
 
Gopu G., Porkumaran K., Neelaveni R. Investigation of Digestive System Disorders with Cutaneous Electrogastrogram (EGG) Signal  An Engineering Approach // European Journal of Scientific Research. 2011. Vol.53. №.2. pp.210221. Популярно о болезнях ЖКТ читайте в разделе "Пациентам"

Авторы: Gopu G. / PorkumaranK. / Neelaveni R. 
Investigation of Digestive System Disorders with Cutaneous Electrogastrogram (EGG) Signal  An Engineering ApproachGopu G. Department of EEE, PSG College of Technology, Coimbatore04, Tamilnadu, INDIA Email: gopugovindasamy@yahoo.com Tel: +919894422311; Fax: +914222461089 Porkumaran K. Department of EEE, Dr.NGP Institute of Technology,Coimbatore,Tamilnadu, INDIA Email: porkumaran@ieee.org Tel: +919894101804; Fax: +914222629368 Neelaveni R. Department of EEE, PSG College of Technology, Coimbatore04, Tamilnadu, INDIA Email: rnv@eee.psgtech.ac.in Tel: +919894365195; Fax: + 91 422 2573833
1. Introduction Electrogastrogram is a noninvasive method of assessing the gastric myoelectrical activity of the stomach [1]. The rapid development in Electrogastrography in the past few years is reflected in the attention among investigators and physician. Currently more research on this interesting field was triggered among the research scholars, investigators, physician, etc. due to disseminated digestive system disorders globally because of fast and junk food habit. The acquisition of gastric activity of the stomach is obtained cutaneously from the human subjects. Unlike other electrophysiological methods, visual analysis is unlikely to provide the investigator with meaningful and accurate information. Therefore, a major breakthrough was noted in the late 1970s after the introduction of computerized frequency analysis, which provided investigators a better and more accurate way of analyzing data [2, 3]. Computerized analysis of EGG with advanced spectral analysis provided the subsequent extraction of more accurate information and clinical experience accumulated in gastrointestinal motility. In this paper the EGG is analysed using three methodologies namely Fast Fourier Transform, Wavelet Transform and Neural Network. The normal subject frequency is found to be in the range 0.048 Hz < f <0.051 Hz and for the abnormal subject the frequency is found to be f< 0.048 Hz and f >0.051 Hz from FFT analysis. The normal subjects have the percentage of error value at 22.21% whereas the abnormal subject have the percentage of error value at 15.45% for frequency below 0.048 Hz and the percentage of error value range from 25% to 41% for frequency above 0.051 Hz depending on disorders from the wavelet analysis. Back Propagation Network [BPN] training algorithm namely trainrp, traincgf and trainoss satisfies the condition but trainrp is more suitable for the classifications of normal subjects and abnormal subjects in the Neural Network analysis. 2. Proposed Electrogastrogram Recording Setup Electrogastrogram recording setup is shown in the Figure 1. The bio signal from the stomach due to motility is tapped with active electrodes or Ag/AgCl surface electrodes cutaneously [4, 5, 6, 17]. The electrodes output is given as an input to the Signal Conditioning Unit [SCU] which consists of Instrumentation amplifier, Band pass filter, Notch filter and Gain control. SCU includes amplification, filtering, converting, range matching, isolation and any other processes required to make sensor output suitable for further processing. In SCU, an instrumentation amplifier is used to amplify the potential detected by the electrodes. An amplifier accepts a voltage signal as an input and produces a linearly scaled version of this signal at the output. It is a closedloop fixedgain amplifier, usually differential, and has high input impedance, low drift and high commonmode rejection ratio over a wide range of frequencies. A bandpass filter is a device that passes frequencies within a certain range and rejects frequencies outside that range. Notch filter is known as bandcut filter or bandreject filter. The function of this filter is to remove some frequency portion of a signal. It is used to reduce or prevent feedback. Gain control is an adaptive system found in many electronic devices. The average output signal level is fed back to adjust the gain to an appropriate level for a range of input signal levels. Signal conditioning unit primarily utilized for data acquisition, in which sensor signals must normalized, filtered to levels suitable for analogtodigital conversion to recording and analyzing using computer processor. A Datascope is used for capturing and analyzing EGG signals. It is an 8 channel data acquisition system which amplifies the data and converts into digital format, which would be input to the personal computer through a serial port (RS232).The personal computer acts as a monitoring, analyzing and display device. The electrode position and real time EGG acquisition is shown in Figure 2. Figure 1: Electrogastrogram recording setupFigure 2: Real Time Electrogastrogram recording process3. Acquisition and Analysis of Electrogastrogram The EGG is recorded for about more than hundred subjects, which includes both normal and abnormal subjects [7, 8]. Details of subjects recorded under each abnormality are listed in table 1. Table 1: Sex and Age Distribution of Subjects
The acquired EGG data [10, 11, 12] are analyzed in three phases such as FFT analysis, Wavelet analysis, and neural network analysis. It is compared with normal subjects’ benchmark database for the classification normal subject and subjects with disorders. The results of each analysis is compared individually or combining of all method of analysis to direct the physician to detect the disorders noninvasively as shown in Figure 3. Figure 3: Block diagram for Electrogastrogram Analysis3.1. Analysis of Electrogastrogram by FFT The EGG is analyzed using Fast Fourier Transform and is classified as normal subjects and abnormal subjects based on their signal frequency [13]. The normal EGG signal has a frequency range of about 0.05 Hz (3cpm) and the frequency range below or above this value is considered as abnormal. The algorithm for EGG analysis using FFT follows:
The FFT analysis output is shown in Figure 4. From the output of the FFT it is clearly observed that the frequency for normal subject is around 0.04688Hz and for abnormal subject (Nausea) is around 0.07813Hz. Figure 4: Electrogastrogram Analysis by FFT. A. FFT output for Normal subjects B. FFT output for abnormal subjects. Frequency in Xaxis and Power in Yaxis3.2. Analysis of Electrogastrogram by Wavelet Transform The EGG is analyzed using Discrete Wavelet Transform [15, 16, 18], which has two stage processes namely Decomposition and Reconstruction. Although there are many types of transform in wavelet families, daubechies (db4) is used since it is the best type for EGG signal analysis. The normal and abnormal signals are classified based on their error values obtained after analysis. The algorithm for EGG analysis using wavelet analysis follows:
The db4 wavelet analysis output is shown in Figure 5. From the output of the wavelet analysis it is the normal subjects have the percentage of error value at 22.21% whereas the abnormal subjects have the percentage of error value at 37.12%. Figure 5: Electrogastrogram Analysis by Wavelet. A. db4 wavelet output for Normal subjects B. db4 wavelet output for abnormal subjects3.3. Analysis of Electrogastrogram by Neural Network The EGG signal is trained by feed forward network [9, 14] using nine different BPN algorithms. Among these three algorithms namely Resilient back propagation (RP), FletcherPowell Conjugate Gradient (CGF), OneStep Secant (OSS) gives best result for EGG application by satisfying the following condition 1). If epoch increases, Mean Square Error [MSE] value decreases i.e. Epoch is inversely proportional to MSE, 2). MSE values decreases when epochs increase for normal subjects as shown in table 2. Table 2: Epoch and MSE Values for Different Algorithms
MSE value is considered as a performance of training algorithms for classifying the EGG signals according to the abnormalities. The algorithm for EGG classification using Neural Network is follows:
Normal and Abnormal subjects were trained using nine algorithms and performance for each algorithm calculated as a function of MSE and the same was plotted as a comparison of response that is shown in Figure 6. From the chart it clear that the following algorithms RP, CGF and OSS satisfies the condition and more suit to classify the disorders among the various subjects when compare to other algorithms. Figure 6: Performance of different BPN training algorithms4. Results Electrogastrogram is a noninvasive and inexpensive method of diagnosing the gastric disorders. About 120 subjects EGG of different age group is taken for analysis. Figure 7: Result of FFT AnalysisAnalysis of EGG signal using FFT is shown in Figure 7. From the above graph, we can observe the frequency of normal signal approximately 0.05Hz. EGG signal have frequency greater than or lesser than 0.05Hz is said to be the abnormal signal. Figure 8: Classification using WaveletFigure 9: Error comparison for Wavelet AnalysisThe EGG classification using wavelet analysis is shown in Figure 8.From the figure, it is found that the percentage error of normal signals lies from 20 to 22. For the abnormal signals percentage error may be lesser than or greater than normal signal range. The error comparison is shown in Figure 9. Figure 10: Electrogastrogram Analysis by Neural Network. A. Performance of RP Algorithm, B. Performance of CGF Algorithm, C. Performance of OSS Algorithm for the classification subjectsThe performance RP, CGF and OSS algorithm for EGG signal analysis are shown in Figure 10 as A, B and C respectively. By comparing the results, we can conclude MSE values and Epochs are inversely proportional. MSE values decreases when epochs increase for normal subjects. However, for abnormal subjects the percentage variation of MSE values with respect to epochs is less compared to normal subjects. The performance comparison of different algorithm is shown in table 3. Table 3: Performance Comparison of Different Algorithm
5. Conclusion In this paper, we introduced a novel approach for analysis and classification of EGG signals. Here three different engineering methods are used for the analysis of EGG signals. They are FFT analysis, Wavelet analysis and Neural network analysis. FFT classify the signals accurately based on the frequency. Wavelet uses the percentage of error for classifications and neural network classify the data according to their % MSE and epochs. By comparing the results of these three methods, neural network classify the EGG signal of normal and abnormal subject at fair amount of accuracy. In neural network analysis, BPN network is used. This network is trained and analyzed using nine different training algorithms. By comparing the result of all algorithm, three algorithm namely Resilient back propagation (RP), FletcherPowell Conjugate Gradient (CGF), OneStep Secant (OSS) performance classify the normal and abnormal subjects in best way. Among these three algorithm ,RP algorithm is more suitable for the classification with an accuracy of 93.33 percentage compared to other algorithms and it is the best choice for the analysis and classification EGG signal with the help of BPN network. From this, it is concluded that the method adopted here is able to classify the signal accurately with less time and the physician is directed for the investigation of digestive disorders noninvasively. Acknowledgements The authors acknowledge their indebtedness to the following medical experts Dr T S Chandrasekar, Gastroenterologist, MedIndia Hospitals, Coimbatore, Dr L Venkatakrishnan, M.D., D.M.,D.N.B., Head of Gastroenterology Dept., Dr.J.Krishnaveni, M.D.,D.N.B., Gastroenterologist from PSG Hospitals, Coimbatore, and Dr M G Shekar, M.S., D.N.B., M.R.C.S., Laparoscopic surgeon Stanley Medical College and Hospital, Chennai for their support and for permitting us to use the facilities at the hospitals for live testing of the recording setup and sharing valuable patient database with us. References [I] W.C.Alvarez, 1922. “The Electrogastrogram and what it shows”, JAMA 78, pp.1116–1118. [2] C.F Code and J.A Marlett, 1974. “Modern Medical Physiology: Canine Tachygastria”, Mayo clinic proceeding 49, pp.325332. [3] R.L.Telander et al, 1978. “Human gastric atony with tachygastria and gastric retention”, Gastroenterology 75, pp.495501. [4] A.J.P.M. Smout, E.J. Van Der Schee, and J.L.Grashuis, 1980. “What is measured in electrogastrography?”, Digestive Diseases and Sciences, pp. 253. [5] T.L.Abell, J.R.Malagelada, 1988. “Electrogastrography: Current assessment and future perspectives”, Digestive Diseases and Sciences 33, pp.982–992. [6] J. Chen, R.W. McCallum, 1991. “Electrogastrography: measurement, analysis and prospective applications”, Medical &Biological Engineering & Computing 29, pp.339–350. [7] G.Riezzo, F.Pezzolla, J. Thouvenot, et al, 1992. “Reproducibility of cutaneous recordings of electrogasography in the fasting state in man”, Pathology Biology 40, pp.889–894.
[8] M.P. Mintchev, Y.J. Kingma, K.L. Bowes, 1993. “Accuracy of coetaneous recordings of gastric electrical activity”,Gastroenterology 104, pp.1273–1280.
[9] Laurene Fausett, 1994. “Fundamentals of Neural Networks”, Prentice Hall.
[10] B.Pfaffenban, R.J.Adamek, K.kuhn, et.al, 1995. “Electrogastrography in healthy subjects. Evaluation of normal values influences of age and gender”, Digestive Diseases and Sciences 40, pp.445450. [11] J.D.Z.Chen, Z.Lin, I.Pan, et al, 1996. “Abnormal gastric myoelectrical activity and delayed gastric emptying in patients with symptoms suggestive of gastro paresis”. Digestive Diseases and Sciences 41, pp.15381545. [12] J.D.Z.Chen, 1998. “Noninvasive Measurement of gastric Myoelectrical Activity and its Analysis and Applications”, Annual International Conference of the IEEE Engineering in Medicine and Biology Society 20(6), pp.28022807. [13] A.Leahy, K.Besherdas, C.Clayman, I. Mason, O.Epstein, 1994. “Abnormalities of the Electrogastrogram in functional dyspepsia”. American Journal of Gastroenterology 94(4), pp.1023–1028.
[14] L.OhnoMachado, 1996. “Medical Applications of Artificial Neural Networks”, Connectionist Model of Survival. Ph.D Dissertation, Stanford University.
[15] HanChang Wu, KuangChing, et al, 1998. “Power distribution analysis of Cutaneous Electrogastrography using discrete wavelet transform”, International Conference of the IEEE engineering in medicine and biology society 20(6), pp.32273229.
[16] Stephen Mallat, 1998. “A wavelet Tour of Signal Processing”, Second Edition, Academic Press, New York. [17] D.Z.Chen, Zhiyue Lin, 2006. “ElectrogastrogramEncyclopedia of Medical Devices and Instrumentation”, Second Edition, edited by John G. Webster, John Wiley & Sons. [18] K.P.Soman, K.I. Ramachandran, 2000. “Insight to wavelets from Theory to Practice”, Prentice Hall of India .Pvt.Ltd, New Delhi. 
