Detection Of Gastrointestinal Multiple Medical Signals And Research On Methods Of Signal Feature Extraction

The study of human gastrointestinal tract(GIT)is continuously meet human needs,and it can also help us to improve the way of life.With the increasing pressure of life and work,and diversification of diet,the incidence of gastrointestinal illness gradually rising in the world,and 30 ~ 40% were diagnosed with gastrointestinal motility disorder(GIMD)diseases.Although the diseases are not fatal,it will bring inconvenience and pain to the patients.Traditional intubation methods can detect the pressure or pH of the GIT,but these methods can bring discomfort to patients,and cannot realize the detection of small intestine.As the progress of science and technology and the requirement of improving the quality of GIMD patients life,combined with the micro sensors and radio frequency(RF)communication technology,detection of gastrointestinal multiple medical signal and data analysis of GIT have become an important content of medical engineering field.Under the supports of the National Natural Science Fund of China(No.31170968)and the Research Fund of Shanghai Committee of Science and Technology(No.14441902800,No.15441903100),this paper conducts the study and implementation on the multiple medical signal detection of GIT and the feature extraction methods of the obtained signals.Combined with the hardware design,human experiments,and algorithm theory,the design work of multiple medical signal detection system,and medical signal analysis were accomplished.After the human experiments,the function of detection system was verified and the feature extraction methods were studied with multiple medical signals.To sum up,the work of this paper including the following aspects:1.Based on the purpose of this study and the technical methods,combined with the design of hardware module unit and software control program,a multiple medical signal detection system was developed.Specifically,the components of the detection system include a low-power wireless capsule,a portable data recorder,a wireless capsule positioning device,and a workstation.Sensor modules(pH,pressure,and temperature sensors);a low-power consumption application specific integrated circuit(ASIC);a RF communication module and spiral antenna;batteries module;and shell were applied in a low-power wireless capsule.The programming inside a wireless capsule including sensor signal setting,zero adjustment,and programmable gain amplifier(PGA),A/D conversion(ADC),and the bidirectional communication control.Low-power wireless capsule that can complete gastrointestinal multiple medical signal acquisition and wireless data transmission is the core part of system.The portable data recorder that can be used for data receiving and saving is mainly composed of micro controller unit(MCU),RF communication module and antenna,multimedia memory card(MMC),and battery modules.In view of the wireless capsule positioning in colon,a wireless capsule positioning device was developed.This paper proposes an analysis method based on the multi-point magnetic field detection,and verifies the feasibility of this method.The workstation can be used to analyze the data achived by human experiments.Data reading,preprocessing,and feature selection,feature evaluation,design of the classifier processing,and corresponding conclusions were accomplished in the workstation.The workstation can intuitively show the characteristics of the data and is the foundation platform of medical signal analysis.2.As an organic whole,the statistical characteristics of the gastrointestinal multiple medical signals(temperature,pH,and pressure)can reflect the overall situation of human gastrointestinal motility.Based on the data obtained by human experiments using multiple medical signal detection system,the feature extraction and cluster analysis of gastrointestinal medical signals were studied.Firstly,in order to limit the impact of singular value data,add missing data,and filter out the high frequency components,preprocessing was accomplished.Second,extracting the statistical characteristics of the medical signal,combining with the intra-class dispersion matrix and inter-class dispersion matrix,the optimal characteristics can be selected.Third,weighted fuzzy c-means(WFCM)clustering analysis was used to classify the gastrointestinal motility of subjects.Using the 20 cases of human gastrointestinal signals,76 original features that include the 20 interval distribution characteristics of the temperature,pH,and pressure values,the standard deviation,variance,maximum,and minimum values of the temperature,pH and pressure values,the entire gastrointestinal emptying time were extracted.Based on the 13 optimal features achieved in the evaluation of the features,all samples can be divided into 3 groups and the WFCM clustering algorithm can identify the patients who lack GI motility with a recognition rate reaching 83.3%.3.Gastric motility disorder is a common disease,clinical manifestations include gastroparesis,indigestion,etc.Gastric motility that indicates the internal pressure changes and contraction activity of stomach is complex and nonlinear.Analyzing the movement function of the human stomach is the basis of disease diagnosis.The gastric internal pressure changes are caused by a variety of factors that contain the stomach contractions,relevant muscle movements,breathing exercises,noise,etc.By using a joint algorithm includes phase space reconstruction(PSR)and fast independent component analysis(FastICA),the problem of pressure signal separation can be solved,and the pressure change of gastric contraction peristalsis was separated effectively.Coupled with short time Fourier transform,the peak frequency of seperated signal that can be apply to evaluate the gastric motility of subjects and provide support for clinical diagnosis was calculated.The 24 cases of human experiments were found to have either normal gastricmotility(75.0%),bradygastria(20.8%),or tachygastria(4.2%).During the digestive period,the center frequency of the DMC was 3 cpm.Meanwhile,the lowest and highest frequencies reached 2 cpm and 4 cpm,respectively.Interestingly,a harmonic wave with 6 cpm was also observed.During the interdigestive period,IDMC had 3 phases:(I)the stationary phase,during which the frequency was maintained steadily at 3 cpm;(II)several discontinuous irregular contractions appeared,the frequency became unstable,and the center frequency started falling,a phase that lasted about 32 minutes;and(III)a series of strong contractions emerged,and the frequency spectrum was more confused,a phase that lasted 18 minutes.Additionally,empirical mode decomposition(EMD)time-frequency method was used for anlyzing the gastric motility.After the EMD decomposition and Hilbert transform,the Hilbert spectrum showed that the IMF2~IMF4 of intrinsic mode functions(IMFs)can indicate the chara-cteristics of gastric motility,and it can help us to further understand the gastric motility.4.Colonic motility disorder is a common clinical symptom,but the understanding of it is still very limited.The colonic pressure is a time-varying and non-stationary signal,and it is relatively complex.In addition,the individual difference of colonic pressure signal is large.The features extracted for the classification were the time domain,frequency domain,chaos-fractal,wave number,motility index,area under curve,and some others.For the feature selection methods,the Relief algorithm and the support vector machine(SVM)-based recursive feature elimination(SVM-RFE)were adopted.Specially,the SVM-optimized by the genetic algorithm(GA)or particle swarm optimizer(PSO)-was applied for the classification of healthy subjects and slow transit constipation(STC)patients.Using the Relief and SVM-RFE algorithms,high-performing classification features can be selected effectively.For the subjects with constipation and healthy individuals,the selected and distinguishing features were probability distribution(F22,F27),mean magnitude at peak frequency of d4(F37),number of wave peaks per hour(F40),areas under the curve(F42),and GI transit time(F43).The test accuracy calculated by GA-SVM was 88.0%,and it was relatively higher compared with that of the PSO-SVM.Additionally,combined with the EMD decomposition technique and different entropy measurement method,the complexity of colonic pressure signal can be measured.The time entropy(TE),power spectrum entropy(PSE),and approximate entropy(AE)were used for quantitative analysis of the complexity of colonic motility.Specifically,the results of clinical trial showed that the average value of TE,PSE and AE of STC patients were significantly lower than the healthy subjects.Compared with healthy subjects,the standard deviation of TE of STC patients was significantly higher.In this paper,the multiple medical signal detection system,feature extraction,and subjects classification of gastrointestinal medical signals were discussed deeply.This work can provide design experience and a theoretical basis for GI motility analysis.Finally,the content of this dissertation is summarized.With the experience of hardware design and algorithm analysis,the new direction in the future are put forward.