ICS Dr. G. Roscher GmbH

Recognition of ERP

Home

Products

Projects

Publications

ICS and more

 

A System to Recognise, Estimate and Describe Single Events 

in the Spontaneous Electroencephalogram: 

Example for Single Sweep N1 and P2 Detection

Roscher, G.*, Herrmann, W. M.#, Henning, K.*, Wendt, D.*, Fechner, S.*, Godenschweger, F.*, Weiß, C.*, Abel, E.*, Rijhwani, A.*, Martinez, J.*, Karawas. A.#, Dahan, N.#

 *: ICS Dr. G. Roscher GmbH Magdeburg, Germany

#: Laboratory of Clinical Psychophysiology, Department of Psychiatry (Head: H. Helmchen), Benjamin Franklin Hospital, Free University of Berlin

 

1. Information created by information processes

"Information is Information, not Matter or Energy"

                                                                                                                                                        N. Wiener

Information is created by living organisms through an information process. It can therefore (unlike matter and energy) disappear. Information processes are not necessarily causal. Information is in Jacob's happy phrase "the power to direct what is done" /Jacob, F.: 1973/. 

The development of complex languages is a significant step in the evolution of humans. Speaking and understanding languages is a dynamic process, only realised in this quality of the human brain.  Language is used to describe the outer world and the inner state. In the context of the recognition of the function of the brain we can correlate this description with inner brain state activity. Direct communication in a co-operative manner is the important key to recognising the function of  information processes in man /Roscher, G.: 1987, 1989, 1994/.

Human beings have created technical devices which can carry out goal directed actions, they can carry out information processes, planned by man. Languages are used at high level to transfer human knowledge into technical devices through the use of computers.

 

The direct communication between man and the well designed computer is the innovative way to recognise information processes in man.

 

2. The strategy for psycho physiological experiments to analyse single events

During current psycho-physiological experiments the subject must sit in a dark, electrically and acoustically shielded chamber. The experimenter must use a microphone for communication and video systems for visual observation. The results of the experiment must be carried out on the computer in batch processing after the experiment. Important results could not be discussed with the patient during the experiment and consequently the time course of the experiments could not be modified with respect to these results.

During the course of the tests, it is often useful for the researcher to continue examining the behaviour of the patient, to vary the test in response to individual reactions and previously achieved test results.

Even more precise results can be achieved, if the researcher has access to accurate measurements of the subject's general reactions, heart- and respiration functions, eye-blink frequency and, particularly, EEG readings /Coles, M. G. H.: 1986/, /Lopes da Silva, F. H.: 1986/.

The three components:

- subjective analysis based on direct communication,

- results of psychological tests and

- real time analysis and presentation of signals and reactions, especially of the EEG

together yield an optimal result and may give a better insight to individual single responses.

 

2.1 Technical Specifications

 

The currently constructed system consists of the following components:

1. Amplification System (Verstärkersystem: VS)

2. Multi-Processor System (MPS)

3. Real Time PC (Echtzeit-PC: EPC), Video, Optical-Disc (OD)

4. Evaluation - PC (Auswerte-PC: APC)

5. Patient-PC (Probanden-PC: PPC)

 

1.  The amplification system acquires electrical potentials from the brain of magnitudes ranging in µV. Interesting components, such as Event-Related Potentials (ERP) in the EEG are in the range of  1-10 µV or less. For this application we build up an amplifier system with high precision. The noise voltage is less than 0.5 µV, the common mode rejection is greater than 90 dB.

 

2.  The multi-processor system is capable of carrying out the following tasks:

     Controlling time course of performed stimuli and storing requested reactions of the patient.

     Receiving the EEG data from the amplification system.

     Real-time recognition of EEG-activity.

     Pattern recognition in the EEG and conditionally triggered stimulus output of the patient /Arnold, G. 1991/.

      Implementation of the test procedure through on-line communication with all PC's and the sending of messages concerning the sequence control program.

 

3.  The Real Time PC (EPC) stores the processed data from the experiment on hard or optical disc and holds the distributed database of the system.  Because the EEG data is constantly shown on the screen here. However, the EPC must always be connected to the Multi-Processor System.

 

4.  The Evaluation PC (APC) is the administrators workstation. This is where the actual interaction with the system occurs. This is especially significant during an experiment because the test procedure is controlled from here. The APC is supplied with the latest on-line data from the MPS. It is not requested to process this data immediately because it is able at this time to perform more important tasks, such as statistical evaluation and visualisation of important EEG - activity or reaction and is left free to log out from the Multi-Processor System. Thus there is no real time capability on this PC. This is not however necessary because the Real Time PC records all of the data.

 

5.  The Patient-PC (PPC), which has multimedia capabilities, is used during the experiment to apply stimuli to the subject. It receives all of the requested data on experiment control from the data base shortly before the beginning of an experiment. The time event stimulation is recorded in a program and controlled entirely by the MPS. The Patient PC stores the experiment control data as well as data concerning the results of the experiment.

 

2.2 Psychological testing, using multimedia technology

 

The PC has become established as a readily available and useful computing tool in many applications. A multimedia PC is able to integrate components of text, picture, video and sound in one computer whereby it becomes possible to generate archives and presentations in the aforementioned media. The co-ordination of all these components is ultimately controlled by a relevant software package through which the whole application is developed. These applications can then be used to sequentially control the presentation of visual and acoustic stimuli. Depending on the intentions or requirements of the user the flow of data can be controlled during the running time of a program or generally set beforehand.

The data for stimulation as pictures, sound and video are stored as objects in the database.

The time course of the test can be described in a user friendly manner in a test table. Trial-No., the name of the stimulus object, the required reaction, the duration of the stimulus and the stimulus interval can be written in the table. During the course of the experiment, the experimenter can open a window with a list of test names, click on the test name with the mouse and start the test.

 

2.3 The method of virtual sources estimates for real-time recognition of EEG in amplitude, time and space

 

            "We should make things as simple as possible, but no simpler"               

                                                                                                                        A. Einstein

We developed a method to calculate the electrical field power of a virtual EEG source. We have to call it a virtual source because it gives the exact time of an event but only estimators for the co-ordinates and the field power. These estimators have to be confirmed in subsequent experiments varying the experimental conditions and taking anatomical structures into account using radiological methods.

In order to realise a strategy of psycho-physiological experiments, where an immediate feed back is possible, signal analysis must be carried out in real time.

The FFT and other procedures, based on the FFT have the disadvantage that they only allow statements about a larger time segment of the signal (restriction of the stationary of the signal segment). The system is only capable of statements concerning a segment of time in the past /Beneke, T.: 1994/.

In Event-Related-Potential (ERP) analysis the EEG is interpreted as noise, and only the averaging of many stimulus triggered EEG-samples produce the ERP. During the application of many test events, the status of the subject can change. Observing a series of sweeps one can easily observe single sweeps with large ERPs and others with no detectable ERP response.

The reason for such variability can only be explored when a real time analysis is done, when the subject is simultaneously observed and when the experiment could be interrupted by asking the subjects questions about a particular event.

The availability of high performance computers with graphical colour display gives researchers the capability to represent the distribution of the electrical potential on the head as a map. Such presentations are very impressive and of particular value for researchers, but the maps of an ongoing EEG could not be represented and recognised in real time /Girard, M. H. 1991/.

The same problem emerges in the method of localisation of generators. The algorithms are so complex,  that the now available high performance processors can actualise these algorithms in real time but only with significant latency /Dierks, T.: 1991/, /Gevins, A. S.: 1987, 1988/, /Scherg, M.: 1986/.

The method of virtual sources is a adaptation of the algorithm for localisation and works in the time domain /Goldberg, P.: 1975/.

 

The basic hypothesis for this method of virtual sources is as follows :

 

The same electrical activity in the brain translates consistently to the same electrical activity detected by the electrodes on top of the head and therefore creates the same virtual sources.

 

This strategy led to an extremely short determination of EEG-activity. The lead time was reduced to a matter of milliseconds. The method of virtual sources is based on information theory for:

                    - real-time analysis,

                    - data reduction,

                    - pattern matching and

                    - classification of EEG data in a time and space dynamic.

 

The virtual source represents the following n-tuple:

- space                                                                                                co-ordinate x,

                                                                                                            co-ordinate y,

                                                                                                            co-ordinate z,

- the electrical potential of the activity                                                    p,

- the time point of appearance                                                               t or

- latency after event                                                                               l,

- the duration as a reciprocal of the frequency                                        d,

- the number of electrodes involved in the activity                                   e

and other parameters necessary for the description and identification of the EEG-activity.

In this way the multiprocessor system builds up the ongoing EEG as a sequence of virtual sources. This description is a variant of Lehmanns micro states derived from the ongoing maps /Lehmann D.: 1987, 1991/. The advantage of the sequence of virtual sources is the easy computational handling. This means that virtual sources is a fast way to consistently recognise significant EEG activities in real-time, whereby "significant" means predetermined parameters (filters) such as peaks.

 

2.4 Recognition of EEG-activity using heuristic methods and tools of Fuzzy Logic and of Artificial Intelligence to estimate single Evoked Potential sweeps

 

The advantage of the model of virtual sources is the quick presentation and very good handling by computer. The single EEG activity represented by a virtual source is normally covered by the background noise of the many other electrical processes in the brain. The recognition of these virtual sources has been achieved by using methods of fuzzy logic /Transfertech 1994-1, 1994-2/.

The methods of fuzzy logic give harmonic, gradual transitions in the definition of conditions related to the states of, for example experiments, which is in contrast to the binary (either-or) logic usually used in computer language. In the "world of fuzzy" very bleary evaluations or statements are possible. For example the statement: "This wave is rather like an alpha rhythm" is difficult to translate into classic binary logic. But in terms of fuzzy logic, the user can describe by using a linear transformation:

A wave of 10 Hz, the duration of the half-wave d = 50 msec fit at 100% to alpha rhythm.

A wave of 8 or 12 Hz fit at 50% and of 6 Hz or lesser or of 14 Hz or higher fit at 0%.

This is however the basic rationale behind using fuzzy logic. A wide range of descriptions/parameters are given to represent a statement or phenomenon allowing for more general definitions. Using the technique of fuzzy-logic virtual sources which lie in the range of predefined templates can be searched for and identified in real-time.

These predefined templates are chosen from either the EEG display or the ERP display. The EEG display can be examined stepwise by locating a line cursor and continuously clicking the mouse, each virtual source of the current click can be figured and displayed in a list box. With experience, the research worker can name the virtual sources, make the parameters of the description of the virtual sources fuzzy and store these description under the name in the database. The same steps are possible in an averaged record or an on-line averaged ERP. If the user would like to search for a special pattern in the ERP display, all those virtual sources, which most represent this pattern, should be selected. With the multiprocessor system, the recognition of an ERP or such an EEG pattern can be computed in milliseconds. For more support, the user can start the mapping procedure, to present the distribution of the electrical potential on the brain including the virtual sources.

The templates (selected sequences of virtual sources) have to be selected to best represent the EEG pattern which is intended to be recognised.

The templates for a certain pattern can be stored and utilised for two purposes: either for diagnostic purposes to discover certain patterns or components such as N1/P2 (see fig. 1 and 2), or to trigger stimuli for Evoked or Event Related Potential work.

Using a powerful database system, there is a user-friendly way to train the system to recognise an alpha-rhythm, µ-rhythm, ß-spindles, spikes, K-complexes, eyeblinks, or artefacts.

However, the true value of the system becomes evident, when it is trained to detect and estimate latency and amplitude of a single Evoked Potential sweep, as demonstrated in the attached example (fig. 1).

 

 

 

 

Fig . 1: ERP-Componente N1/P2 in single trials as virtual sources simple reaction test to a tone stimulus

 

Table 1: Numerical description of virtual sources in single trials for N1/P2 (see hand.GIF (969 Byte) Table 1)

 

The figure 1 shows sweep #6 of a 19 lead (10:20 system) EEG record of a healthy subject (alias Katja) under an Evoked Potential experiment. A tone of 1000 Hz was presented, and the subject had to press a mouse button to react. The first line is the trigger for the stimulus onset, and the second line the switch impulse of the mouse button. The sequence of spherical spline maps gives information about the amplitude distribution at an indicated time point after the stimulus trigger impulse. As shown in the sequence of maps at 94 msec there is a field distribution which could relate to N1 and at 168 msec one which could relate to P2.

The N1 and P2 virtual sources are marked as white crosses within the maps. They are also shown in the left, top and front view of the virtual source plotting.

The upper Windows present the numeric expression of the virtual source for N1 and P2, in detail in the following description:

 

            Co-ordinates                                   E l e c t r o d e s

   p        x     y        z        l         d        e      O2     O1     T6     P4     Pz     P3    T5    T4     C4       Cz       C3      T3       F8      F4      Fz          F3      F7       Fp2      Fp1 

-73      19     1     101    23      17      19     -7      -7       -4     -13    -18    -14    -7       -9     -17      -23     -17     -9        -11     -19      -22       -16     -11      -16      -17

 46     -13    -8    108     42      19      19      7       9         7      11     14     11      5        5      15        19      12       2          4       11        11        10        3          4        4

 

One can see that in this case all 19 electrodes are involved in the determination of the virtual sources for N1 and P2. The maximal values in Cz in this trial are:   N1 = -23 µV,   P2 = +19 µV.   The difference:   P2-N1 = +42 µV is named a_diff.

To demonstrate the accuracy of this method the ERP has been evaluated by averaging after artefact rejection of single sweeps against statistical evaluation of the virtual sources in all single trials (sweeps).

 

 

p

x

y

z

l

d

e

a

a_diff.

N1

-27

20

-6

111

23

13

16

-10

 

P2

24

2

-10

118

43

20

16

10

20 µV

 

Values for the virtual sources of the Averaged Evoked Potential

 

 

p

x

y

z

l

d

e

a

a_diff.

N1

-48/22

7/42

-2/21

100/10

22/4

14/7

16/4

-18/7

 

P2

42/23

-5/40

-4/34

103/13

41/6

17/5

15/6

16/8

34 µV

 

Average values and standard deviation of the virtual sources of all single sweeps <mean value>/<std. dev.>

As can be seen from this example, the amplitudes for N1 and P2 are higher if each amplitude in single sweep will be added and the mean amplitude a is given (N1: -18 µV, P2: 16 µV, a_diff.  = 34 µV), if compared to the conventional averaged Evoked Potential (N1: -10 µV, P2: 10 µV, a_diff. = 20 µV). The latencies (l) however are comparable if the single sweep analysis is compared with the averaged Evoked Potential (N1: 23 tacts = 92 msec vs 22 tacts = 88 msec; P2: 43 tacts = 172 msec vs 41 tacts = 164 msec).

Another example is given in figure 2, where sweep # 91 is shown. The stimulus had been triggered based on alpha-activity. The stimulus was given at an alpha-peak /Remond, A.: 1967/.

The reaction time of an alpha triggered stimulus was not significantly different from the stochastical presented stimulus. The ERP's are evaluated in single trials.

 

 

p

x

y

z

l

d

e

a

a_diff.

N1

-39

-69

-5

105

25

12

10

-15

 

P2

53

-23

-15

110

43

17

15

17

32 µV

 

Values for the virtual sources of the Averaged Evoked Potential

 

 

p

x

y

z

l

d

e

a

a_diff.

N1

-74/30

-9/39

-4/16

100/7

25/4

14/4

17/3

-28/10

 

P2

74/30

-4/34

-4/14

100/8

44/5

16/4

16/3

26/10

54 µV

 

Average values and standard deviation of the virtual sources of all single sweeps <mean value>/<std. dev.>

 

In fig. 1 and 2 it is visible, that the ERP-components N1 and P2 are synchronised with the ongoing EEG and the amplitude of the ERP-components is influenced by the EEG. In the case of alpha-triggered ERP, the amplitude is higher than in other cases. This result is in correspondence with the ERP, evaluated in conventional way by averaging /Molenaar, P. C. M.: 1987/.

 

 

 

Fig. 2: ERP-Components N1/P2 in single trials as virtual sources triggered by alpha rhythm (zustandsgetriggert)

 

Table 2: Description of virtual sources in single trials triggered by alpha rhythm N1/P2   (zustandsgetriggert)  (see hand.GIF (969 Byte) Table 2)

 

Tables 1 and 2 show the description of virtual sources in single sweeps.The system presents the following information for further data analysis:

n                  trial number,

p                  power (electrical field power),

x, y, z           co-ordinates (x, y, z),

d                  the duration of the peak detected

                    (d in tact's in 4 msec, d = 12 represents 48 msec duration),

l                   the latency after stimulus in 4 msec units

                    (l in tact's in 4 msec, l = 35 represents 140 msec after the stimulus onset),

e                  the number of Electrodes used for the peak detection (EA), and

a                  the maximal amplitude of the EEG in the electrodes are involved.

 

A sufficient reliable single Evoked Potential Sweep detection is a precondition to correlate parameters of the pre stimulus spontaneous EEG with EP-parameters and the psychological performance of one single sweep /Coppola, R.: 1978/, /Dawson, G. D.: 1947/.

We are currently evaluating an experiment with n = 50 subjects to answer the question whether the pre- stimulus EEG determines the EP/ERP and the psychological performance.

Different N1/P2 and P3 paradigms have been used in this experiment.

The recognition of ERP in a single sweep experiment in real-time, and the possibility for watching and communicating with the subject may give new insight in the information processes in man.

E. Niedermeyer wrote in his introduction:

"This work led us into a 'brave new world' of EEG computerisation and, as early as in 1967, we were told that customary EEG reading would soon be a thing of the past, replaced by a fully automatic EEG interpretation....

It was fond that EEG is by far too complex for such an automation. Its interpretation requires that wonderful computer that is located between the ears." /Niedermeyer, E.: 1994/.

It is not automatic recognition but computer aided analysis, that supports subjective evaluation and psychological tests giving a new approach to the function of the human brain.

The unique property of the human brain evolved through the evolution of language. Speech and the aural recognition of language is a dynamic process. The information is coded in the dynamic of the time course of air pressure fluctuations. It is hypothesised that this dynamic is necessary to use and recognise language and that it is inherent in the higher cognitive functions of the human brain /Basar, E.: 1980, 1988/.

At the beginning of computer science in the early 1960's the information technologists of the time were euphoric about recognising spoken language with computers. Now, 30 years later, the solution is at hand. If information processes in the human brain are coded in the EEG then the task is much harder.  The recognition of information, coded in the dynamics of amplitude, time and space of the EEG is a further challenge which lies with the information technologist.

 

References

 

Arnold, G., Grießbach, G., Kaleta, B.: Implementation of adaptive algorithms for EEG pattern recognition on transputer-aided computers. First International Hans-Berger-Congress, Jena 1991, 21.

Basar, E.: EEG brain dynamics: relation between EEG and brain evoked potentials. Elsevir, Amsterdam 1980.

Basar, E. (Ed.): Dynamics of sensory and cognitive processing by the brain. Springer-Verlag Berlin 1988.

Beneke, T., Fingerling, S., Schwippert, W.: Noch kann es die Natur besser. mc 5/94, S. 106-113.

Coles, M. G. H., Donchin, E., Porges, S. W.: Psychophysiology, Systems, Processes and Applications. Elsevier, Amsterdam, Oxford 1986.

Coppola, R., Tabor, R., Buchsbaum, M. S.: Signal to noise ratio and response variability measurements in single trial evoked potentials. Electroenceph. clin, Neurophysiol., 44 (1978), 214-222.

Dawson, G. D.: Cerebral responses to electrical stimulation of peripheral nerve in man. J. Neurol. Neurosur. Psych., Vol 10 (1947), 134-147.

Dierks, T., Engelhardt, W., Maurer, K.: Dipolquellenberechnung von EEG-Hintergrundaktivität mittels FFT-Approximation und AEP-P300 vor und nach Gabe von Benzodiazepin und deren Antagonisten. Vortrag, 36. Jahrestagung Deutsche EEG-Gesellschaft, Celle 1991.

Gevins, A. S., Remond, A. (Eds.): Methods of analysis of brain electrical and magnetic signals. Handbook of Electroencephalography and clinical Neurophysiology, Revised Series, Vol. 1. Elsevier Amsterdam 1987.

Gevins, A. S., Morgan, N. H., Greer, D. S.: Brain wave source network location scanning method and system. US Patent 4,736,751 12, April 1988.

Giard, M. H., Perrin, F.: Scalp potential and current density mapping: interest for analysis of long-latency auditory ERPs. In: Boelhouwer, A. J. W., Brunia, D. H. M. (Eds.): Proceedings of the first european psychophysiology conference , Tilburg 1991, 27.

Goldberg, P., Samson-Dollfus, D.: A time domain analysis method applied to the recognition of EEG rhythms. In: Dolce, G., Künkel, H. (Eds.): CEAN-Computerized EEG analysis. Fischer, Stuttgart 1975, 19-26.

Jacob, F.: The Logic of Life: A History of Heredity, New York 1973/.

Lehmann, D., Ozaki, H. Pal, I.: EEG alpha map series: brain micro-states by space-oriented adaptive segmentation. Electroenceph. clin, Neurophysiol., 67 (1987), 270-288.

Lehmann, D.: EEG fild maps and mentation: towards the "atoms of thought". First International Hans-Berger-Congress, Jena 1991.

Lopes da Silva, F. H., Strom von Leeuwen, W., Remond, A. (eds.): Clinical applications of computer analysis of EEG and other neurophysiological signals. Handbook of Electroencephalography and Clinical Neurophysiology. Vol. 2, Elsevier Amsterdam, New York, Oxford 1986.

Molenaar, P. C. M., Roelofs, J. W.: The Analysis of Multiple Habituation Profiles of Single Trial Evoked Potentials. Biological Psychology 24 (1987) 1-21.

E. Niedermeyer, F. Lopes da Silva, Electroencephalography. Williams & Wilkins, Baltimore 1994     p. 12

Remond, A., Lesevre, N.: Variations in average visual evoked potentials as a function of the alpha rythm phase ("Autostimulation"). In Cobb, W., Morocutti, C.: The evoced potentials. Electroenceph. clin, Neurophysil., Elsevier, Amsterdam, Suppl. 26 (1967), 43-52.

Roscher, G., Roth, N.: Methodological demands and technical requirements for an effective control of psychophysilogical experiments. In: Kneppo, P., Horsky, I., Kreliöe, I. (Eds.): Proceedings of the 4th IMEKO Conference "Advances in Biomedical Measurements", Bratislava 1987, 294, 294-301.

Roscher, G.: Rechnerunterstützte Arbeit - Neugestaltung von Informationsprozessen mit qualitativem Erkenntnisgewinn, dargestellt an Beispielen der technischen Produktionsvorbereitung bzw. der Durchführung psychophysiologischer Untersuchungen. Dissertation (B), Magdeburg, 30. 9. 1989.

Roscher, G.: Informationsprozesse in technischen Systemen - Modelle und Methoden zur Erforschung des menschlichen Denkens. In Maas, J. F. (Ed.) Das sichtbare Denken - Modelle und Modellhaftigkeit in der Philosophie und den Wissenschaften. 1994, S. 153 - 179.

Scherg, M., Von Cramon, D.: Evoked dipole source potentials of the human auditory cortex. Electorenceph. clin. Neurophysiol. 65 (1986), 344-360.

Transfertech GmbH: Der Fuzzy Control Manager. Firmendokumentation, Braunschweig 1994-1.

Transfertech GmbH: Der Neuro Control Manager. Firmendokumentation, Braunschweig 1994-2.

 

Home

Products

Projects

Publications

ICS and more