ANALYSIS OF APPLIED REFERENCE LEADS INFLUENCE ON AN EEG SPECTRUM

The article presents the analysis of various kinds of montage applied in the analysis of electroencephalographic data. The present study covers analysis of data obtained as a result of EEG examination of several persons with the age of 23–24 years. Every person has undergone an examination with appliance of 21 electrodes in accordance with classic EEG record method applying the 10–20 system. The examined persons were asked to perform the following activities: sitting steadily with opened eyes, sitting steadily with closed eyes as well as typical cognitive activity i.e. silent reading of given text. Every fragment has been subjected to initial data analysis (filtering, artifact correction) and used for spectral analysis afterwards. The whole analysis has been realized on the basis of four EEG montage examples (two bipolar and two monopolar). For each of them comparative analysis concerning the following points has been carried out: EEG spectra as well as other measures, such as activity maps, diffraction histograms for separate waves and spectrum charts for the selected electrodes.


Introduction
Electroencephalography (EEG) is non-invasive diagnostic method used for examination of brain bioelectrical activity by the means of a device called an electroencephalograph. This device records electrical changes appearing in the millions of brain neuron cells. EEG electrodes are placed at the surface of the head. Specially designed cap is used to keep them in proper position.
The first study on human EEG was conducted in 1929 by the German psychiatrist Hans Berger. This discovery provided a new neurologic and psychiatric diagnostic tool designed especially for purposes of neurologic diagnosis and neurosurgical operative procedures planning [14].
Nowadays EEG recordings might be visible at computer screen and saved to a computer memory. Currently available EEG analysis software make it possible to analyze previously recorded data repeatedly and with different settings. One of such setting is the location of reference electrode (electrodes) know as a montage [17]. It is possible to apply different kind of montages [12,18]. Their schemes might have significant impact on the EEG data analysis [16]. This paper is dedicated to analysis of the different montage types application to the typical EEG analysis.

Electroencephalography
Electroencephalography is the research method aimed at studying bioelectrical brain function as well as the functional state of the central nervous system. EEG is useful in the diagnosis of many nervous system diseases e.g.: epilepsy, insomnia, headaches and migraines, brain tumors and injuries. Also EEG is very important for monitoring and diagnosis of consciousness disturbances as well as emotional disorders and attention deficit disorders causing difficulties in learning process. BCI (Brain-Computer Interface) is widely applied for EEG. It enables to establish the connection with external world through detecting brain waves. A special game based on EEG exists, it is aimed at training concentration by strengthening the desired brainwave rhythms. Network authentication or identification of a person is based on unique characteristics of processed EEG signal. In this purpose the system consisting of computer and special sensors transmitting the collected signal to the evaluation unit is used. The kit allows to maintain effective communication between computer and brain in on-line mode, so for example applets in a smartphone can be controlled by mood and emotions [10].
Currently EEG research methodology is well established. In standard clinical study 19 electrodes are placed on the head surface. This distribution is widely known as 10-20 system recommended by the International Federation of Clinical Neurophysiology. It consists of eight electrodes on each hemispheres and three electrodes in the midline [4].
Correct electroencephalography record of a health adult has some characteristic features. Human sitting with eyes closed has a dominant rhythm in the form of regular alpha wave activities. It properly differentiated spaces i.e. with an amplitude decreasing from occiput anterior. Leads in front leading dominated by lowvoltage operation beta.
Under normal physiological conditions brain waves are found in the frequency range 1-100 Hz and an amplitude between 5 and several hundred μV. In the case of pathology (for example, damaged cells or impairment of the chemical conduction) some of these pathologies might occur in EEG record: delay or acceleration of the speed of flow, increase or decrease of the amplitude, change of their shape or configuration [7].
There are several EEG waves occurring in typical record (Fig. 1). There are several wave types amongst them:  delta waves (δ),  theta waves (θ),  alpha waves (α),  beta waves (β),  gamma waves (γ). Theta waves (θ) are the waves with frequency values from 4 Hz to 8 Hz. Theta activity may be observed during hypnotic states such as trance, hypnosis and light sleep. They are associated with 1( nap) and 2 ( light sleep) stages of NREM sleep. Different kind of theta is associated with cognitive activity and particularly with attention as well as with memory processes. Consciousness at this frequency makes possible control of physical pain and even bleeding in extreme cases. Thinking becomes restless at this frequency and ability of logical thinking disappears, what can be seen clearly on the example of thinking activity during dreaming.
Alpha waves (α) have frequency of 8 Hz to 13 Hz. Amplitude is approximately 30-100 μV. Alpha waves are clearly visible in the absence of visual stimuli (when eyes examined person are closed). Their suppression occurs during visual perception. Alpha waves are also associated with the state of relaxation and a reduced level of cognitive activity, e.g. during lying with closed eyes before going to sleep and in the morning during awakening. This wave type is used in the techniques of accelerated learning.
Beta waves (β) are the wave type with a frequency from 13 to 30 Hz, their amplitude is below 30 μV. They reflect the commitment of the cerebral cortex in cognitive activity. Beta waves of small amplitude appear during the periods when the attention is focused. Additionally they can be caused by various pathologies, and chemicals substances such as benzodiazepines. Benzodiazepines are drugs with anxiolytic, sedative, hypnotic, anticonvulsant action.
Gamma waves (γ) have the frequency value around 26-100 Hz. Gamma activity is associated with higher cognitive processes such as perception and sensory memory. Gamma waves with the frequency of about 40 Hz are related to perceptual consciousness (connected with impressions and perception) as well as to integration of various sensory modalities (sight, hearing, sense of smell and touch) [2].

EEG configuration
EEG signal is received as voltage difference between two electrodes. There are many possibilities to make connections of electrodes. These connections are known as montages, which is a design of electrodes derivations. Nowadays recording EEG data on computer makes it possible to apply many types of montages in one study [11], in order to obtain various results of EEG test [9]. Two types of montages are commonly distinguished: bipolar montage and referential montage. Obviously, there are more montage types available [3]. Among them one can find (Fig. 2):  Bipolar montage. In Bipolar montage voltage difference is measured by connection two adjacent electrodes. For instance Fp1 is connected together with F7 or F3. This type is particularly used to test selected fields activity of nerve cells, not entire brain.  Referential montage (Monopolar montage). In this case voltage difference is measured between electrodes and a special electrode, which is called the referential electrode. This montages are applied for examining the activity of all nerve cells.
(By using this montage, it is easy to detect various nerve disease of brain, e.g. epilepsy).
 Monopolary montage (M3). In this type monatage electrodes A1 and A2 are defined as referential electrode. The electrodes of the left hemisphere of the brain and electrodes Fz, Pz are linked with A1, and with A2 the right hemisphere and electrode Cz.  Average Reference montage (M4). It is necessary to mention Average Reference montage, in which values off all electrodes (without reference electrode) are summed up and averaged. Finally, difference between average and referential electrode is measured.  Laplacian montage. This kind of montage is very similar to common average reference montage, but the difference is that there are local weighted average reference. It means, that average of neighboring electrodes are taken into account. It is very useful especially for low or moderate artifacts [6].

Spectral analysis
Spectral analysis is used for data analysis in various domains. Spectrum analysis is analysis in terms of a spectrum of frequencies based on time series. It may have application for instance in engineering, astronomy and geophysics. Increasingly spectral analysis is applied in medical diagnostics.
A human brain emits specific waveforms, which may be gathered by using EEG (electroencephalography). Spectral analysis is used for examination of brain activity.
Spectral analysis is a mathematical approach to quantify the EEG. It decomposes a signal into its constituting frequency components. Quantitative electroencephalography (QEEG) has been accepted for clinical application for example in epilepsy.
The transition from the time domain to the frequency domain is performed using the Fourier transform. Each of spectral analysis is based on the Fourier transform. Fourier analyzer transforms a signal over time into a frequency. It is a tool to analyze neural signals.
There are several methods of spectral analysis. First of them is harmonic analysis. Harmonic analysis is applied in case if signal is composed of a sum of individual oscillatory components. There are also Blackman-Tukey Transformation and MESA -Maximum Entropy Spectral Analysis.
The fast Fourier transform is the method for obtaining the EEG spectrum [1].
A fast Fourier transform (FFT) algorithm computes the discrete Fourier transform (DFT). It manages to reduce the complexity of computing the DFT [15].
The sampling frequency must be consistent with the sampling theorem. If the sampling theorem is not respected, aliasing appears. Aliasing causes that signals to become indistinguishable.
In the frequency domain spectrum of the signal is multiplied in the sampling.
If the maximum frequency sampling signal (width of the spectrum) exceeds a half the sampling frequency, next recurring spectrums begin to overlap. According to the theorem of Kotelnikov-Shannon the original signal can not be reproduced with distorted samples. Figure 3 shows spectrum analysis in the time and frequency domain.

Testing procedure
Participants took part in experiment in a quiet testing room in dedicated laboratory with artificial lighting. The experimenter explained the procedure. The participants were instructed to minimize body movement to minimize potential artefacts in their EEG signals and to fixate on the centre of the monitor.
The 21 channel EEG amplifier -Mitsar EEG 201was applied to record EEG data. EEG amplifier transmitted data to computer with the native software (EEG Studio). The room was illuminated with standard fluorescent light and outside light was blocked to remain stable conditions for experiment's durations.
The samples were recorded with frequency of 500 Hz. The signal was gathered from nineteen using cup-shaped electrodes located over the user scalp according international 10-20 EEG standard (Fig. 4). Electrodes were placed in a special EEG cap to keep them in the right position. In addition, the ground electrode was placed in the centre of the frontal lobe and two reference electrodes (A1, A2) were placed on ears.

Fig. 4. EEG electrodes localization
The signal was gathered from four participants. They were asked to perform standard activities:  sitting quietly motionless looking straight aheadresting state data-sitting with eyes closed and open,  sitting performing cognitive processing action: reading book quietlycognitive activity action data,  artefacts generating like blinking, speaking, hyperventilationnoised data. EEG signal preprocessing covered signal filtration (including frequency filtration) and artefact elimination. As the EEG signals have low amplitude, which is usually less than 100 µV, its quality and high signal-to-noise ratio is important.
Frequency filtration was applied to remove noisy part of the signal containing low level of information (such as signal frequencies lower than 3 Hz or higher than 45 Hz) as well as signal interfering with external electronic devices (signal with a frequency of 50-60 Hz).
Another important preprocessing step is artefact correction. It enables flattering unwanted signals. Depending on their origin, they can be divided into two groups: technical artefacts related to the registration signal and biological artefacts originating from the examined person. The first group is related to electric field noise and inaccurate attached electrodes. The second group of artefacts is related to the physiology of a patient: sweating, blinking and eye movement, tongue, head and neck muscles movements as well as electrical activity of the heart.
In the study Principal Component Analysis (PCA) [8] and Independent Component Analysis (ICA) [5] were applied to remove the artifacts [13]. Depending on needs, ICA or both, PCA followed by ICA were applied.

Analysis of the montage influence on the EEG spectrum
EEG spectra in the study were generated based on FFT method. The analysis was performed regarding four different EEG montages originating from different montage types including these presented at Fig. 1. All applied montages were generated based on the same EEG recording fragments where 21 electrodes were used. From the long, lasting almost one hour recordings several seconds fragments were chosen for the analysis. Different activity fragments, such as sitting with eyes open, closed and cognitive action processing were analyzed separately.
The analysis process covered noised EEG signal filtration and artifact correction with PCA and/or ICA method.
The analysis covered comparison of several aspects. The analysis included data visualization (such as tables and charts, brain activity maps, raw data presentation). Several comparison methods were applied in the study. They are:  Brain activity maps generated for both, signal amplitude and power,  EEG asymmetry visualization,  EEG spectra calculated for both, signal amplitude and power,  Histograms presenting EEG waveform distribution,  Chosen waveform ratio (such as alpha/theta ratio).
Brain activity records obtained as a result of brain examination with monopolar electrode montage can be presented as a map of activity, whereas the map of brain activity as per examination with bipolar electrode montage cannot be made due to the fact that for bipolar montage the voltage difference of every electrode is measured in relation to two neighboring electrodes, so there is no single reference level to build up a potential map, because a potential is always relative and needs a single specified reference point.
The analysis covered a series of different results. Some of them, the most significant are presented in Fig. 5 -Fig. 15. Fig. 5-7 present histograms obtained for one example EEG recording. Histograms were generated for both, power and amplitude spectra, whereas figures present results for power spectra obtained for example electrodes (Fp2,Cz,T4) and for all montages. Fig. 8-9 show maps of EEG power and amplitude spectra. Maps were generated for two of montages: monopolar (M3) and monopolar average (M4). Presented fragments are related to eyes opened, eyes closed and reading. The scale for each single map was unified (for power spectrum 0.4 µV 2 , for amplitude spectrum 0.7 µV). Fig. 9-14 were generated for both, power and amplitude spectra for the selected electrodes (Fp2,Cz,T4,P3). For each user activity fragment was shown for the same scale which was mentioned previously. For each fragment single, the most characteristics waveform was chosen. Alpha waveform was presented for eyes closed, beta 1 waveform was shown for eyes opened and beta 2 was presented for reading.
All results presened in Fig. 3-13 proove that there are visible differences between particular montages. Results visible in Fig. 3-5, obtained for unified scale, indicate that montage M3 (monopolar with ears reference) is the most accurate for observing general spectral analysis results for the whole surface of the head and for all electrodes analysed simultanously. Other montages, especially bipolar, are dedicated rather to analysing particular, usually narrow area of the head. The most visible differences might be noticed for eyes closed fragment, where one can notice differences in waveform distribution among different montages.
More general character of monoplar mongates are confirmed also by results presented in Fig. 6-7, where brain activity maps are presented. These results might be generated only for monopolar montages. Differences in map coloring are due to unified scale obtained for results of both montages. Obtained results, analysed also for other scales (not shown) indicate similar conclusions and no significant differences between activities of particular brain areas. Results of Fig. 10-15 present values of the most characteristic waveforms for each single fragment (eyes opened, eyes closed, reading) for four chosen electrodes. For both, amplitude and power spectrum it might be noticed, that results for M3 montage (monopolar montage with ears reference) occurred to be the most informative. Noticeable differences between electrodes and particular waveforms might also be observed for the M4 montage (monopolar montage with average reference). Bipolar montages occurred to be less significant.

Summary
The paper present the analysis of various kinds of EEG montages and its application to typical EEG data analysis. The case study presented in the paper covered analysis of data obtained as a result of EEG examination of several persons with the age of 23-24 years. Data were gathered with classic EEG record method applying the 10-20 system. The examined conducted several activities: sitting steadily with opened eyes, sitting steadily with closed eyes as well as typical cognitive activity i.e. silent reading of given text. After initial data analysis (filtering, artifact correction), the main analysis was performed. It covered EEG spectra analysis, histogram analysis, brain activity and waveforms comparison done for four EEG montage examples (two bipolar and two monopolar). Presented results show that the most informative and simultaneously the most versatile results were obtained for monopolar montage, especially for the montage with ears reference. Bipolar montages occurred to have some limitations regarding generating brain activity maps. What is more, the results prove that bipolar montages are less useful in general, covering the whole spectrum of electrodes, brain activity.