COLE MODEL ANALYSIS OF EBIS NEONATAL CEREBRAL MEASUREMENTS

Transcription

1 COLE MODEL ANALYSIS OF EBIS NEONATAL CEREBRAL MEASUREMENTS PRATHAMESH SHARAD DHANPALWAR & XINYUAN CHEN MASTER DEGREE THESIS 1 ECTS, 9-1 SWEDEN MASTER THESIS N 1/1, BIOMEDICAL ENGINEERING

2 Cole Model Analysis of EBIs Neonatal Cerebral Measurements Prathamesh Sharad Dhanpalwar Xinyuan Chen Master thesis Subject Category: Medical Technology, Signal Processing University College of Borås School of Engineering SE- 1 9 BORÅS Telephone pratham_maverick@yahoo.co.in; Xinyi86-8@hotmail.com Examiner: Supervisor: Fernando Seoane Martínez Fernando Seoane Martínez Date: 7th September 1 Keywords: Electrical Bioimpedance Spectroscopy, Cole Model, Hook Effect, Signal Analysis, neonates. II

3 Dedicated To Parents & Everyone Who Have Made This Possible III

4 ABSTRACT The concept of Electrical Bio Impedance prevails in this thesis. The EBI measurement which is used for obtaining the body composition is, by virtue of time becoming of great use as its one of the easiest method of finding out the body composition. In simple words, EBI is the opposition offered by the body to the current. It is just like another analysis tool. The result is only as good as the test is done. In this thesis, we have done the analysis on the neonatal EBI measurements of two kinds. In this work, 93 measurements are obtained from 1 babies and 3 measurements are obtained from 7 babies have been analyzed with the purpose of obtaining reference values for the spectrum of complex EBI. The analysis uses both statistical and model approach of obtaining reference values and in order to fit the given data, Cole model analysis is used. Filters were applied to get the highest degree of correctness. In the due course of the filtering, it was found that the measurements from some babies have been deleted. The Standard Error of Estimation (S.E.E.) is a parameter used for obtaining the further reliable and most probable output. The further analysis is done using MATLAB and the results are been compared to the previous analysis report. IV

5 ACKNOWLEDGEMENT Thanks to Almighty. The existence of this thesis work would not be possible without the constant support, great effort and unmatchable gesture and encouragement from our supervisor Prof. Fernando Seoane Martinez. We deeply appreciate the assistance and help he provided in accomplishing this thesis. In the due course of this work, we learnt a lot from him and which is priceless. We are very grateful to our parents, siblings and family members who were always there by our side and prayed for our welfare. We would also like to thank our colleagues and friends who always made things easier and simpler for us and made us happy. Prathamesh Sharad Dhanpalwar & Xinyuan Chen 9 th October, Borås V

6 Table of Contents CHAPTER INTRODUCTION INTRODUCTION MOTIVATION GOAL WORK DONE STRUCTURE... CHAPTER... 3 ELECTRICAL BIOIMPEDANCE INTRODUCTION ELECTRICAL PROPERTIES OF TISSUES ELECTRICAL EQUIVALENT OF CELL.... FREQUENCY DEPENDENT EBI Cole equation..... Dispersion in tissue.... EBI MEASUREMENTS Hook effect Origin and its effect Correction and Compensation CEREBRAL MONITORING Cellular damage Causes Need for cerebral monitoring... 9 CHAPTER FLOW OF ANALYSIS EBI SPECTROSCOPY MEASUREMENTS COLE FUNCTION FITTINGS Bioimp Software FILTERING ANALYSIS MATLAB EBI MANIPULATION AND DESCRIPTIVE STATISTICS Data processing in MATLAB MATLAB functions Histogram creation Subject selection Best measurement selection... 1 CHAPTER... 1 RESULTS... 1 VI

7 .1 HISTOGRAMS OF F CHAR FOR EACH SUBJECT Sick neonates with good outcome Sick neonates with no recovery HISTOGRAMS OF THE 6% OF F CHAR Fittings for sick neonates with good outcome Fittings for sick neonates without recovery....3 MINIMUM S.E.E. CRITERION FOR BEST FITTING ANALYSIS REFERENCE VALUES... 7 CHAPTER... 9 DISCUSSION, CONCLUSIONS AND FUTURE WORK DISCUSSION CONCLUSION FUTURE WORK... 9 REFERENCES APPENDIX A... 3 FLOW CHART OF MATLAB WORK... 3 APPENDIX B MATLAB FUNCTIONS VII

8 Chapter 1 INTRODUCTION 1.1 Introduction Bioelectrical Impedance analysis (BIA) is a common method used to assess the tissue and body composition and analyzing them. It is a wide spread and widely accepted technology used for different applications like medical diagnosis, research, patient monitoring and treatment. This method has become popular because of its simplicity and portability. Among the different methods available for EBI spectroscopy, Cole model offers the possibility to represent and experimental measurement with only four parameters. This project focuses on the spectrum of complex EBI of the neonatal brain. 1. Motivation In the past years, research on EBI spectroscopy has been made on neonatal brain in animals and concluded that it changes with hypoxia. This brings out the need to investigate if the EBI changes in asphyxiated brains in humans neonates which in turn opens the doors for development in new tools screening and monitoring neonates suffering from neonatal asphyxia. 1.3 Goal The main goal of this project throws light on obtaining the spectral reference values of EBI and Cole parameters from a set of EBI spectroscopy measurements from both healthy and sick neonates. The secondary goal is also to present a novel way of performing the analysis of EBI values between healthy and sick neonates and correlate them. In this thesis, much work is done on obtaining the reference values which indicate the neonates to the best. 1. Work done EBI spectroscopy data from sick neonates has been analyzed and characterized. The EBI characterization has been done based in the Cole parameters using the Bioimp software MS Excel and MATLAB. Both groups of data have been characterized and the S.E.E of the produced curve fitting is taken to be the quantitative measure of correctness. The set of Cole parameters obtained from classes have been analyzed to look for possible classification features. 1

9 1. Structure The whole thesis report is organized in six chapters and the references. Chapter 1 is the introduction part of the thesis work. Chapter gives a brief background of the electrical properties of the tissues, Cole model, EBI measurements and the bases about how EBI technology can detect tissue injury. Chapter 3 discussed briefly the sector of the population and how the measurements were taken, as well as the equipment used. Furthermore, it describes the method used to realize the study. Chapter contains the results obtained during the analysis. Chapter discusses the final results obtained from the analysis. The last Chapter contains the conclusion and proposed future work.

10 Chapter ELECTRICAL BIOIMPEDANCE.1 Introduction Electrical bioimpedance (EBI) is commonly measured for estimating body and tissue composition. Though the electrical properties of the tissue were discussed from the late 18 th century, the study and advancement in EBI were made in 197s. The analysis is very straight forward. The better the tests are carried out, the better are the results though some care has to be taken while doing the experiments and tests. The opposition offered by the body tissues to the fluids flowing inside the body is called body impedance. Analysis indicates that measurements of body impedance can be used to acquire fluid volumes, for example, total body water (TBW) in the neonate [1].. Electrical properties of tissues The electrical properties of tissue and the cell suspension have been used for a wide range of biomedical applications. Much research work has been done until now and is still undergoing to find out the applications of EBI in various fields. Diagnosis and treatment of various physiological conditions with weak electric currents, radio-frequency hyperthermia, electrocardiography, and body composition are some of the application worked and practiced on till now []. The electrical conductance of the biological tissue depends on its constituents. Many cells together constitute a tissue and each cell has a lipid bi-layer plasma membrane with protoplasm comprising of cytosol, organelles and the nucleus of the cell. Figure.1: The composition of cell The cell conducts the electric current as the ions in it act as charge carrier and the tissue fluid along with cell fluid act as electrolyte where the free movement of charge takes place-giving rise to electric current. This is as same as that of electrons in metals. 3

11 Table.1: Typical ion concentrations in mammalian cytosol and blood Ion Concentration in cytosol (millimolar) Concentration in blood (millimolar) Potassium 139 Sodium 1 1 Chloride 116 Bicarbonate 1 9 Amino acids & proteins Magnesium.8 1. Calcium <. 1.8 (Data obtained from [3]) Every living organism also has dielectric properties at every level (cellular, molecular, tissue). This is produced by polar elements that can orient their electrical pole in the direction of a gradient field and therefore the dielectric properties are contributed by cell constituents and cell organelles. Thereby, this dielectric medium with lipid bilayer act as a capacitor with an approximate capacitance of.1 F/m []. Figure.: Bi-lipid layer in cell membrane The dielectric properties also depend upon the composition of cells, its organelles, structure of tissue, etc. Therefore, by the above-mentioned properties, a cell can be assumed to be as similar to a simple electrical circuit with a capacitor and with its conductance.

12 .3 Electrical equivalent of cell In reality, most materials including biological tissues act as both dielectric conductor as it has immobile charges that can be polarized but they do not move and free charges that contribute to the current flow. The electrical properties of tissue are given by the cell and tissue constituents and their structure. An equivalent model of cell is represented in the below figure.3. Figure.3: Electrical Equivalent of cell. The circuit (a) is the one relying on cell membrane and the intra- and extra- environment of it. The circuit (b) is the equivalent of circuit (a) and the circuit (c) is the simplification of the circuit (b). Frequency dependent EBI For living tissues, both the permittivity and the conductivity are depending on frequency. This dependence is named dispersion. The dispersion can be sorted into four major classes, α-dispersion, β-dispersion, γ-dispersion, and δ-dispersion. At the frequency range of the Beta dispersion, the Cole model equation was found. Four parameters are applied in this equation and it can plot the Cole plot...1 Cole equation The EBI measurement of any biological tissue is given by the Cole equation Where R = Resistance at DC frequency R = Resistance infinite frequency τ = Time constant (RC) and α = 1 in a typical RC circuit (Cole, 19) (.1).. Dispersion in tissue According to Grimnes and Martinsen, for all the biomaterials, the dispersion is evident as a function of frequency under electrical examinations, which is also evident from the above mentioned

13 Cole equation as the equation has which is equivalent to. Thereby, the dispersion here occurs as a function of frequency []. The dispersions are classified into four types and they are α-dispersion, β-dispersion, γ-dispersion, and δ dispersion. More information can be found on this in (Grimnes & Martinssen 8) nd edition. Figure.: Dielectric dispersions (permittivity on the left and conductivity on the right) in the brain grey matter, illustrated from []. Table.: Dielectric dispersions (Grimnes & Martinsen, 8 nd edition) Type Frequency Main Mechanism range α mhz-khz Counter ion effects near the membrane surfaces, active cell membrane effects and gate channels, intracellular structures and ionic diffusion, dielectric losses. β.1 1 MHz Maxwell- Wagner affects, passive cell membrane capacitance, intra cellular organelle membranes, protein molecule response. γ.1 1 GHz Dipolar mechanisms in polar media such as water, salts and proteins.. EBI measurements From the mid 9 s, the measurement of EBI has been playing a crucial role in the medical field. In order to know the correct EBI measurements or in other words to do the correct EBI analysis, one must know the details of the specific conductivities and relative permittivity of biological tissue. The table below can give a further more insight on this. 6

14 Table.3: Data Ranges of Specific Conductivities and Relative Permittivity of Tissues at Low-Frequencies Specific conductivity Relative permittivity Tumor.. 6 (at 1 khz) Fat.. 1 (at 1 Hz) Muscle Transversal..1 1 (at 1 Hz) Longitudinal (at 1 Hz) Skin (dry) (at 1 Hz) Stratum corneum.1 1 (at Hz) Lower-lying layers.7 1 (at Hz) Bone (d.c.) Blood (at 1 khz) Heart.6. 7 (d.c.) Kidney.6 3 (d.c.) Liver.3. 1 (d.c.) Lung (inflated)..9 1 (d.c.) Spleen.3 (d.c.) Gray matter.33 (d.c.) White matter.3 3 (d.c.) (Data obtained from [6]) Due to its safety and non- invasiveness, EBI has become the subject for many researchers and scholars. Though EBI is a simple measurement method it might suffer from severe measurement artifacts. One of the most commons artifacts as the known is capacitive leakage artifact of Hook effect...1 Hook effect Though the EBI measurement is simple, it has a commonly encountered artifact, i.e. Hook effect and in order to use it, the measurements have to be rid off this artifact. This effect is caused due to the capacitances at high frequencies. (Buendia, 9)..1.1 Origin and its effect The Hook effect usually is associated more to reactance than to resistance. It arises from the creation of a current divider circuit by an impedance measurement load with a parasitic capacitive pathway in parallel to each other. Parasitic capacitances are always present in any type of AC circuit and their influence, become critical at high frequencies. The Hook effect is named after its resemblance with the hook like shape in its impedance plot. 7

15 Figure.: The Hook effect to reactance (a) and phase (b) The Hook Effect must be corrected and compensated since it influences further analysis based of EBI, especially when doing Cole-based analysis and obtaining the Cole parameters, fittings and its estimations []...1. Correction and Compensation To compensate the hook effect, the obtained EBI measurements are multiplied with an exponential factor, exp [-jωtd]. But this only compensates the phase error. This compensation method using a scalar Td is called Td compensation and it is well spread despite its limitation to compensate the estimation error in the module..6 Cerebral monitoring There are several ways of monitoring the brain, both invasively and non-invasively. Cerebral monitoring using EBI measurements belong to the non-invasively class. Several investigations have been carried out for the past years, but for the past years, extensive research work has been done on the EBI-based cerebral monitoring from it. In order to monitor the cerebral EBI measurements successfully, many issues come across that has to be taken care off like instrumentation, electrodes, measurement analysis, etc. It has been found that changes in EBI measurements occur when the brain suffers from injuries or cell swelling [7]..6.1 Cellular damage The most sensitive tissue is the neural tissue because its functions depend on various metabolic pathways, oxygen supply, stress, genetics, surrounding environment and many more. Whenever the brain cells face these factors, they try to adapt themselves to get stability and this effort of it is called as dominated cellular adaptation. But this adaptations last only up to an extent and after that they suffer injuries if the limits are crossed and it becomes difficult for the cells to reverse the adaptation process and the cells are damaged. 8

16 .6. Causes The damages can have many reasons like insufficient oxygen, stress, insufficient blood, etc. but the most common among all is the injury caused by hypoxia, which is nothing but the loss of oxygen. The most common origin of hypoxia is Ischaemia. Hypoxia is the condition faced when oxygen supply is insufficient to brain and Ischaemia is the condition where blood is insufficient. Lack of oxygen supply leads to cellular swelling and this condition is named as cellular Oedema [7]. Figure.8: Intra cerebral hemorrhage.6.3 Need for cerebral monitoring As per the earlier studies and surveys, it can be inferred that the cerebral monitoring is very important for decreasing the death rate of the patients dying due to cerebral hemorrhage. Studies indicate that the cerebral damage occurs if neurons undergo injury, and then there arises a change in its electrical properties, thereby producing a change in EBI measurements of the brain. [] The injuries can refer to oedema, hypoxia/ischaemia, perinatal asphyxia, etc. In order to avoid them, if EBI successful, EBI measurements could be used for brain monitoring- and early detection of brain damage.[8] 9

17 Chapter 3 FLOW OF ANALYSIS 3.1 EBI Spectroscopy Measurements Complex EBI measurements obtained from 19 sick neonates have been analyzed. The measurements belong to two different groups, 1 out of 18 babies were born sick but show good outcome, and the other 7 did not recover and parish. The group with good outcome contained 93 measurement files and the group with bad outcome contained 3 measurement files. Both set of files have been analyzed by fitting the data to the Cole extended model to get the best fittings. Using a standard error estimate parameter does the validation of the BIA equations in a sufficiently large number of subjects. The reason behind using this feature is that it is very safe for the patient. And by far, the complex studies related to the brain can be done easily by the EBI spectroscopy. Thus, the best fittings were selected from each of the subjects and descriptive statistics were performed on them to characterize the sets. 3. Cole function fittings The Cole parameters were obtained by fitting the EBI measurements to the Cole extended model using the batch-processing tool of the BioImp software. To perform a fitting the Bioimp software allows setting 3 different parameters: frequency range of the fitted data, the rejection limit of the EBI data considered for the fitting and the value of the parameter Td for Td compensation. Since the data of this study contained several artifact and there was no method with valid scientific grounds to select the best parameters for each fittings. Therefore, a brute force fitting approach was used fitting the EBI data with several combinations of the 3 parameters according to the following list: a) Frequency ranges (- KHz, 7-6 KHz, 1- KHz, - KHz, - KHz). b) Rejection limits (%, %). c) Td correction values (ns, ns, 1ns, 1ns, ns). In this way, a total of 61 files were obtained and were segregated for the analysis work Bioimp Software The software used for the manipulation and fitting of the EBI measurements is the Bioimp Version... developed by ImpediMed Ltd. This software tool is the standard software accompanying to the SFB7 EBI spectrometer manufactured by Impedimed [9]. 1

18 3.3 Filtering analysis To ensure that all the data obtained with the Cole fittings were filter to remove inaccurate fittings and failed fittings. The obtained data were filtered according to filtering parameters contained in Figure 3.1. Cole fitted Data <f char < 6 khz X centroid <Ω SEE<% Filtered Data Figure 3.1: Flow of data filtering in Excel When using BioImp the upper frequency limit of the Cole fitting process was set as high as 6kHz and all the measurements presented a characteristic frequency within the frequency limits of the selected frequency range, therefore no characteristic frequency over 6kHz must be produced; According to the characteristics of Cole Model, the center of the semi-circle of the Cole plot must be always below the resistance axis, which means that the reactance center of the circle must be always smaller than zero. In addition, to filter all the fitting with a S.E.E value larger than % ensures that only accurate fittings are further analyzed. 3. MATLAB EBI manipulation and descriptive statistics This section of analysis has been done by writing MATLAB scripts in order to analyze the obtained Cole fitted data. Especial attention has been paid to the analysis of the characteristic frequency Data processing in MATLAB Each subject generated a histogram indicating the characteristic frequency values obtained with the Cole fittings. Only the subjects with certain even distribution of values around the centroid were selected for further analysis. After such filter, 9 sick subjects with good outcome and 6 acute subjects remain for further analysis. The next step is to select the 6% of the fittings producing a characteristic frequency value near the main lobe of the distribution. Finally, the measurement with the smallest S.E.E. value among for each of the subject was selected. All the selected measurements were stored in a file named Small_SEE. The final section of the analysis performing the descriptive statistics of the 11

19 Cole parameters produces with the set of selected-files. Filtered Data Centroid fitting 6% fitting Smallest S.E.E Best Measurement Figure 3.: Flow of data processing in MATLAB Note: As one of the important parameters of Cole model, the alpha parameter is not produced by the Bioimp software, the alpha parameter is calculated from the parameters of R, R and the radius of the obtained Cole plot as follows in the label (3.1). (3.1) 3.. MATLAB functions The analysis work in MATLAB was done with four functions. For the two groups of babies, the inputs and outputs are not the same when the same function is called Histogram creation A histogram is generated for each subject by applying the MATLAB function hist (). One parameter should be calculated first, to obtain the number of bins-to represent the histogram. This is realized by calling the BIN_NUMBER () function. The character frequency values are the inputs and the main output is the number of bins when the bin size is equal to Subject selection After the histograms were generated, subjects to be included for further analysis must be selected. To decide which subjects remain for further analysis, the distribution of the values for the characteristic frequency is studied and compare with the average value. The CENTROID () function was called to calculate the centroid value of the distribution characteristic frequency values for each subject. The character frequency values and the variable conserving the number of data points in each segment were applied to acquire the centroid value. The function SUBJECT_SELECTION () was called when the centroid value had been obtained. A Boolean value was returned indicating that whether the centroid value is inside or outside the main distribution lobe Best measurement selection In order to choose the best measurement, SELECTION () function was called. It performs the calculation of the frequency range containing 6% of the values for the characteristic frequency around the center frequency of the distribution of values for each subject. That is that in a subject 1

20 with a center frequency of 8 khz and the 6% of the characteristic frequencies are distributed around the center value from 17 to 388, as indicated in Figure 3.3, only the fitting generated with a characteristic frequency within that range would evaluated for the best fit. Figure 3.3: Schematic diagram presents the method to calculate the range of 6% for f c 13

21 Chapter RESULTS This chapter presents the results obtained from the processing and analysis of the EBI data as presented in the previous chapter. Firstly, the histograms showing the distribution the values of the characteristic frequency obtained for the obtained Cole fittings for all the EBI measurements ARE PLOTED BY SUBJECT. Secondly, the histograms indicating the frequency range containing the 6% of the fittings around the center frequency of the histogram are shown. And finally, the Cole parameters together with other impedance parameters from the selected best fittings are represented..1 Histograms of f char for each subject After filtering the Cole Fitting as indicated in section 3...3, different distributions of characteristic frequency were obtained for each subject. Figures.1 to.18 present the obtained histograms for all the 18 subjects. In this section, the histograms with the characteristic frequency are shown for each neonate with good outcome..1.1 Sick neonates with good outcome The nine following figures include the histograms of the remaining subjects: Baby8 ******* Centroid Frequency(kHz) Figure.1: Histogram of obtained f char for subject 8 with good outcome The distribution in this histogram looks like a Gaussian distribution with one main lobe. The centroid value equals to 83.3 khz. The value at khz indicates those non-fit results 1

22 1 1 1 Baby1 ******* Centroid Frequency(kHz) Figure.: Histogram of obtained f char for subject 1 with good outcome The shape of this histogram is similar like Gaussian distribution but most of the frequencies are located on the right side of the centroid. The centroid value is khz. One main lobe exists. Baby3 ******* Centroid Frequency(kHz) Figure.3: Histogram of obtained f char for subject 3 with good outcome This histogram consists of several lobes spread over wide range and the centroid is at 1.9 khz Baby36 ******* Centroid Frequency(kHz) Figure.: Histogram of obtained f char for subject 36 with good outcome Subject 36 have four main lobes and several sides lobes. And the centroid locates at 8.78 khz. 1

23 Baby1 ******* Centroid Frequency(kHz) Figure.: Histogram of obtained f char for subject 1 with good outcome Subject 1 only have one main lobe and centroid is at 9.1 khz Baby ******* Centroid Frequency(kHz) Figure.6: Histogram of obtained f char for subject with good outcome This characteristic frequency distribution has one main lobe. The centroid is at 9. khz Baby ******* Centroid Frequency(kHz) Figure.7: Histogram of obtained f char for subject with good outcome This histogram has one main lobe and many frequencies spread around it. Few frequencies are located away from this main lobe. The centroid is at 7.6 khz. 16

24 1 Baby7 ******* Centroid Frequency(kHz) Figure.8: Histogram of obtained f char for subject 7 with good outcome The histogram distribution has a main lobe and all the frequencies are almost concentrated. The centroid is at 7.39 khz Baby7 ******* Centroid 1 1 Frequency(kHz) Figure.9: Histogram of obtained f char for subject 7 with good outcome In this histogram, many frequencies form a Gaussian distribution with only one main lobe. Few frequencies are outside and the centroid is 8. khz. Ideally, all figures should follow the Gaussian distribution. As many noise and interference exist everywhere in practice, it is difficult to obtain pure Gaussian distribution. Therefore, it is impossible to decide which kind of subjects can be left by comparing the shapes of the distribution for f char. Another method to choose suitable subjects is to check the position of the centroid values. According to all the above figures, the centroid values are all placed among the first five peaks of f char values. Subjects with such kind of centroid values can be remained for future analysis. The following two subjects should be removed from further analysis as they have odd distributions: 17

25 Baby6 ******* Centroid Frequency(kHz) Figure.1: Histogram of obtained f char for subject 6 with good outcome Here, the main lobe is found to be at initial frequencies and a side lobe near the central frequencies. The centroid is.86 khz, which is outside the first five peaks. 1 1 Baby1 ******* Centroid Frequency(kHz) Figure.11: Histogram of obtained f char for subject 1 with good outcome This histogram has only one main lobe and several spurious side lobes. However, the centroid valued 7. khz is still outside the first five peaks..1. Sick neonates with no recovery This section will give all the histograms related to subjects that didn t recover finally. The first seven histograms belong to the subjects that will be left for next stage: 1 1 Baby3 ******* Centroid Frequency(Hz) Figure.1: Histogram of obtained f char for subject 3 with acute sick In this histogram, only one main lobe exists. The shape looks like a Gaussian distribution and the centroid is at 17. khz. 18

26 6 Baby ******* Centroid Frequency(Hz) Figure.13: Histogram of obtained f char for subject with acute sick This subject has a histogram with one main lobe and the centroid is at 11.3 khz. 3 3 Baby3 ******* Centroid Frequency(Hz) Figure.1: Histogram of obtained f char for subject 3 with acute sick In this distribution, on the right side of the centroid (. khz), most frequencies exist and spread in a wider range than the other side. 3. Baby8 ******* Centroid Frequency(Hz) Figure.1: Histogram of obtained f char for subject 8 with acute sick This subject has only one lobe and intervals between peaks are quite large. The centroid is 3. khz. 19

27 Baby3 ******* Centroid Frequency(Hz) Figure.16: Histogram of obtained f char for subject 3 with acute sick The histogram of this subject contains one main lobe and several sides lobes. The centroid is.8 khz. 8 7 Baby63 ******* Centroid Frequency(Hz) Figure.17: Histogram of obtained f char for subject 63 with acute sick This histogram looks like a perfect Gaussian distribution, with one main lobe at the centroid, i.e khz. The following figures represent the histograms of subjects with odd distribution of f char values: 3 Baby ******* Centroid Frequency(Hz) Figure.18: Histogram of obtained f char for subject with acute sick As this figure shows, the centroid value (6. khz) of this subject does not fall in the range of the first five peaks. According to the selection criterion, this subject should be removed from further work. However, compared with other subjects in the same group, this histogram looks more like a Gaussian distribution. An exception should be given to this subject because if the bin size was

28 changed, it is possible that this subject could be remained and used in further analysis. It can be observed that the values for centroid are located the range of the first five peaks of f char. In the two classes, totally 3 subjects have such kind of centroid values. These subjects were deleted from further analysis. However, the shapes of histograms should also be considered. If the histogram of a subject follows Gaussian distribution or looks similar, this subject could also be kept.. Histograms of the 6% of f char In this section, figures indicate the 6% of the fittings producing a characteristic frequency value near the main lobe of the distribution. This range gives the most important and useful information of subject. The red star line in each figure indicates the best measurement...1 Fittings for sick neonates with good outcome The following figures indicate the figures of all remained 9 sick neonates with good outcome. 7 6 Baby8 ******* Best measurement Frequency(kHz) Figure.19: Histogram of the 6% of f char for subject 8 with good outcome The histogram starts at 7.18 khz (7 khz in histogram) and ends at 381. khz (381 khz). The value of the best measurement is khz. 1 1 Baby1 ******* Best messurement Frequency(kHz) Figure.: Histogram of the 6% of f char for subject 1 with good outcome This subject has the characteristic frequency range spanning from 113. khz (113 khz) to 7.3 (7) khz. The frequency value for the best measurement is.18 khz. 1

29 ******* Best measurement Baby Frequency(kHz) Figure.1: Histogram of the 6% of f char for subject 3 with good outcome The histogram starts at 9.1 khz (9 khz) and ends at.1 khz ( khz) khz is the frequency value for the best measurement Baby36 ******* Best measurement Frequency(kHz) Figure.: Histogram of the 6% of f char for subject 36 with good outcome In this histogram, the characteristic frequencies range from 3.1 khz (3 khz) to khz (11 khz). The frequency value for the best measurement of this subject is. khz Baby1 ******* Best measurement Frequency(kHz) Figure.3: Histogram of the 6% of f char for subject 1 with good outcome This subject s characteristic frequency distribution starts at.6 khz ( khz) and ends at 88.1 khz (88 khz) with the best measurement frequency value at 6.63 khz.

30 8 7 Baby ******* Best measurement Frequency(kHz) Figure.: Histogram of the 6% of f char for subject with good outcome Here, the characteristic frequency values of the subject starts at 8.8 khz (8 khz) and ends at 1.9 khz (1 khz). The best measurement s character frequency value is khz. 3. Baby ******* Best measurement Frequency(kHz) Figure.: Histogram of the 6% of f char for subject with good outcome The characteristic frequency values ranges from.33 khz ( khz) to 1. khz (1 khz) with the best measurement s frequency value being equal to 3.71 khz Baby7 ******* Best measurement Frequency(kHz) Figure.6: Histogram of the 6% of f char for subject 7 with good outcome This histogram starts from 6. khz (6 khz) to 11. khz (11 khz). The best measurement s character frequency is 67. khz. 3

31 Baby7 ******* Best measurement Frequency(kHz) Figure.7: Histogram of the 6% of f char for subject 7 with good outcome In this histogram, the characteristic frequencies are from 37.3 khz (37 khz) to 67. khz (67 khz) and the best measurement s characteristic frequency value is equal to 7.78 khz... Fittings for sick neonates without recovery This section will display figures of those babies that didn t recover. 1 1 Baby3 ******* Best measurement Frequency(Hz) Figure.8: Histogram of the 6% of f char for subject 3 with acute sick The two edges of the histogram are khz (79 khz) and khz (168 khz). The characteristic frequency of the best measurement is khz. 9 8 Baby ******* Best measurement Frequency(Hz) Figure.9: Histogram of the 6% of f char for subject with acute sick The range of the characteristic frequencies is from 6.31 khz (6 khz) to khz (16 khz) khz is the frequency value for the best measurement.

32 3 Baby3 ******* Best measurement Frequency(Hz) Figure.3: Histogram of the 6% of f char for subject 3 with acute sick In this histogram, the frequency range starts at 1. khz (1 khz) and ends at.3 khz ( khz). The frequency of the best measurement is 1.37 khz. 3. Baby8 ******* Best measurement Frequency(Hz) Figure.31: Histogram of the 6% of f char for subject 8 with acute sick This characteristic frequency range starts from.3 khz ( khz) and ends at.7 khz ( khz). 39. khz is the frequency of the best measurement. Baby3 ******* Best measurement Frequency(Hz) Figure.3: Histogram of the 6% of f char for subject 3 with acute sick The left edge and the right edge in this histogram are 7.16 khz (7 khz) and 73.1 khz (73 khz), respectively. The frequency of the best measurement locates at 31.3 khz.

33 8 7 Baby63 ******* Best measurement Frequency(Hz) Figure.33: Histogram of the 6% of f char for subject 63 with acute sick The histogram indicates that the frequency range is from 6.68 khz (6 khz) to khz (18 khz) with the frequency of the best measurement being at 79.8 khz. 3 3 Baby ******* Bestmeasurement Frequency(Hz) Figure.3: Histogram of the 6% of f char for subject with acute sick This histogram shows that the frequency range is from 6.3 khz (6 khz) to 9.6 khz (9 khz) and frequency value of the best measurement is 8.67 khz. For each subject, measurements whose characteristic frequency values are inside the 6% range were chosen for future analysis work. Other parameters of chosen measurements will be acquired and listed later..3 Minimum S.E.E. criterion for best fitting analysis The criterion to select the best fitting measurement has followed the minimum value for Standard Error of Estimate for the Cole fitting. The following two tables display the values of the Cole parameters, the characteristic frequency and other EBI parameters for the best-performed fit received for each subject of the two classes. The alpha values are also included in the tables. 6

34 TABLE.1: PARAMETER VALUES OBTAINED IN THE ANALYSIS With Good Outcome Subject X centre R centre Radius SEE R R Z char f char α Mean TABLE.: PARAMETER VALUES OBTAINED IN THE ANALYSIS With Acute Sick Subject X centre R centre Radius SEE R R Z char f char α Mean Reference values After all the best measurements were selected, the maximum, minimum, mean and standard deviation values were calculated for all obtained parameters to get the impedance reference values. Those values are shown in Table.3 and Table.. TABLE.3: ANALYSIS FOR OBTAINED PARAMETER VALUES With Good Outcome X centre R centre Radius SEE R R Z char f char α Max Min Mean SD

35 TABLE.: ANALYSIS FOR OBTAINED PARAMETER VALUES With Acute Sick X centre R centre Radius SEE R R Z char f char α Max Min Mean SD

36 Chapter DISCUSSION, CONCLUSIONS AND FUTURE WORK.1 Discussion As the measurements were taken from sick and healthy neonates, the analysis becomes little different with each of them. As the group of healthy neonates remain healthy, the analysis will become easy, as the constraints do not change by virtue of time. But, in case of the neonates who become healthy, the constraints and parameters change over a period of time. Also, there are slightly less number of measurements available for acute sick neonates. This is due to the reason that they die in short period of time. The reasons can be many like perinatal asphyxia or more appropriately hypoxic-ischemic encephalopathy (HIE), which causes high mortality and morbidity for long period of time. Neonates with good-outcome present a minimum to maximum range of the characteristic frequency twice as large as the range presented by the acute sick neonates. This might be caused by the fact that since the acute sick neonates is born with a sick brain and they do not recover. Therefore all the EBI measurements are taken on a sick brain. In the other hand, the neonates with good outcome are supposedly with a sick brain, but with time the brain heals and the newborn live longer than the acute neonates. Therefore, it becomes much easier to analyze the healthy neonates. Thereby, the measurements of the brain of neonates with good outcome are taken over a changing brain and are expected to present a wider range of characteristic frequency. The observed distribution of values for characteristic frequency of the neonates with good outcome suggests that maybe there is a time shift from lower frequencies to larger. This is something that cannot be proven in the present study but it deserves to be investigated further.. Conclusion From the distribution of the obtained characteristic frequency, it is clear that the mean of the values for both cases is easy to differentiate. However, the characteristic frequency values of the two groups have the largest difference. If this references value was used, it is much easier to distinct one group from another. Sure both classes present overlapping values, but it is clearly seen that the concentration of values of acute sick for the characteristic frequency occur at much lower frequencies than for the neonates with good outcome..3 Future work As it has been mentioned before there is some overlapping between the values of the characteristic frequencies that avoid making a perfect distinction between classes. Since the EBI data 9

37 taken from neonates with good outcome might contain measurements from different brain status, sick to healthy. A more detailed time analysis on the obtained measurement from neonates with outcome could confirm that such time component is present in the good outcome case. The later is better important to build a predictor of outcome. 3

38 REFERENCES [1] G. N. H. a. B. E. Lingwood, "Reference values for whole body and cerebral multi-frequency bio-impedance data in neonates less than 1 h postpartum," Physiological Measurement, vol. 7, pp , 17 July 6 6. [] K. R. F. a. H. P. Schwan, Dielectric properties of tissues: New York: CRC Press, [3] H. F. Lodish, Molecular cell biology: New York: Scientific American Books, [] F. Seoane, et al., "Electrical Bioimpedance Cerebral Monitoring. A Study of the Current Density Distribution and Impedance Sensitivity Maps on a 3D Realistic Head Model," in Neural Engineering, 7. CNE '7. 3rd International IEEE/EMBS Conference on, 7, pp [] Ø. G. M. Sverre Grimnes, BIOIMPEDANCE AND BIOELECTRICITY BASICS, Second ed.: Academic Press,ELSEVIER, 8. [6] D. M. a. F. X. Hart, Electric Properties of Tissues: John Wiley & Sons, 6. [7] F. Seoane Martínez, "Electrical bioimpedance cerebral monitoring," Doctoral University College of Borås. School of Engineering, University College of Borås, Borås, 7. [8] S. G. a. Ø. G. Martinsen, "Journal of Electrical Bioimpedance (JEB) a new, open access, scientific journal," Journal of Electrical Bioimpedance, vol. 1, p. 1, 1. [9] Manual, "Bioimp,"... ed, p. BioImp Body Composition Analysis Software for the Imp SFB7. 31

39 APPENDIX A Flow chart of MATLAB work SubjFilteredData Data Separation per Subject Histograms of f char Subject Selection Histograms of All Selected Subjects Fixed Rang of f char (6%) Measurements inside Fixed Range Sorting in Ascending Order Best Measurements (Minimum S.E.E) Maximum, Minimum, Mean and Standard Deviation of Parameters Figure A.1: Flow of MATLAB work 3

40 APPENDIX B MATLAB functions Table B.1: MATLAB functions 1) Name: BIN_NUMBER () Inputs: Subject name (a), f c (b) Description: When called, calculate the number of bins with the bin size equals to 1. ) Name: CENTROID () Inputs: f c, the distribution of f c Description: When called, achieve the centroid value. 3) Name: SUBJECT_SELECTION() Inputs: f c, the distribution of f c, centroid value, length of f c Outputs: f c of the number of bins (a), the number of bins (b) Outputs: Centroid value Description: When called, decides whether the subject can be remained for further analysis. ) Name: SELECTION () Inputs: The distribution for f c with order number, subject name (a), original data, start and end points for each subject (b) Outputs: Boolean value, the distribution of f c with order number Outputs: Best measurement, mean, measurements inside the 6%, f c, the number of bins, summary of all percentage inside 6% Description: When this function is called, best measurement with the smallest SEE is selected and the preparation for creating new histogram and final analysis work is also done. Note: - (a) and (b) refers to the group of sick neonates which become healthy and group of sick neonates which remain sick. 33