested and at last DMeyer wavelet packet threshold de-noising method is used to improve the de-noising result. Then we proposed a LIP (Lowest point In one Period) algorithm to improve the baseline wander removal algorithm. We also recovered some signals from heavy noise by ICA transform.
In the feature extraction, as the algorithm based on grads got a lot of problems on the Cun and Chi, we propose a new algorithm based on the ILP (Intersection points of Lines and Pulse waveform) to extract the time domain features from the pulse waveform, Gaussian function was used to fit the mean amplitude of the sub sensors and the delta of the Gaussian function was selected to stand for the width of the pulse. We also extract some other new features including area features, frequency domain features and strength features.
To classify these signals we tried the SVM algorithm, ANN algorithm and KBANN algorithm. We also did some experiment on joint feature classification. The result shows that the joint feature using ANN can be more accuracy. However as we can add some domain knowledge, the KBANN algorithm gets the best result.
本文主要研究了基于多点复合式压力传感器的多路脉象信号中特征提取和模式分类的一些技术。这些信号包括寸、关、尺三路主信号以及关部的七路副信号。
在预处理阶段我们通过试验分析了近二十种小波基的性能,最终选择DMeyer小波进行小波包阈值消噪,消噪效果较以往的sym6小波消噪有所改进。我们还提出了LIP(Lowest point In one Period)算法用以改进基于周期起点插值的去基线漂移算法在多路信号应用中对噪声的鲁棒性。由于多路脉象信号具有一定的信息相关性,我们在三路主信号上利用ICA变换后去除噪声信号在反变换的办法恢复了一些被噪声湮没的脉搏波信号。
在特征提取阶段由于基于差分的时域特征提取算法在应用于多路信号时该算法的准确性受噪声影响比较严重,我们提出了基于ILP(Intersection points of Lines and Pulse waveform)的时域特征提取算法,耐噪声能力得到了很大提高。为了定义脉搏波的宽度特征,我们用高斯函数拟合七个副探头的平均幅值然后用拟合得到的高斯函数的delta来代表脉搏波的宽度。最后我们还提取了四个新的面积特征,十维频域特征以及脉搏波的强度特征。
在模式分类阶段我们尝试了支持向量机,神经网络以及基于知识的神经网络算法,我们还对不同种类的特征进行了联合特征的分类实验,结果表明在用神经网络分类时联合特征可以有效的提高识别精度。在基于知识的神经网络应用中由于我们利用了领域知识来改进性能使其络获得了这些分类方法之中最高的识别精度。
Pulse diagnosis is the most distinctive one among the traditional Chinese medicine. Arterial pulse was used as a guide to diagnose and treat different diseases so it is necessary to apply the advantage of modern science and technology to pulse diagnosis.
This thesis is mainly worked on the feature extraction and pattern classification technique of the pulse signal gathered by multiple probes. In the preprocessing a lot of wavelet function was tested and at last DMeyer wavelet packet threshold de-noising method is used to improve the de-noising result. Then we proposed a LIP (Lowest point In one Period) algorithm to improve the baseline wander removal algorithm. We also recovered some signals from heavy noise by ICA transform.
In the feature extraction, as the algorithm based on grads got a lot of problems on the Cun and Chi, we propose a new algorithm based on the ILP (Intersection points of Lines and Pulse waveform) to extract the time domain features from the pulse waveform, Gaussian function was used to fit the mean amplitude of the sub sensors and the delta of the Gaussian function was selected to stand for the width of the pulse. We also extract some other new features including area features, frequency domain features and strength features.
To classify these signals we tried the SVM algorithm, ANN algorithm and KBANN algorithm. We also did some experiment on joint feature classification. The result shows that the joint feature using ANN can be more accuracy. However as we can add some domain knowledge, the KBANN algorithm gets the best result.
引文
1黄世林,孙明异.中医脉象研究.人民卫生出版社,1986:1-6
2李时珍.瀕湖脉学白话解.人民卫生出版社,1978:11-62
3赵恩俭.中医脉诊学.天津科学技术出版社,1990:1-18
4费兆馥,张志枫.中医脉诊的图像化和定量化.自然杂志,1995,17(5):269-274
5王炳和,王海燕,职利琴等.高血压患者脉象信号的采集与频谱分析.声学技术,1998,18(3):135-138
6孟兆辉,白净,王苏中等.高血压病人的光电容积脉搏波的频域分析.北京生物医学工程,2002,21(1):1-4
7 K.Q. Wang, L. Wang, L.S. Xu, D.H. Wang. SVM classification for discriminating cardiovascular disease patients form non-cardiovascular disease controls using pulse waveform variability analysis. 17th Australian Computer Society (ACS) Australian Joint Conference on Artificial Intelligence,2004:109-119
8岳沛平.脉象与脉图在心血管疾病临床诊断中的应用.中医文献杂志,2006(1):31-33
9沈海东.脉象信号与心血管状态关系的研究.上海交通大学硕士论文.2007
10 W. A. Lu, Y. Y. Lin Wang, W. K. Wang. Pulse Analysis of Patients with Severe Liver Problems– Studying Pulse Spectrums to Determine the Effects on other Organs. IEEE Engineering in Medicine and Biology. 1999,18:73~75
11罗志昌.脉搏波的工程分析与临床应用.科学出版社,2006
12黄世林,孙明异.中医脉象研究.人民卫生出版社,1986:1-6
13 Wang B.H., Xiang J.L. Yang Y. Evaluation of Transfer Function of human pulse system bases on signal detection. Chinese Science Bulletin,1999,44(17):1566-1571
14林钦裕.中医脉诊脉波信号撷取系统之研究.中医药年报,1999,17(3):1301-1312
15 S.H.Yoon, Y.Koga, I. Matsumoto and E. Ikezono. An objective method of pulse diagnosis. American Journal of Chinese medicine.1987,15(3-4):147-153
16 B. Michael, M. Michael. Instrument assisted pulse evaluation in the acupuncture practice. American Journal of Acupuncture, 1986,14(3):255-259
17 Y.Z. Yoon, M.H. Lee, K.S.Soh. Pulse type classification by varying contact pressure. IEEE Engineering in Medicine and Biology,2000(19):106-110
18 S.He.etal. Objectifying of pulse taking. Journal of Japanese Eastern Medicine Society.1977,27(4):7
19 W.K. Wang, H.L. Chen, T.L. Hsu, et al. Alternations in human subjects by three Chinese herbs. American Journal of Chinese Medicine.1994,22(2):97
20 J.L. Jin, J. Min-suk. Development of a Radial Artery pulse wave transducer for diagnostic of human body constitution. Proceedings of the 32nd international symposium on robotics, Seoul, Korea, April 2001:19-21
21 X.F. Teng, Y.T. Zhang. Continuous and noninvasive estimation of arterial blood pressure using a photoplethysmographic approach. Proceedings of the 25th Annual International Conference of the IEEE EMBS, Cancun, Mexico, September 17-21,2003:3153-3156
22姬军,俞梦孙等.基于无线传感器网络的脉搏波传导时间监测系统.航天医学与医学工程,2006-06
23赵同光.微机在脉象测量分析中的应用.鞍山钢铁学院学报,1995,18(2):27-32
24 BH Wang, JL Xiang. Detecting system and power spectral analysis of pulse signals of human body. Proceedings of IEEE international conference on signalprocessing, Oct 1998:1646-1649
25 QingLi Guo, KuanQuan Wang, DongYu Zhang, NaiMin li. A wavelet packet based pulse waveform analysis for cholecystitis and nephritic syndrome diagnosis. International Conference on Wavelet Analysis and Pattern Recognition, August 30-31,Hong Kong, 2008
26徐礼胜.中医脉象数字特征研究.哈尔滨工业大学博士论文,2006:42-46
27 KARIM G O, DAVID J A. Noise reduction in multichannel neural recordings using a new array wavelet denoising algorithm. Neurocomputing. 2001, 38(4):1687~1693
28 THOMAS T L, ANTONY C S. Detection of transients in 1/f noise with the undecimated discrete wavelet transform. IEEE Transactions on Signal Processing. 2000, 48(5):1458~1462
29郭晶,孙伟娟.小波分析理论与MATLAB7实现.电子工业出版社,2005:110~117
30郑建国,石智,权豫西.非平稳信号的小波包阈值去噪方法.信息技术. 2007,(3):16-18
31 Donoho.D.L. Denoising by soft thresholding. IEEE Transaction on information Theory. 1995, 411(3):613~627
32 Donoho D.L. Adapting to unknown smoothness via wavelet shrinkage. J Amer starist assoc. 1995,90(432):613~627
33曾法力.小波包分析在齿轮故障诊断中的应用.武汉科技大学硕士学位论文. 2005:30~31
34 Cichocki A , Amari S. Adaptive Blind Signal and Image Processing. Wiley , 2003
35 Cardoso J F. Blind Signal Separation : Statistical Principles. IEEE Proceedings , 1998 , 86(10) :2009 - 2025.
36杨福生,洪波.独立分量分析的原理与应用.清华大学出版社2006
37张洪渊,贾鹏,史习智.确定盲分离中未知信号源个数的奇异值分解法.上海交通大学学报,2001,35(8):1155-1158
38陈锡明,黄硕翼.盲源分离综述——问题、原理和方法.电子信息对抗技术第23卷.2008年3月第2期
39张治国,牛欣等.脉宽的可视化考察. Chinese Journa1 of Informationon TCM,2008:5-14
40岳全中,朱永利.基于支持向量机多分类器的电流互感器状态评估,中国高等学校电力系统及其自动化专业第二十四届学术年会论文集(上册),2008:117-120
41 Tom M. Mitchell.著,曾华军等译. MACHINE LEARNING,机械工业出版社,2008:240:247
2李时珍.瀕湖脉学白话解.人民卫生出版社,1978:11-62
3赵恩俭.中医脉诊学.天津科学技术出版社,1990:1-18
4费兆馥,张志枫.中医脉诊的图像化和定量化.自然杂志,1995,17(5):269-274
5王炳和,王海燕,职利琴等.高血压患者脉象信号的采集与频谱分析.声学技术,1998,18(3):135-138
6孟兆辉,白净,王苏中等.高血压病人的光电容积脉搏波的频域分析.北京生物医学工程,2002,21(1):1-4
7 K.Q. Wang, L. Wang, L.S. Xu, D.H. Wang. SVM classification for discriminating cardiovascular disease patients form non-cardiovascular disease controls using pulse waveform variability analysis. 17th Australian Computer Society (ACS) Australian Joint Conference on Artificial Intelligence,2004:109-119
8岳沛平.脉象与脉图在心血管疾病临床诊断中的应用.中医文献杂志,2006(1):31-33
9沈海东.脉象信号与心血管状态关系的研究.上海交通大学硕士论文.2007
10 W. A. Lu, Y. Y. Lin Wang, W. K. Wang. Pulse Analysis of Patients with Severe Liver Problems– Studying Pulse Spectrums to Determine the Effects on other Organs. IEEE Engineering in Medicine and Biology. 1999,18:73~75
11罗志昌.脉搏波的工程分析与临床应用.科学出版社,2006
12黄世林,孙明异.中医脉象研究.人民卫生出版社,1986:1-6
13 Wang B.H., Xiang J.L. Yang Y. Evaluation of Transfer Function of human pulse system bases on signal detection. Chinese Science Bulletin,1999,44(17):1566-1571
14林钦裕.中医脉诊脉波信号撷取系统之研究.中医药年报,1999,17(3):1301-1312
15 S.H.Yoon, Y.Koga, I. Matsumoto and E. Ikezono. An objective method of pulse diagnosis. American Journal of Chinese medicine.1987,15(3-4):147-153
16 B. Michael, M. Michael. Instrument assisted pulse evaluation in the acupuncture practice. American Journal of Acupuncture, 1986,14(3):255-259
17 Y.Z. Yoon, M.H. Lee, K.S.Soh. Pulse type classification by varying contact pressure. IEEE Engineering in Medicine and Biology,2000(19):106-110
18 S.He.etal. Objectifying of pulse taking. Journal of Japanese Eastern Medicine Society.1977,27(4):7
19 W.K. Wang, H.L. Chen, T.L. Hsu, et al. Alternations in human subjects by three Chinese herbs. American Journal of Chinese Medicine.1994,22(2):97
20 J.L. Jin, J. Min-suk. Development of a Radial Artery pulse wave transducer for diagnostic of human body constitution. Proceedings of the 32nd international symposium on robotics, Seoul, Korea, April 2001:19-21
21 X.F. Teng, Y.T. Zhang. Continuous and noninvasive estimation of arterial blood pressure using a photoplethysmographic approach. Proceedings of the 25th Annual International Conference of the IEEE EMBS, Cancun, Mexico, September 17-21,2003:3153-3156
22姬军,俞梦孙等.基于无线传感器网络的脉搏波传导时间监测系统.航天医学与医学工程,2006-06
23赵同光.微机在脉象测量分析中的应用.鞍山钢铁学院学报,1995,18(2):27-32
24 BH Wang, JL Xiang. Detecting system and power spectral analysis of pulse signals of human body. Proceedings of IEEE international conference on signalprocessing, Oct 1998:1646-1649
25 QingLi Guo, KuanQuan Wang, DongYu Zhang, NaiMin li. A wavelet packet based pulse waveform analysis for cholecystitis and nephritic syndrome diagnosis. International Conference on Wavelet Analysis and Pattern Recognition, August 30-31,Hong Kong, 2008
26徐礼胜.中医脉象数字特征研究.哈尔滨工业大学博士论文,2006:42-46
27 KARIM G O, DAVID J A. Noise reduction in multichannel neural recordings using a new array wavelet denoising algorithm. Neurocomputing. 2001, 38(4):1687~1693
28 THOMAS T L, ANTONY C S. Detection of transients in 1/f noise with the undecimated discrete wavelet transform. IEEE Transactions on Signal Processing. 2000, 48(5):1458~1462
29郭晶,孙伟娟.小波分析理论与MATLAB7实现.电子工业出版社,2005:110~117
30郑建国,石智,权豫西.非平稳信号的小波包阈值去噪方法.信息技术. 2007,(3):16-18
31 Donoho.D.L. Denoising by soft thresholding. IEEE Transaction on information Theory. 1995, 411(3):613~627
32 Donoho D.L. Adapting to unknown smoothness via wavelet shrinkage. J Amer starist assoc. 1995,90(432):613~627
33曾法力.小波包分析在齿轮故障诊断中的应用.武汉科技大学硕士学位论文. 2005:30~31
34 Cichocki A , Amari S. Adaptive Blind Signal and Image Processing. Wiley , 2003
35 Cardoso J F. Blind Signal Separation : Statistical Principles. IEEE Proceedings , 1998 , 86(10) :2009 - 2025.
36杨福生,洪波.独立分量分析的原理与应用.清华大学出版社2006
37张洪渊,贾鹏,史习智.确定盲分离中未知信号源个数的奇异值分解法.上海交通大学学报,2001,35(8):1155-1158
38陈锡明,黄硕翼.盲源分离综述——问题、原理和方法.电子信息对抗技术第23卷.2008年3月第2期
39张治国,牛欣等.脉宽的可视化考察. Chinese Journa1 of Informationon TCM,2008:5-14
40岳全中,朱永利.基于支持向量机多分类器的电流互感器状态评估,中国高等学校电力系统及其自动化专业第二十四届学术年会论文集(上册),2008:117-120
41 Tom M. Mitchell.著,曾华军等译. MACHINE LEARNING,机械工业出版社,2008:240:247