CA2051683C - Cerebral biopotential analysis system and method - Google Patents

Abstract

EEG leads are connected to a patient's head (14) by a set of surface electrodes which transmit signals over a patient cable (16) to a 19-channel EEG data acquisition system (12). Data acquisition system (12) filters, amplifies and digitizes the EEG
waveforms and sends the digitized data to the microcomputer (18) via high speed synchronous serial line (26). Microcomputer (18) processes the serial data stream to generate computed data arrays. These arrays are then used in conjunction with predetermined reference arrays derived from clinical studies to produce diagnostic indices.
These indices are displayed on graphics display (20).
Printed output of the diagnostic index is available on the hard copy output device (22) which is connected to the microcomputer (18). Interaction between the operator and the system is provided by means of a keyboard (24) and a pointing device (28) with feedback provided by graphics display (20).

Description

CEREBRAL BIOPOTENTTAL ANALYSTS SYSTEM AND METHOD
CROSS REFERENCE TO RELATED PATENT
The subject matter of this application is related to that of United States Patent 4,907,597, issued March 13, 1990, which is also assigned to the assignee of the present applica-tion.
BACKGROUND OF THE INVENTION
The present invention relates to a real-time, high-frequency, high-resolution cerebral biopotential analysis system and method, and more particularly to a computer-based biopotential diagnostic system and method for quantitatively determining, in a noninvasive manner, cerebral phenomena that can be ascertained by analyzing cerebral electrical activity.
Despite a considerable expenditure of time and effort, current approaches to the quantitative, noninvasive assessment of cerebral electrical activity, as displayed in an "EEG"
waveform, have not been successful in fully extracting all of the information which is present in this complex waveform. A
great need remains for an accurate, sensitive, reliable, and practical neurologic profiling technology. In particular, contemporary intra-operative EEG monitoring techniques have not been widely adopted due to their inherent limitations. Indeed eighty percent (80$) of all medical malpractice suits are believed to be related to post-anesthesia morbidity and mortality, and if such EEG monitoring techniques were reliable, they certainly would have been adopted.

WO 90/i 1718 PCT/US90/01378 2~J~.6~s A number of devices known in the prior art ars capable of tracking cerebral activity qual:.~atively. Techniques in-volving the use of the "classical", conventional analog EEG
are restricted to analyses in the time domain, and require considerable training for adequate interpretation.
Moreover, since the frequency resolution of the human eye at standard speeds and gain is 30 - 60 Hz, much high fre-quency content is invisible. Thus visual EEG assessment is better characterized as being an art rather than a science.
Zn fact, it has been shown that the average correlation be- ' tween seven experienced readers did not exceed 56 per cent.
The use of frequency (power spectrum) analysis of the EEG in the 1960's introduced the notion of some basic processing of the signal prior to visual inspection and led to the application of frequency aaalysis of the EEG to various cerebral monitoring problems. In the past 25 years at least 100 papers have been published in the medical literature describing applications of power spectral ' analysis for purposes such as assessing the depth of anes-thesia and cerebral ischemia under various intraoperative conditions. United States Patent No. 4,557,270 issued to John also describes the use of power spectral analysis to evaluate cerebral perfusion during open heart surgery.
Several recent studios, however, have shown many deficien-cies in the use of power spectral analysis to monitor cerebral perfusion and to determine post operative neurologic outcome. In addition, neither power spectral analysis nor any other monitoring technique has been shown to be reliable, and this is demonstrated by the fact that the well-accepted Hasvard Medical School Anesthesia Monitor-ing Standard does not include any type of intraoperative neurologic monitoring due, in all likelihood, to the com-plexity of interpreting raw EEG data and the unreliability -3- ..
of existing automated systems utilizing power spectral or time-domain analytic techniques.
The discharge of thousands of bio-electrically active cells in the brain, organized in larger, interacting neural centers contributes to the formation of an electrical sig-nal with a wide frequency spectrum and extremely complex dynamics. Embedded in that signal is information regarding frequency content, non-linearities, and phase relationships arising from the complex neuronal firing patterns that take place. Because of the complexity of the EEG signal, conven-tional time and frequency modes of analysis have not been adequate to fully profile its behavior. In the Fourier transforsa of the second order autocorrelation function (the power spectrum) processes are represented as a linear summa-tion of statistically uncorrelated sine-shaped wave com-ponents. Contemporary approaches to monitoring the EEG by means of the power spectrum have thus suppressed informa-tion regarding non-linearities and inter-frequency phase relationships and are of limited utility in representing the EEG's dynamic structure. Furthermore the high frequency low amplitude elements of the EEG have bees discarded to data by the filtering and sampling characteristics of known analysis techniques.
Because the EEG has a wide spectrum and is highly dynamic and non-linear, the phase relationships within the EEG, especially is the higher frequencies, must carry a grant deal of diagnostic information regarding cerebral !unction. The Fouriar transform of the third order autocor-relation function, or autobispectrum, is an analytic process that quantifies deviation from normality, quadratic eon-linaaritiss and inter-frequency phase relationships within a signs!. Ths Fourier transform of the third order crosscorrelation function, or crossbispectrum, is an ~~J~~~a.~
analytic process that provides similar infosznation between two signals.
Autobispectral analytic techniques have been applied to the EEG signal and the basic bispectral properties of the conventional EEG focusing on frequencies below 32 Hz have bees investigated. Such studies have also bean con-ducted to search for changes between waking and sleeping by means of autobispectral analysis. Autobispectral analysis and power spectral analysis have also been used in an at-tempt to show that the EEGs of monozygotic twins are similar in structure.
To date, no previous study has examined the high fre-quency (greater than 32 Hz) content of the EEG and found in-formation of diagnostic value. It also does not appear that any study has shown autobispectral or crossbispectral analysis to be of any value for any diagnostic purpose and certainly neither of these anal~rtic techniques have been shown to have any value in quantifying depth and adequacy of anesthesia, pain responses induced by surgical stress, cerebral ischamia, consciousaess, degrees of intoxication, ongoing cognitive processes or interhemispheric dynamic phase relations.
It is therefore a principal object of the present in-vention to provide a noninvasive high resolution high fre-quency electro-encephalographic system and method capable of recognizing and monitoring physical phenomena that are reflected in cerebral electrical activity.
Another object of the present inveatioa is to provide a aoaiavasive electroeacephalographic system and method capable of datermiaiag sad monitoring depth sad adequacy of anesthesia, pain responses during surgical stress, acute W090/11718 Z~ ~~-~''~~' PCT/US90101378 cerebral ischemia, lev~1 of consciousness, degrees of in-toxication and normal or abnormal cognitive processes.
SUMMARY OF THE INVENTION
Accordingly, the system and method of the present in-vention utilizes a suitable electrode and amplifier system to obtain 19 unipolar EEG signals from regions of interest on both left and right hemispher~s of a subject's brain.
Band-pass filtering of 2 - 500 Hz is ust~d to obtain signals with a high frequency content. High gain amplifiers maxi-mize the dynamic range for the high frequency, lorr energy gave components of the signals. The system applies digital sampling techniques to the signals and transmits digitized data over a high speed serial limo to a host computer. The system divides a 32 second long data aegmant from each lead into 128 consecutive 0.25 second intervals. The system nor-malizes all 19 unipolar leads by the standard deviation, and then characterizes tho dynamic phase relations within the signal by processing for autobispectral variables using either a Fast Fourier Transform (FFT) based approach, or a parametric cubic fitting approach. Similarly three cor-responding left and right hemisphere data pairs are normal-ized in the same manner and dynamic phase relations between two hemispheres are then characterized by processing for crossbispectral estimates utilising either the FFT or parametric based techniques. The outcome is a set of two dimensional arrays representing the dynamic interactions be-tween all the possible combinations of frequencies (frequen-cy pairs) in the spectrum of interest. For each unipolar lead, three arrays are produced: autobicoharence, autobispectral density and autobiphase. Three arrays are also generated for each bipolar data set: crossbicoherence, crossbispectral density and crossbiphase.

Each of the autobispectral and crossbispectral arrays contains 16,512 data points. Although all, or nearly all, of these values can be expected to change from normal during different interventions or due to differing disease states, in the preferred embodiment only those points which show the great-est fidelity in tracking the particular diagnostic determination in question are utilized to create a diagnostic criterion. The ensemble of points most sensitive to a particular intervention or ongoing physiologic process can be used to create a clinically useful single-number index from the computed bispectral arrays. The system uses these indices as a diagnostic figure of merit for the assessment of depth and adequacy of anesthesia, pain responses during surgical stress, acute cerebral ischemia, level of consciousness, degree of intoxication and normal or abnormal cognitive processes. This approach makes it possible for any, even unskilled, operator to meaningfully interpret the output of the diagnostic device.
In situations where continuous monitoring is required, indices can be continuously displayed on a video terminal thereby enabling the operator to interactively evaluate regions of interest. For record keeping purposes, index values and other pertinent variables can be sent to a hard copy output device or stored on a disk.
In accordance with the present invention, there is provided a method of noninvasively detecting cerebral phenomena comprising the steps of: acquiring electroencephalographic signals through at least one electrode from a body surface of a subject being analyzed; filtering said electroencephalo-graphic signals to obtain filtered signals having frequencies -6a-between 2 and 500 hertz; dividing said filtered signals into a plurality of equally sized data records; characterizing dynamic phase relations within said filtered signals by processing said filtered signals to generate bispectral values; comparing said generated bispectral values to reference values to derive a diagnostic index that quantifies the detected cerebral phenomena.
In accordance with another aspect of the invention, there is provided the method of noninvasively detecting cerebral phenomena as defined above where said step of generating said crossbispectral density values comprises the steps of: comput-ing fast Fourier transforms Xi(f) and Yi(f) of said data records i; computing power spectra PXi(f) and PYi(f) of said data records by squaring the magnitude of elements of said fast Fourier transforms Xi(f) and Yi(f) respectively; computing for at least one electrode pair an average complex triple product of all data records acquired by said at least one electrode pair; computing for said at least one electrode pair an average real triple product of all data records acquired by each of said at least one electrode pair; computing for said at least one electrode pair a crossbispectral density value as the absolute value of the average complex triple product for said electrode pair.
In accordance with a further aspect of the invention, there is provided a system for noninvasively detecting cerebral phenomena comprising: means for acquiring electroencephalo-graphic signals through at least one electrode from a body surface of a subject being analyzed; means for filtering said -6b-electroencephalographic signals to eliminate those signals having frequencies less than 2 hertz or frequencies greater than 500 hertz; means for dividing said filtered signals into a plurality of equally sized data records; means for generating bispectral values capable of characterizing dynamic phase relations within said filtered electroencephalographic signals;
means for comparing said generated bispectral values to refer-ence values in order to derive a diagnostic index that quantifies the detected cerebral phenomena.
These and other objects and features of the present invention will be more fully understood from the following detailed description which should be read in light of the accompanying drawings in which corresponding reference numerals refer to corresponding parts throughout the several views.

wo 9oimns 2~ ~~~~~ ~ PCT/US90/01378 BRIEF DESCRIPTION OF THE DRAWING, Fig. 1 is a schematic vier of the system of the present invention for detecting cerebral phenomena in a non-invasive manner;
Fig. 2 is a schematic view of a 19 channel EEG data ac-quisition system including a serial interface utilized in the system of Fig. 1;
Fig. 3 is a schematic viarr of the microcomputer used to calculate and display the EEG bispectrum in the system of Fig. 1;
Fig. 4 is a schematic view of the processing opera-tions pesformed by the system of Fig. l:
Fig. 5 is a flow chart of the operations of the monitor module shops in Fig. 4;
Fig. 6 is a view of a sample display represaatation of bispectral values generated by the system of Fig. 1;
Fig. 7 is a flog chart of the operations of the ac-quisition and EEG rax data management module of the system shoxa in Fig. 4;
Fig. 8 is a floe chart of frequency domain based method for producing autobispectrum or crossbispectrum used by the system of Fig. 1;
Fig. 9 is a flow chart of a parametric based method for producing autobispectrum or crossbispectrum in the sys-tem of Fig. 1;

WO 90/11718 , ' PCT/US90/01378 ~~J~ ~~~ ~ -8-Fig. 10(a) is an illustration of s graph showiag a bispectral density array generated by the system of Fig. 1;
Fig. 10(b) is an illustration of a graph showing a biphase array generated by the system of Fig. 1;
Fig. 10(c) is an illustration of a graph showing a bicoherence array generated by the system of Fig. 1;
Fig. 11 is a flow chart of the diagnostic index genera-tion module shown in Fig. d;
Figs. 12(a) - 12(c) era illustrations of arrays of bispectral density values for three different states of one patient;
Figs. 13(a) - 13(b) are graphs of statistical arrays generated by the system and method of the present invention;
Fig. 14 is an annotated continuous autobispectral den-sity diagnostic index graph for ono load generated by the system of Fig. 1.

W090/11718 ~~ ~~ f~~~ PCT/US90/01378 -9- , . I 1. ,.

Referring to Fig. 1 the apparatus of the present inven-tion includes a 19 channel EEG data acquisition system 12 connected to a microcomputer 18 through a high speed serial interface 26.
The EEG leads are connected to a patient's head 14 by a set of surface electrodes. The International 10/20 electrode system and nomenclature is preferred. The EEG sig-nals are picked up by the electrodes and transmitted over a patient cable 16 to the EEG data acquisition system 12.
The data acquisition system 12 filters, amplifies and digitizes the EEG waveforms and sends the digitized data to the microcomputer 18 via a high speed synchronous serial line 26. In addition, the serial lima 26 can be used to download filtering, gain and sampling rata instructions to the data acquisition unit 12.
The microcomputer 18 processes the serial data stream in order to generate all computed data arrays. Those arrays are then used in conjunction with predetermined reference w arrays derived from clinical studies to produce diagnostic indices which indicate tho status of tho patient. These in-dices are displayed oa the graphics display 20. Printed out-put of the diagnostic index is also available on the hard copy output device 22 which is connected to the microcom-puter 18. Interaction between the operator and the acquisi- ' tion arsd analysis components of the system is provided by mesas of a keyboard 24 and pointing device 28 with feedback on the graphics display 20.
The 19 channel data acquisition system 12 is shown is greater detail in fig. 2. The EEG surface potential, '~~rJ~.E~~

detected by a surface eleetrode mounted on the patient head 14, passes through an electrosurgery protection circuit 30, a defibrillator protection circuit 32, and an amplifier/
filter circuit 36 before being passed on to the multi-chan-nel analog to digital convertor 38.
The electrosurgery protection circuit 30 includes a radio frequency (rf) filter, which limits the rf current through the patient leads 16 to less than 100 microamperes and thus protects the patient 15 from rf burns and proteets the amplifiers 36 from damage resulting from exceeding the absolute maximum input voltage specified by the manufac-turer. This circuit can be an LC section circuit consist-ing of a generic inductor connected in series to a generic capacitor which is then connected to ground.
The defibrillator protection circuit 32 limits the voltage to the amplifiers 36 to a safe level when a defibrillator is applied to the patient 15 and discharged.
This circuit can consist of a neon light bulb and or a ' parallel variable resistor connected in series to a grounded resistor. ..
The amplifier/ filter circuitry 36 is controlled by the microprocessor 34 for default gain and filtering levels or alternate gain and filtering levels as requested by the operator. Preferred gain and filtering settings are dis-cussed later. This circuit section consists of three stages: the first is a pre-amplifier stage that can be as- , .
sembled using a wide variety of high impedance pre-amplifiers such as those sold by National Semiconductor, Sunnyvale G; the second is a programmable filters stage which can utilize components from Frequency Davicas, Haver-hill 1~1; the third stage is a programmable amplifiers stage xhich can be assembled from operational amplifiers used in wo 9oimn8 2~~~fif~3;~ PCT/US90/0137$

conjunction with a multiplying digital to analog (D/A) con-verter both components can be supFlied by National Semicon-ductor. The multiplying D/A is used to aet the gain to the appropriate levels requested by the microprocessor 34.
The high impedance pre-amplifier of each channel will saturate to either the positive or negative supply voltage if the input of the pre-amplifier is not tezminated. This will lead to large positive value or a large negative value at the output of amplifier/ filter section 36. Such value will be used to identify lead failure.
The output of all 19 channels of the amplifier/ filter 36 is fed to a multi-channel analog to digital converter (A/D) 38 which is under microprocessor 34 control for sam-pling rate settings. The analog signals are converted to digital data format suitable for input to a computer. A/D w converters sold by Analog Devices, Norwood 1~ can be used for this purpose.
The multi-channel A/D converter 38 is optically coupled to data bus 40 by optical isolator 42. All control lines to the sample and hold circuits, the multiplexer and the A/D convertor 38 are also optically isolated by optical isolator 44. Any known optical isolators can be used for this purpose.
All DC power lines going to the amplifiers 36, sample and hold circuits, multiplexer and A/D convertor 38 are also isolated from the AC power line with a DC/DC convertor 46 in order to provide complete patient isolation from ground. DC/DC converters available from Burr Brown can be usad.for this purpose.

~~:W.~~;y -12-The basic instructions for controlling operation o!
the microprocessor 34 are stored in a road only memory (ROM) 48. The random access memory (RAM) 50 is used as a buffer memory for data and a portion of the RAM 50 can also b~ used as program memory when a control program is being downloaded from the microcomputer 18.
Serial interface 52 operates under the control of the microprocessor 34. The serial interface 52 is optically coupled with optical isolators 54 to high speed synchronous serial drivers 56 to provide a synchronous serial link be-tureen the 20 channel data acquisition system 12 and any com-patible high speed synchronous serial interface card on any computer. The serial lines are isolated by optical isolators 54 and DC/DC convertor 58 to provide increased patient safety and to protect the host computer 18 from any transients.
Tha host or microcomputer 18 of >3'ig. 1 is shown in greater detail in 13'ig. 3. Tho entire microcomputer system runs under control of a microprocessor 62 with the program memory for the microprocessor 62 being stored in ROM 64.
The RAM 66 is used for storage of intermediate data. The , mass storage device 84 is used for storing clinical databases as wall as archiving patient data.
In a preferred embodiment, the microcomputer 18 con-tains an array processor 68 (such as the Vortez sold by SRY
of Lowall, MA) on which comploz arithmetic calculations can b~ performed oa entire arrays of data simultaneously. The protorred embodiment also includes a math coprocessor 70 which is connected directly to microprocessor 62. Tho math coprocessor 70 is used for scalar and graphic calculations while the array processor 68 is used to calculate bispactral and other data vectors.

wo 90~1171s ~~~~~~~ PCT/US90/01378 ,.
A graphics controller 72 operating under program con-trol of the microprocessor 62 drives a graphics display 20.
A keyboard controller 74 interfaces directly with the operator's keyboard 24. A serial port 80 interfaces with a pointing device 82.
Operator control of the entire acquisition, analysis and display procedure is controlled by the keyboard 24 and pointing device 82 ~rith feedback on the graphics display 20. One high speed synchronous serial port 76 is provided to interface with the 20 channel data acquisition system 12. Port 76 can be used to send control data to the system (e. g., filtering, gain, sampling rate, start/ stop acquisi-tion, perform self diagnostics) and to receive EEG data from the system, as well as to download program data to the system. Another serial or parallol port 78 is provided to drive a hard copy output device 22 for printing desired diagnostic indices.
Referring now to Fig. 4, a block diagram of the system operations and the method of the present invention is described. As meationed above, the aystam and method of the present invention computes dynamic phase and density rela-tions of EEG signals from a preselected number of leads (19 unipolar and 6 bipolar in the described embodiment). Single number diagnostic indices are then generated from the data arrays by utilizing predetermined reference arrays. The results are quantitative iadices useful for analyzing cerebral electrical activity as it relates to, for example, ;..
the assessment of depth and adequacy of anesthesia, pain responses during surgical stress, acute and chronic cerebral ischemia, level of consciousness, degree of ' cerebral intoxication and normal or abnormal cognitive processes.
~.:. ~ . -z,. : . s -:... ~. _-.,,~, <~~. . . .

2051~~~y -14-Tho monitor module 402, handles the overall operations of the system via integration of data and process informa-tion from the user interface module 404, acquisition and rev EEG data management module 406, bispectral processing module 408 and diagnostic index derivation module 410. A
detailed illustration of module 402 can be found in Fig 5.
The user interface and display management module 404 represents the means through which the operator controls and interacts with the system during the course of a proce-dure. This includes, but is not limited to, entry of infer-mation regarding the patient, typo of diagnostic procedure being carried out, lead and acquisition settings; con-tinuous display of acquisition status, lead integrity, and diagnostic indices corresponding to regions probed by each electrode; and requests for printing and archiving results to disk. Module 404 directly interacts with the monitor module 402. Tho operations handled by module 404 can bo achieved under ono of many commercially available environ-ments such as Microsoft's Windows.
The acquisition and raw EEG data management module 406, handles all of the raw EEG data checking and process-ing prior to bispoctral analysis. This includes, but is not limited to, continuous acquisition of EEG data and the verification of the integrity of the data; preparing all unipolar EEG data for autobispectral processing; preparing _ ..
all bipolar EEG data for crossbispoctral processing. Module 406 directly iatoracta with the monitor module 402. A more detailed description of module 406 is provided below in con-nection with fig 9.
The bispoctral processing module 408 controls the generation of all data arrays measuring dynamic phase and :~,~ .

W090/1171$ ~~ ~.~,y.E;~si PCT/US90/01378 density relations within the EEG. This infozmatioa can be organized in both autobispectral sad crossbispectral arrays utilizing either an FFT based or parametric based approach.
The tasks performed by thin module include, but are not limited to: Fourier transformation; and the generation of power spectrum, autobispectral densiWr, crossbispectral den-sity, autobicoherence, crossbicoherence, antobiphase, and crossbiphase. Module 408 directly interacts with the monitor module 402, and a more detailed description of module 408 is provided belor in connection with Figs. 8 and 9.
The diagnostic indez derivation module 410 generates the data values utilized in the diagnostic process. The task includes, but is not limited to, identifying frequency pairs of interest through the use of predetermined clinical reference arrays and creating a diagnostic index from the values in the bispectral data arrays at the frequency loca-tions defined by the reference array. Module 410 directly interacts with the monitor module 402, and a more detailed description of module 410 is provided below in connection with Fig. 11.
Referring now to Fig. 5, the operation of the monitor module 402 will nox be discussed. In step 502, the data ar-rays used to store the digitized EEG, the 128 0.25 second EEG data records, and the bispectral data of each lead are initialized. Tho data files required for storage and files containing data bases required for the computation of ding-nostic indices are also opened is the initializing step 502.
Ia step 504 the system requests the information re-quired to start the acquisition and diagnostic process from the user via the user interface module 404. This requested information includes patient descriptive statistics (sax, ;..... ..Z. ,~~.:~. , ,. , ,.Y.., ..... _.:I9':.'~ .. '~A

2~5~~,~~~
age, clinical symptoms ~tc..), typo of diagnostic procedure to ba coaductod, and the loads used for autobispectral analysis and the leads used for crossbispoctral analysis.
The system includes a default mode of operation and in this default mode the system continuously monitors the depth and adequacy of anesthesia, and any pain responses during surgical stress utilizing a default autobispectral density database. Default band pass filtering is perforsaed from 2 to 500 Hz; the default sampling rate is set at 2000 Hz; and default gain is automatically adjusted to achieve maximum dynamic range in each lead. The following discus-sion of the monitor module 402 will utilize the default set-tings of the system.
The EEG signals measured by leads Fpi, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, p3, Pz, P4, T6, Ol, and 02 (A1 or A2 for reference) era used for autobispectral analysis.
The EEG signals measured from the differbnce of leads F7 and T3 (F7-T3) and the difference of leads F8 and T4 (F8-T4) originate from the area covered by the frontal left hemisphere and frontal right hemisphere regions respective-ly. These signals from F7-T3 and F8-T4 are paired and used for crossbispectral analysis. In this way, the interhomis-pheric relationships for the frontal region can be ex-amined. Similarly, pairing C3-Cz vith C4-Cz and T3-TS with T4-T6 for crossbispectral analysis purposes allows for the examination of the interhamispharic relationships of the oc-cipital and parietal regions respectively.
Zn stop 506, 128 0.25 second buffers of artifact free raw EEG data are acquired. 7U.1 channels transmitting ar-y 2~a'~ ~~3 tifactual data era properly signaled to the operator to cor-rect the problem.
The system, in step 508, computes autobispectral ar-rays for leads Fpl, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, F4, T6, Ol, 02, and crossbispectral ar-rays for leads F7-T3 paired with F8-T4, T3-T5 paired with T4-T6, and C3-Gz paired with C4-Cz. Other leads may, of course, be used in the computation of theca arrays, and two different approaches for bispectral computation will be dis-cussed below with reference to Figs. 8 and 9.
In step 510, the single number diagnostic indices from all generated autobispectral and crossbispectral arrays are computed. Autobispectral density and crossbispectral den-sity clinical reference arrays era utilized in these diag-nostic index computations. The goneration of the reference arrays is discussed later. The system instantaneously dis-plays, in step 512, all computed diagnostic indices for all leads being analyzed. In step 511, the system checks for an exit request, and if such a request has not been made, the system, in step 516, acquires a new 0.25 second buffer and repeats steps 508 through 51'. Ia step 518, requested prin-touts are produced, results are stored to disk for archival purposes sad all files era closed. In step 520, the process is terminated.
A sample display representation generated by the sys-tem is shows is Fig. 6. Represeatations of the patieat's head era shows on the graphics display in Fig 6(a) and Fig.
6(b). The first illustration Fig. 6(a) is divided into nineteen seetions each representing the region probed by an electrode. The second illustration Fig. 6(b) is divided into three horizontal sections representing combined left WO 90/11718 ._ PCT/US90/01378 2~51~;~3;~ -18-and right hemisphere activity probed by the group of electrodes in that region.
For head representation Fig. 6(a), each section con-tains a compressed continuous tracing 602 of the computed diagnostic index utilizing the unipolar EEG data aequired ' ' from the electrode in that site. For head representation Fig. 6(b), each section contains a compressed continuous tracing 604 of the computed crossbispectral diagnostic index utilizing bipolar EEG data acquired from several electrodes in that site.
At the request of the operator any site can be dis-played as an enlarged view 606 for closer examination. The background of the tracing of each site (such as 602 or 604) is color coded to reflect the possible values alloyed for in the range of the selected diagnostic index. The most ' ' currant value of the diagnostic index for that site will dictate what color is displayed in the background (e.g. Red = lowest value to Green = highest value). This will ' facilitate the examination of the patient's status at a dis-tance. Each site will be covered by a large "X" 608 if a lead fail or an artifact was detected for any of the leads contributing to the data required to generate the diagnos-tic index at that site. ' Referring to Fig. 7, the acquisition and raw EEG data management module 406 will now be described in greater detail. Ia stop 702, the system checks whether the 0.25 second buffer for which data is to be acquired is the first buffer being filled for that run, and if it is, the acquisi-tion system 12 in stop 704 is supplied with requested fil-tering, gain, sampling rate and load selection information.
The default sottinga are band pass 2 - 500 Hz for filtor-W090/11718 ~~~~'~ ~~?~~ PCT/US90l01378 -19- . ,..
ing, 50,000 for gain, 2,000 samples/sec for sampling rate and signals from all 19 leads are acquired.
In step 706, the acquisition system 12 acquires data for each 0.25 second buffer for all requested leads and transfers this data to the host computer 18. The system detects lead fai r during the acquisition cycle in step ?08 by checking for very large positive or negative values.
Also in step 708 a publicly available algorithm is used to check for artifact in each lead. In step 710, leads generating failed and artifactual data are marked for the monitor module 402.
In step 712, the system normalizes the records of data acquired from all artifact free leads by subtracting the mean of the samples in each record from each sample in that record, and then dividing the sample by the standard devia-tion of the records. This normalization sets the variance in each record to 1 and has the effect.of weighing each record equally during bispectral averaging. The process is therefore lass dependent on the absolute polder spectral den-sity at any frequency band.
In step 714, each 0.25 second record from each of the leads Fpl, Fp2, F7, F3, Fz, Fd, F8, T3. C3, Cz, C4, T4, T5, P3, Pz, P4, T6, Ol, 02 is assigned to an Xi(t), where Xi(t) are the iadividual time series records provided fos autobispectral processing. Also in step 714, the froatal loft hemisphere time series, Xi(t), from F7-T3 and the fron-tal right hemisphere time series, Yi(t). from FB-T4 era provided for crosabispectral processing. Similarly, by pair-ing loads C3-Cz vith C4-Cz and T3-TS with T4-T6, the cross bispectrum of the left and right occipital and loft and right parietal regions respectively can be processed. It should be anted that for autobispectral analysis Yi(t) is ~G~51.fs8 set to equal X;(t) and in all cases the index i denotes the record number from 1 to 128.
In step 716, a circular buffer mechanism is used for storing the appropriate X;(t) and Y;(t) records for each lead. The buffer is updated by storing the most recently acquired data record in the location of the least recently acquired data record. In step 718, the program checks whether the circular buffer has 128 acquired data records to start bispectral analysis, and if there are 128 data records in the buffer, operation of the system returns to the monitor module 402 in step 720.
Referring now to Fig. 8, the frequency domain based procedures for producing the autobispectrum or the crossbispectrum will nov ba discussed. In step 802, the sys-tem chocks whether the computation to ba performed is an autobispectral or crossbispectral computation.
Autobispectral analysis is a special case of ' crossbispectral analysis and therefore different rules of symmetry apply.
In step 804, the system sets the following symmetries in order to proceed with autobispectral computation:
fl + f2 < N/2 where N ~ 512 (0.25 sacs * 2000 samples in a pieferrad em-bodimant), and 0 < f2 < fl gi~t~ a Yi(t) ___~ Xi(f~ = yi(f~ , WO 90/11718 2~~~~ ~ y PCT/US90/01378 -21- . ._ ~ .
where f1 and fz (also refasrad to as Fz and Fz or Frequency 1 and Frequency 2) denote the frequency pairs over which bispectral computation will ba carried out. Xi(t) and Yi(t) denote the individual time aariaa records used for bispectral computation. Xi(f) and Yi(f) denote the Fourier transform of the time series records and i denotes the record number and in thin embodiment ranges from 1 to 128.
In step 806, the following symmetries are adhered to for crossbispactral analysis:
fl + f2 < N/2 0 < f1 < N/2 0 < f2 < N/2 -2f2 < f1 Xi(t) * Yi(t) ___~ Xi(f) * Yi(f) where all variables represent the same values as they do for autobispactral analysis, except that for crossbiapectral analysis Xi(t) and Yi(t) represent in-dividually derived time aeries records from left and right hemisphere loads respectively.
The fast Fourier transform (FFT) Xi(f) and Yi(f) of esch record of the 128 selected records for that lead, is computed using a standard IEEE library routine or any other publicly available routine in step 808.

WO 90/11718 PCI'/US90/01378 IG~J~.~.2W i Zn Step 810, the power spectra Pxi(f) and Byi(f) of each record of the 128 selected records for that load is computed by squaring the magnitudes of each element of the Fourier transform Xi(f) and Yi(f) respectively.
The system computes the average complex triple product in step 812 by utilizing the following equations where bci(fl.f2) is an individual complex triple product from one record in a given lead and BC(fl.f2) is the average complex triple product for that same lead:
bci (fl. f2) = Xi (fl) * Yi (f2) * Yi (fl+f2) -where Y~(fl+f2) is the complex conjugate of Yi(fl+f2), and BC (fl, f2) ' 128 ~ bCi (f 1. f2) i=1 The average real triple product is computed in stop 814 by using the following equations where bri(fl.f2) is an individual real triple product from ono record in a given load and 8R(fl.f2) is the average real triple product for that same load:
bri (fly f2) i pxi (~1) * pyi (f2) * pyi (fl+f2) HR(fl, f2) ' 128 ~ bri (fl. f2) i = 1 In step 816, the array of auto/crossbispoctral density values (BD(fl,f2)) is computed using the following equation:

WO 90/11718 ~~ ~~ f,~ ~ y PGT/US90/01378 BD (t1. f2) = I BC (fl. f2) I
In step 818, the system computes the array of auto/crossbiphaae values (~(fl,f2)) using the following equation:
~p(f1. f2) = tan-1 [Im(BC (fl, f2) ) /~ (BC (fi. f2) ) ]
0 < ~ < 2n (radians) In step 820, the system computes the array of auto/crossbicoherence values (R(fl.f2)) using the following equation:
R(f1. f2) = BD (f1. f2) / [BR(f1. f2) ] 1/2 0 < R < 1 In step 822, the system returns the requested autocross bispectral density, bicoherence, biphase arrays to the monitor module 402.
Nov turning to Fig. 9, a parametric based method for producing the autobispectrum and the crossbispectrum will now be described. In stops 902, 904, and 906 the system sets the symmetries and time aeries records is the same man-ner as described above in steps 802, 804, and 806 respec-tively. The power spectra of Xi(t) and Yi(t) are estimated is steps 908, 910, and 912. This estimation method includes txo major stages, the Autoregressive (AR) model order selec-tion and the power spectrum computation for Xi(t) and Yi(t). Ia step 908, the system computes two sequences of autocorrelstions. tR2xim)~ and (R2Y(m)~ using the following .
equation.

~~- ~ -24-~a~~~J~i y M N-Iml R2z(m) = M ~N ~ ~ zi(t)zi(t+m), z = X, Y, and m = 0, 1, . . . , L
i=1 t=0 where M is the number of records of each load (128 in our case), and N is the number of samples par record (512 in our case), and L is much greater than the possible AR fil-ter order (we choose 50).
Tha Final Prediction Errors, FPEx(m) and FPEY(m) are calculated fos all orders, m = 1, 2, ... L, by performing a Levinson recursion function on each autocorrelation se-quence in step 910 in order to find the order of the AR fil-ter. The locations, Qx snd QY, of the minimum of FPEx(m) and FPEY(m) respectively are chosen to be the orders of the AR filters of power speqtra of~Xi(t) and Yi(t) respective-ly, i.e., FPEx(Qx) = min ~FPEx(m)} and FPEr(Qy) = min {FPEr(m)}
Once the orders of the 1~R filters for power spectra are chosen, the autocorrelation sequences, (R2x(m)} and (R2y(m}}. are entered into Leviason recursion with order Qx and QY. respectively, instead of L. The coefficients, (cix, i=0, 1, ...,Qx? and (ciY. i = 0,1, ... ,QY}, obtained from the recursion are the coefficients of the 71R filters for poxes spectra o! Xi(t) tad Yi(t) respectively. Then, is step 912, the power spectra Px(f) snd PY(f) are computed as the prediction error (~) divided by square of the mag-nitude of the Fourier transfosm of the coefficients, i.e., ai Pz(~ = Oi , Z = X, Y.
I 1 + ~ Ciz a ~2Z'1 12 i = 1 Tha system estimates the aato/cross bispectrum in steps 914, 916, and 918. The estimation process includes two major stages: the order selection and bispectrum com-putation. In step 914, two seqveacos of third-order mo-ments, (Rgx(T)) and (R3Y(i)) aro computed using the following equation.
M s2 R3z(Z) = M ~N ~ ~ Zi(t)zz(t~L), z = X, Y, and t = -L, . . . , L
i=1 s=sl where s1 = max (1,1-t) , s2 = ~a (N, N-T) , and I. is much groator than the possibly AR filter orders (e. g. 50).
In step 916, two super matrices Tx and Ty are formed as follows.
R3z(-L) R3z(-L+1) ... R3zt0) R3z~-Irl) R3z~-L) ... R3zt 1) z X, Y.
s R3z~-2L) R3 z(-2L+1) ... R3 z{-L) from tho assumption rro made about the AR filter of bispoctrum, the orders Ox and OY of tho AR filtors of bispectra o! Xi(t) ~d Yi(t) era the rsaks of the supor matrices Tx and TY. Thereforo, Ox and OY era choson by using singular value decomposition. Having found tho or-ders, wo obtain the coefficients of the AR filters of ~(~ ~~ ~;;t~~'y bispectra by solving the following linear system of equa-tions:
R3z(0) R3z(1) ... R3z(~z) 1 pz R3z(-1) R3z(0) ... R3z(~z-1) blz 0 _ = X. Y .
R3 z(-~z~ R3z(Wz+1) ... R3z(0) bp=z 0 Where the skewness ((3z) and the coefficients (biz. . . . , bozZ). s = X, Y, can be obtained by solving the linear sys-tem of equations.
The autocross bispectrum of Xi(t) and Yi(t) are com-puted in step 918 as the cubic root of the triple product of the skeWnessas (~i~i~iY)~ divided by the triple product of the Fouriar transforms of the AR filter coefficients (az(f)). i.e., BC (fl, f2) _ (~x~r~Yw~ Hx(f1)H~f2)Hrtf1+f2) oZ
Hz (f ) = 1 + ~ biz a ~Zsti~ . Z = X~ 1,.
i = 1 and BR(fl,f2) is the real triple product for that same lead:
BR(fl,f2) = px(f1) * pz(fz) * pY(f1+f2) After obtaining power spectrum and suto/crosa bispectrum, the system computes the bispectral density array, the biphase, and the bicoherenc~ in step 920 the WO 90/11718 ~~J~~S~3 ~ ~ PCT/US90/01378 _27-same way as in steps 816, 818, 820. Ir step 922, the aystam returns to the monitor module 402 the r~questad bispactral density, biphase, and bicoherence arrays.
For illustration purposes Fig. 10 contains sample autobispectral arrays showing frequency pairs 0 < fl < 128 Hz, and 0 < f2 < 64 Hz. A bispectral density array is shown in Fig. 10(a) where the Z axis represents the magnitude in decibels (db) of the coupled interaction between all ap-propriate frequency pairs fl and f2. Recall that the fre-quency pairing scheme must adhere to the symmetry rule:
fl + f2 < N/2 where H = 256 Hz in this case. A bicoherenca array is shown in Fig. 10(c) where the Z axis represents the normal-ized magnitude in percent (%) of the coupled interaction be-tween all appropriate frequency pairs fl and f2. A biphase array is shown in Fig. 10(b) whore the Z axis represents the phase in radians of the coupled interaction between all appropriate frequency pairs fl and f2.
Referring now to Fig 11. a more detail~d description of the diagnostic index generation module 410 will now ba provided. In stop 1102, the system identifies the typo of diagnostic assasamant in progress. Ia a preferred embodi-m~nt the four possible options are:
1. Depth of anesthesia/ pain Z surgical stress.
2. Cerebral ischamia.
3. Cerebral intoxication (alcohol, narcotics).

4. Cognitive process evaluation.

~~~~f~~.y -28-In step 1104, the system identifies the typo of bispectral array to pass for use in the diagnostic index computation after a user selects a specific lead and array typo as described above with respect to the user interface module 404. There era three (3) poasibla options for each unipolar lead: autobispectral density; autobiphasa;
autobicoherenca. There are also three (3) possible options for each set of bipolar leads: crossbispactral density;
crossbiphase; crossbicoherenca. Since there era 57 (3 X 19 leads) different types of autobispactral and 9 (3 X 3 sets) types of crossbispectral arrays for each one of the 4 diag-nostics, the total number of databases is 264.
In step 1106, the appropriate roferanco array is retrieved from resident memory (or from disk). Each refer-ence array gill coatain the locations of the frequency pairs which era most sensitive to the assessmont is progross (the generation of the roferonco arrays and the selection of defaults will be discussed later). In stop 1108, the system adds all data points in the bispectral array at the locations identified by the retrieved refer-ence array. A counter (NP) of the total number of points added is kopt. In step 1110, the sum of the data points is divided by NP to obtaia the single number diagnostic index.
In step 1112, the program returns to the monitor module 402.
The prodotorminod clinical reference arrays referred to above ire critical to tho device's ability to achieve clinically relevant diagnostic efficacy. Ia the following section we discuss the process adopted for generating these clinical reference arrays. Since a total number of 276 pos-sible reference arrays exist, only one will be discussed in detail. J111 otbor reference arrays ire acquired is a similar fashion. For illustration purposes the generation W0 90/11718 2~ ~~ fT~.~' .4 PCT/US90/01378 -29- _ ._ . .. _ , of the autobispectral density reference array for monitor-ing depth of anesthesia with lead ".'3 will be reviewed.
In a first study EEG potentials from a small group of medically healthy surgical patients (N) with no known neurological disorders are recorded during routine surgery.
The acquisition procedure described previously is followed, with the following exception:
-Band pass filter 0.1 - 500 Hz For all patients, two minutes of artifact free EEG
data era acquired under each of the following conditions:
- Pre-operative: awake ("control") - Deep anesthesia; defined by conventional clini-cal standards (intervention or disease state) - Post-operative: alert in the recovery room (recovery from intervention, or after treatment of disease state) An autobispectral density array is generated for lead T3 from each one of the three recordings for all patients, yielding a total of 3N arrays. The arrays are grouped in 3 sets of N arrays. The first representing the control state, the second representing intervention, and the third repre-aantiaQ recovery.
A paired Student's t test is performed on each of 16,512 data points, comparing the first and second array.
The resulting 16,512 t values are stored in a two dimen-sional array identical is structure to that o! the bispectral deaaity array. A second paired Student's t test is carried out on each of the 16,512 data points, comparing the socond and third arrays. The resulting 16,512 t values ~~:~~~3~i. i -30-era stored in a second txo dimensional array identical in structure to that of the bispectral density array.
All t values not meeting a specific significance test or a specific confidence interval in either array are set to 0. In the preferred embodiment all locations xith a t value not corresponding to a p < 0.0001 are set to 0. Each t value from the first t array (Tl(fl,f2)) is compared with its corresponding t (T2(fi,f2)) fsom the second t array.
Ona of the following conditions must be met:
Tl (fl. f2) < 0 < T2 (fi. f2) or T2 (fl. f2) < 0 < Tl (fl, f2) If neither one the two conditions is mat at a particular frequency pair fl, f2 then Ti (fl, f2) = 0 and T2 (fl, f2) = 0.
Tho application of the above conditions has the effect of identifying all of the frequency pair locations that change significantly by shoxing a consistent increase in bispectral density value xith anesthesia followed by a decrease xith recovery, or a decreaao xith anesthesia fol-loxed by an iacroaso xith recovery.
rinslly, the absolute values o! the t values in each 11,12 location from the first t array are added to their counterpart is the second t array to form a third t array.
The third t array is an average of the first txo and can be visually inapoctod for highly aonsitive regions.
The last step involves sorting the third t array for the most sensitive ensemble of frequency pair locations. In the prelersed embodiment this xould consist o! the top 25%

1V090/11718 ~~ ~~ ~~';~ y PC1'/US90/01378 of all t values. The locations fl,ty of the most sig-nificant t values meeting all of tho above conditions era stored in resident memory (or oa disk) as one of the predetermined reference arrays. This reference array will be accessed by the diagnostic index derivation module 410, for autobispectral density diagnostic index generation during anesthesia/surgery for the location probed by lead T3.
For any particular diagnostic task and any particular lead there are 6 possible bispectral arrays (autobicoherence, autobispectral density, autobiphase, crossbicoherence, cross bispectral density, and crossbiphase) which could be oxamined for diagnostic poten-cy. To rank order the reference arrays with respect to diagnostic efficacy a second prospective study is con-ducted. Tha conditions under which the study is conducted are identical to those of the first except that: a) the fre-quency pair locations of interest have already bean iden-tified and era now followed prospectively and b) the size of the study group is now sufficiently large so that sample variation of bispectral arrays more closely approximates the true variance within the population undergoing the in-tervention or suffering from the disease.
Thus for the example of anesthesia monitoring the EEG
recording starts during the awake/control state and con-tinues uninterrupted through the end of recovery. Con-tinuous surgical notes are maintained throughout the operation.
After the completion of the study, continuous diagnos-tic indices are generated for the le:ds of interest for each of auto or cross bispoctral density. biphise ind bicoherence arrays. The continuous trends are annotated ~~~~~1 7 with the intraoperative notes. A sufffciantly largo group of prospective patients (determined by a statistical power test) is used to determine which continuous diagnostic index exhibits the greatest diagnostic ~fficacy on clinical grounds. The particular bispectral array used to generate this best diagnostic index during a particular diagnostic procedure is programmed into the system as the default array for diagnostic assessment.
The following non-limiting example is provided solely for illustrative purposes. Twenty (20) patients undergoing elective surgery for a variety of orthopedic and gynecologic conditions were studied. Standard EEG leads were placed in 16 locations according to the International 10/20 system. Raw EEG signals were acquired, bawd-pass fil-tered (0.1 - 110 Hz) and digitised at a sampling rate of 256 Hz. EEG recordings were obtained from all patients prior to the induction of anesthesia. Patients were than anesthetized using standard techniques with a variety of anesthetic agents. Continuous EEG recordings were obtained during the period of anesthesia induction until the patient was judged to be adequately anesthetized for surgery by clinical assessment. Intermittent EEG recordings ware then obtained during the course of the operation. During the period of recovery from anesthesia another continuous EEG
recording was taken. A final recordiag was obtained when the patient was deemed to be "awake" in the recovery room.
Detailed clinical intra-operative notes of patient status were maintained during all phases o! EEG recording for sub-sequent correlation with bispectral parameters.
In 10 patients the entire available frequency spectrum (0.1 to 110 HZ) was examined for statistically aigaificant changes in autobispectral density values from the awake state to the deeply anesthetized state and back to the W090/11718 ~~'~~'~~'~ PCT/US90/01378 awake state. Figs. 12(a)-12(c) show average bispoctral den-sity arrays (from 10 patients) for each of the three states of consciousness. The method for determining statistical significance was as outlined above. Figs. 13(a)-13(b) show the statistical arrays generated by the technique of the present invention: the average t array for these 10 sub-jects for lead T3 for locations corresponding to a p < 0.05 (t > 2.26) (Fig. 13(a)) and the average t array for loca-tions corresponding to a p < 0.0000003 (t > 10.0) (Fig.
13(b)). Each t value in the array reflects tba consistency of change in a bispectral density value through the three states for one frequency pair location across all 10 patients. Zt is north noting that virtually no data points are significant with a p < 0.0000003 (t > 10.0) in the fre-quency pair band of F1 below about 24 Hz and F2 below about 2 Hz in Fig. 13(b). On the other hand 7,168 locations were found to change with a p < 0.0000003 (t > 10.0) in the fre-quency pair band of F1 above about 24 Hz and F2 above about 2 Hz. The top 25% (7,168/4 = 1,792 points) most sig-nificant high frequency locations worn used to define the reference array. The autobispectral density index was cal-culated for each subject from the points defined by the ref-erence array as described above is the detailed description of the invention. This autobispactral density index was , then calculated as a continuous function for the continuous EEG recordings to assess its behavior during induction, intra-op~rativoly, and during recovery. Tha correlation with clinical events during the operation ryas noted.
A sample annotatQd continuous autobispectral density diagnostic index for lead T3 during surgery in a prospec-tive subject is shown in Fig. 14. Tha index varies between 30 and 5 decibels and is qvit~ sensitive to the patient's state of consciousness and the onset of painful stimuli.
Specifically, the index drops with the induction of saes-~~J~.E~~3s thetic agents (pentothal and othrano) to the patient, and the index level rises as the patient's lag is being prepped for surgery. Zn addition, the index approaches its highest value when the patient is awake in the recovery room and most likely experiencing post-operative stress. (The gaps seen in the index plot correspond to time periods when EEG
recordings mere not being taken).
Similarly the above analytic process is used to generate the reference databases for cerebral ischemia, degrees of intoxication and normal or abnormal cerebral processes. In quantitatively detecting any of these cerebral phenomina, the system compares a number of autobispectral and crossbispectral EEG data from subjects in the normal state to clinically identified extremes of a certain physiologic state (awake vs anesthetized, sober vs intoxicated, perfused vs ischemic, at rest mentally vs thinking, normal vs retarded, etc..). Tho comparison util-izes a statistical approach to identify the bispoctral data poiats that are most sensitive to the particular physiologic state in question. Tho frequency pair loca-tions of the most sensitive data points are identified and stored in a database for reference purposes. When a diag-nosis is to be carried out, the average of all the data points defined by the reference array is obtained for the subject undergoing the study. This average is used as a diagnostic index cad is compared to a list of indices char-acteristic o! etch state by the operator or the system.
In addition to quantifying the depth and adequacy of anesthesia, pain responses during surgical stress, acute and chronic cerebral ischemia, level of consciousness, de-gree of cerebral intoxication and normal or abnormal cogni-tive processes, the system and method of the preseat invention may also be used to assess s myriad of cerebral 2~5'~ f ~'3 phenomena based on the acquisition and processing o! BEG
signals into various bispoctral arrayb which are than com-pared to appropriate reference arrsys.
Although bispectral analytic techniques in the frequen-cy domain have been applied to the EEG signal, as was dis-cussed in the Background above, parametzic approaches to the estimation of bispectral values have not. Furthermore no bispectral technique has aver been demonstrated to be useful for any diagnostic purpose. Other techniques for the quantification of the depth of anesthesia or the detection and quantification of cerebral ischemia intraoperatively remain qualitative and limited in their overall utility and acceptance in practice. Specifically, the system and method of the present invention examines various bispectral values across all frequency pairs in a frequency range hitherto ig-pored by those knowledgeable in the art and uses the summed degree of changes is autobicoherance/autobispectral dan-sity/autobiphaso, crossbicoheronce/crossbispoctral don-sity/crossbiphase at a limited number of frequency locations as an index of physiological perturbation. The system and method utilize various bispectral arrays of defined clinical populations to define the locations of the subset of frequencies used to calculate this index. Refer-ence clinical arrays are further utilized to assess the meaning o! this index and to measure the aignificaaco of deviations of this indes from nosmality. This allows the quantitative gauging of the disturbances in cerebral func-tion, whether due to anesthesia, intoxicants or ischemia w for any particular EEG lead position. The system and method disclosed heroin also define the graphic display o! the diagnostic index, whether oa graphics screQa or on paper, whether is real-time or in digital archive.

jt~''~3.~~~.~.~~'~, i ifhila the foregoing invantios has boon described with reference to its preferred ambodimanta, various alterations and modifications will occur to those skilled in the art.
A11 such alterations and modifications are intended to fall within the scope of the appended claims.

Claims (70)

THE EMBODIMENTS OF THE INVENTION IN WHICH AN EXCLUSIVE
PROPERTY OR PRIVILEGE IS CLAIMED ARE DEFINED AS FOLLOWS:

1. A method of noninvasively detecting cerebral phenomena comprising the stops of:
acquiring electroencephalographic signals through at least one electrode from a body surface of a subject being analyzed;
filtering said electroencephalographic signals to obtain filtered signals having frequencies between 2 and 500 hertz;
dividing said filtered signals into a plurality of equally sized data records;
characterizing dynamic phase relations within said filtered signals by processing said filtered signals to generate bispectral values;
comparing said generated bispectral values to reference values to derive a diagnostic index that quantifies the detected cerebral phenomena.

2. The method of noninvasively detecting cerebral phenomena of claim 1 wherein the step of acquiring electroencephalographic signals further comprises the step of attaching electrodes to a head of the subject being analyzed in order to obtain a unipolar electroencephalographic signal from each region of interest of both left and right hemispheres of the subject's brain to which said electrodes are attached.

3. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values generated in said step of characterizing said dynamic phase relations are autobispectral density values.

4. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values generated in said step of characterizing said dynamic phase relations are autobicoherence values.

5. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values generated in said step of characterizing said dynamic phase relations are autobiphase values.

6. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said step of acquiring electroencephalographic signals further comprises the step of attaching electrodes to the head of the subject being analyzed in order to obtain bipolar data sets of electroencephalographic signals from left tad right hemispheres of the subject's brain to which said electrodes are attached.

7. The method of noninvasively detecting cerebral phenomena of claim 6 wherein one bipolar data sot is acquired from a frontal left hemisphere of the subject's brain and another bipolar data set is acquired from a frontal right hemisphere of the subject's brain.

8. The method of noninvasively detecting cerebral phenomena of claim 6 wherein one bipolar data set is acquired from a left occipital region of the subject's brain and another bipolar data set is acquired from a right occipital region of the subject's brain.

9. The method of noninvasively detecting cerebral phenomena of claim 6 wherein one bipolar data set is acquired from a left parietal region of the subject's brain and another bipolar data set is acquired from a right parietal region of the subject's brain.

10. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values generated in said step of characterizing said dynamic phase relations are crossbispectral density values.

11. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values generated in said step of characterizing said dynamic phase relations are crossbicoherence values.

12. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values are generated in said step of characterizing said dynamic phase relations are crossbiphase values.

13. The method of noninvasively detecting cerebral phenomena of claim 3 where said step of generating autobispectral density values comprises the steps of:
computing fast Fourier transforms X i(f) and Y i(f) of each of said data records i;
computing polder spectra P Xi(f) and P Yi(f) of said data records i by squaring the magnitude of each element of said fast Fourier transforms X i(f) and Y i(f) respectively;
computing for at least one electrode an average complex triple product of data records acquired by said at least one electrode;
computing for said at least one electrode an average real triple product of data records acquired by said at least one electrode;
computing for said at least one electrode an autobispectral density value as the absolute value of said average complex triple product for said electrode.

14. The method of noninvasively detecting cerebral phenomena of claim 13 further comprising the step of computing an autobiphase value ~(f1,f2) for at least one electrode such that:
~ (f1, f2) = tan-1 [Im (BC(f1, f2))/Re(BC(f1, f2))]
where BC(f1,f2) is the average complex triple product for an electrode, and f1 and f2 designate limits of the frequency range over which the autobiphase computation is carried out.

15. The method of noninvasively detecting cerebral phenomena of claim 13 further comprising the step of computing an autobicoherence value R(f1,f2) for at least one electrode such that R(f1, f2) = BD (f1, f2) / [BR(f1, f2)]1/2 where BD(f1,f2) is the autobispectral density value for said electrode, BR(f1,f2) is the average real triple product for the same electrode, and f1 and f2 designate limits of the frequency range over which the autobicoherence computation is carried out.

16. The method of noninvasively detecting cerebral phenomena of claim 10 where said stop of generating said crossbispectral density values comprises the steps of:
computing fast Fourier transforms X i(f) and Y i(f) of said data records i;
computing power spectra P Xi(f) and P Yi(f) of said data records by squaring the magnitude of elements of said fast Fourier transforms X i(f) and Y i(f) respectively;
computing for at least one electrode pair an average complex triple product of all data records acquired by said at least one electrode pair;
computing for said at least one electrode pair an average real triple product of all data records acquired by each of said at least one electrode pair;
computing for said at least one electrode pair a crossbispectral density value as the absolute value of the average complex triple product for said electrode pair.

17. The method of noninvasively detecting cerebral phenomena of claim 16 further comprising the step of computing a crossbiphase value ~(f1,f2) for said at least one electrode pair such that:
~(f1, f2) = tan-1 [Im(BC(f1,f2))/R~(BC(f1,f2))]
where BC(f1,f2) is the average complex triple product for an electrode pair, and f1 and f2 designate limits of the frequency range over which the crossbiphase computation is carried out.

18. The method of noninvasively detecting cerebral phenomena of claim 17 further comprising the step of computing a crossbicoherence value R(f1,f2) for said at least one electrode pair such that R(f1, f2) = BD (f1, f2) / [BR(f1, f2)) 1/2 where BD(f1,f2) is the crossbispectral density value for an electrode pair, BR(f1,f2) is the average real triple product for the same electrode pair, and f1 and f2 designate limits of the frequency range over which crossbicoherence computation is carried out.

19. The method of noninvasively detecting cerebral phenomena of claim 3 wherein said step of generating autobispectral density values comprises the steps of:
computing autocorrelation sequences R2x(m) and R2Y(m) of all data records acquired by at least one electrode;
determining the orders and coefficients of parametric models for power spectra of data records acquired by said at least one electrode;
computing power spectra P X(f) and P Y(f) of data records acquired by said at least one electrode;
computing third order moment sequences R3x(t) and R3Y(~) of data records acquired by said at least one electrode;
determining the orders and coefficients of parametric models of the bispectra of data records acquired by said at least one electrode;
computing for said at least one electrode a bispectrum of data records acquired by said at least one electrode.

20. The method of noninvasively detecting cerebral phenomena of claim 19, wherein said bispectrum is autobispectrum and further comprising the step of computing an autobispectral density value for at least one electrode as the absolute value of the bispectrum of all data records for said electrode.

21. The method of noninvasively detecting cerebral phenomena of claim 19 further comprising the step of computing an autobiphase value ~(f1,f2) for said at least one electrode such that:
~(f1, f2) = tan-1 [Im(BC(f1, f2))/Re(BC(f1,f2))]
where BC(f1,f2) is the bispectrum for an electrode, and f1 and f2 designate limits of the frequency range ever which the bispectral computation is carried out.

22. The method of noninvasively detecting cerebral phenomena of claim 20 further comprising the steps of:
computing for at least one electrode a real triple product of data records acquired by said at least one electrode;
computing an autobicoherence value R(f1,f2) for said at least one electrode such that R(f1, f2) = BD(f1,f2)/[BR(f1,f2))]1/2 where BD(f1,f2) is the autobispectral density value for an electrode, BR(f1,f2) is the real triple product for the same electrode, and f1 and f2 designate limits of the frequency range over which bispectral computation is carried out.

23. The method of noninvasively detecting cerebral phenomena of claim 19 wherein said bispectrum is crossbispectrum and further comprising the step of computing a crossbispectral density value for each electrode pair as the absolute value of the crossbispectrum of all data records for each said electrode pair.

24. The method of noninvasively detecting cerebral phenomena of claim 23 further comprising the step of computing a crossbiphase value ~(f1,f2) for each of said at least one electrode pair such that:
~ f1, f2) = tan 1 [Im(BC(f1, f2))/Re(BC(f1,f2))]
where BC(f1,f2) is the crossbispectrum for an pair, and f1 and f2 designate limits of the frequency range over which the bispectral computation is carried out.

25. The method of noninvasively detecting cerebral phenomena of claim 24 further comprising the step of computing a crossbicoherence value R(f1,f2) for each of said at least one electrode pair such that R(f1, f2) = BD(f1,f2)/[BR(f1, f2)] 1/2 where BD(f1,f2) is the crossbispectral density value for an electrode pair, BR(f1,f2) is the real triple product for the same electrode pair, and f1 and f2 designate limits of the frequency range over which bispectral computation is carried out.

26. The method for noninvasively detecting cerebral phenomena of claim 1 wherein said step of acquiring electroencephalographic signals further comprises the step of analyzing said signals to determine lead failure.

-46-~

27. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said stop of comparing further comprises the steps of:
organizing said generated bispectral values is at least one array of bispectral values;
selecting a physical phenomena to be diagnosed;
retrieving an appropriate bispectral reference array from a resident memory, said reference array containing frequency pairs that are most sensitive to the physical phenomena to be diagnosed;
adding data values in locations of each of said at least one array of bispectral values that are identified by the retrieved reference army is being locations containing data of significance to obtain a sum of said significant locations;
averaging the values stored is said significant locations to generate a diagnostic index relating to the cerebral phenomena to be detected.

28. The method of noninvasively detecting cerebral phenomena of claim 1 further comprising the steps of:
generating three arrays of bispectral data for each of three different states of the subject;
performing a paired Student's t test comparing data in a first and a second array of said three arrays of bispectral data to produce a first t array and performing a paired student's t test comparing data in said second and a third array of said three arrays of bispectral data to produce a second t array;
comparing data values in said first t array with data values in corresponding locations in said second t array;
identifying those corresponding locations in said first and second t arrays that differ by more than a preselected amount, said identified locations representing those locations that are significant for detecting the cerebral phenomena.

29. The method of noninvasively detecting cerebral phenomena of claim 1 wherein the cerebral phenomenon being detected is the depth of anesthesia in the subject being analyzed.

30. The method of noninvasively detecting cerebral pheomena of claim 1 wherein the cerebral phenomena being detected are pain responses during surgical stress in the subject being analyzed;

31. The method of noninvasively detecting cerebral phenomena of claim 1 wherein the cerebral phenomenon being detected is acute ischemia or infarction in the subject being analyzed.

32. The method of noninvasively detecting cerebral phenomena of claim 1 wherein the cerebral phenomenon being detected is the level of consciousness of the subject being analyzed.

33. The method of noninvasively detecting cerebral phenomena of claim 1 wherein the physical phenomenon being detected is the degree of cerebral intoxication of the subject being analyzed.

34. The method of noninvasively detecting physical phenomena of claim 1 wherein the physical phenomena being detected are normal or abnormal cognitive processes.

35. A system for noninvasively detecting cerebral phenomena comprising:
means for acquiring electroencephalographic signals through at least one electrode from a body surface of a subject being analyzed;
means for filtering said electroencephalographic signals to eliminate those signals having frequencies less than 2 hertz or frequencies greater thaw 500 hertz;
means for dividing slid filtered signals into a plurality of equally sized data records;
means for generating bispectral values capable of characterizing dynamic phase relations within said filtered electroencephalographic signals;
means for comparing said generated bispectral values to reference values in order to derive a diagnostic index that quantifies the detected cerebral phenomena.

36. The system for noninvasively detecting cerebral phenomena of claim 35 further comprising a plurality of said means for acquiring electroencephalographic signals, each of said means for acquiring electroencephalographic signals being connected to said means for filtering.

37. The system for noninvasively detecting cerebral phenomena of claim 36 wherein said plurality of said means for acquiring electroencephalographic signals is a plurality of electrodes attachable to a head of a subject being analyzed to obtain a unipolar electroencephalographic signal from each of a plurality of regions of interest on both left and right hemispheres of the subject's brain.

38. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said means for acquiring electroencephalographic signals comprises:
a plurality of surface electrodes for mounting on a surface of a head of the subject being analyzed;
means for providing electrosurgery protection including a radio frequency filter for limiting radio frequency current through said electrodes;
means for providing defibrillator protection for limiting voltage to said amplifier during a discharge;
means for amplifying said filtered signals for a high gain in order to maximize the dynamic range for high frequency, low energy wave components of said filtered signals;
means for feeding said signals to an analog-to-digitial converter to convert said signals to digital signals.

39. The system for noninvasively detecting cerebral phenomena of claim 35 further comprising means for analyzing said signals received by each electrode in order to detect electrode failure.

40. The system for noninvasively detecting cerebral phenomena of claim 35 Wherein said bispectral values are autobispectral density values and further comprising means for organizing said autobispectral density values in at least one array of autobispectral density values.

41. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said bispectral values are autobicoherence variables and further comprising means for organizing said autobicoherence values in at least one array of autobicoherence values.

42. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said bispectral values are autobiphase values and further comprising moans for organizing said autobiphase values in at least one array of autobiphase values.

43. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said bispectral values are crossbispectral density values and further comprising means for organizing said crossbispectral density values in at least one array of crossbispectral density values.

44. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said bispectral values are crossbicoherence values and further comprising means for organizing said crossbicoherence values is at least one array of autobicoherence values.

45. The system for noainvasively detecting cerebral phenomena of claim 35 wherein said bispectral variables are crossbiphase values and further comprising means for organizing said crossbiphase values in at least one array of autobiphase values.

46. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said means for acquiring encephalographic signals further comprises means for obtaining bipolar data sets of electroencephalographic signals from different regions of a brain of said subject.

47. The system for noninvasively detecting cerebral phenomena of claim 40 where said means for generating at least one array of autobispectral density values comprises:
means for computing fast Fourier transforms X i(f) and Y i(f) of each of said data records i;
means for computing power spectra P xi(f) and P Yi(f) of said data records i by squaring the magnitude of elements of said fast Fourier transforms X i(f) and Y i(f) respectively;
means for computing for said at least one electrode as average complex triple product of data records acquired by said at least one electrode;
means for computing for said at least one electrode an average real triple product of data records acquired by said at least one electrode;
means for computing for said at least one electrode an autobispectral density value as the absolute value of the average complex triple product for said electrode.

48. The system of noninvasively detecting cerebral phenomena of claim 47 further comprising means for computing an autobiphase value .PHI.(f1,f2) for said at least one electrode such that:
.PHI. (f1, f2) = tan -1 [Im (BC (f1, f2)) /Re (BC (f1, f2))]
where BC(f1,f2) is the average complex triple product for an electrode, and f1 and f2 designate limits of the frequency range over which the autobiphase computation is carried out.

49. The system for noninvasively detecting cerebral phenomena of claim 47 further comprising means for computing an autobicoherence value R(f1,f2) for said at least one electrode such that R(f1,f2) = BD(f1,f2)/[BR(f1,f2)]1/2 where BD(f1,f2) is the autobispectral density value for said electrode, BR(f1,f2) is the average real triple product for the same electrode, sad f1 and f2 designate limits of the frequency rangy over which the autobicoherence computation is carried out.

50. The system for noninvasively detecting cerebral phenomena of claim 40 where said means for generating at least one array of crossbispectral density values comprises:
means for computing fast Fourier transforms X i(f) and Y i(f) of each of said data records i;
means for computing power spectra P xi(f) and p Yi(f) of said data records by squaring the magnitude of elements of said fast Fourier transforms X i(f) and Y i(f) respectively;
means for computing for at least one electrode pair an average complex triple product of all data records acquired by each of said at least one electrode pair;
means for computing for said at least one electrode an average real triple product of all data records acquired by for each of said at least one electrode pair;
means for computing for said at least one electrode pair a crossbispectral density value as the absolute value of the average complex triple product for said electrode pair.

51. The system for noninvasively detecting cerebral phenomena of claim 50 further comprising means for computing a crossbiphase value .PHI.(f1,f2) for said at least one electrode pair such that:
.PHI.(f1,f2) = tan -1 [Im (BC (f1, f2) ) /Re (BC (f1, f2))]
where BC(f1,f2) is the average complex triple product for an electrode pair, and f1 and f2 designate limits of the frequency range over which the crossbiphase computation is carried out.

52. The system for noninvasively detecting cerebral phenomena of claim 51 further comprising means for computing an crossbicoherence value R(f1,f2) for said at least one electrode pair such that R(f1, f2) = BD (f1, f2) / [BR(f1, f2)] 1/2 where BD(f1,f2) is the crossbispectral density value for an electrode pair, BR(f1,f2) is the average real triple product for the same electrode pair, and f1 and f2 designate limits of the frequency range over which the crossbicoherence computation is carried out.

53. The system for noniavasively detecting cerebral phenomena of claim 40 wherein slid means for generating at least one array of autobispectral density values comprises:
means for computing autocorrelation sequences R2X(m) and R2Y(m) of all data records acquired by at least one electrode;
means for determining the orders and coefficients of parametric models for power spectra of data records acquired by said at least one electrode;
means for computing power spectra P X(f) and P Y(f) of all data records acquired by said at least one electrode;
means for computing third order moment sequences R3X(t) and R3Y(t) of data records acquired by said at least one electrode;
moans for determining the orders and coefficients of parametric models of the bispectra of data records acquired by said at least one electrode;
means for computing for said at least one electrode a bispectrum of data records acquired by said at least one electrode;

54. The system for noninvasively detecting cerebral phenomena of claim 53 wherein said bispectrum is autobispectrum and further comprising means for computing an autobispectral density value for each electrode as the absolute value of the bispectrum of data records for said electrode.

55. The system for noninvasively detecting cerebral phenomena of claim 53 further comprising means for computing an autobiphase value .PHI.(f1,f2) for at least one electrode such that:
.PHI.(f1, f2) = tan -1 [Im(BC (f1, f2) ) /Re (BC(f1, f2))]
where BC(f1,f2) is the bispectrum for an electrode, and f1 and f2 designate limits of the frequency range over which the autobiphase computation is carried out.

56. The system for noninvasively detecting cerebral phenomena of claim 54 further comprising:
means for computing for at least one electrode a real triple product of all data records acquired by said at least one electrode;
means for computing an autobicoherence value R(f1,f2) for said at least one electrode such that R(f1, f2) = BD (f1, f2) / [BR(f1, f2)] 1/2 where BD(f1,f2) is the autobispectral density value for an electrode, BR(f1,f2) is the real triple product for the acme electrode, and f1 and f2 designate limits of the frequency range over which autobicoherence computation is carried out.

57. The system for noninvasively detecting cerebral phenomena of claim 53 wherein said bispectrum is crossbispectrum and further comprising means for computing a crossbispectral density value for an electrode pair as the absolute value of the bispectrum of data records for said electrode pair.

58. The system for noninvasively detecting cerebral phenomena of claim 53 further comprising means for computing a crossbiphase value .PHI.(f1,f2) for at least one electrode pair such that:
.PHI.(f1, f2) = tan -1 [Im(BC(f1, f2))/Re (BC(f1, f2))]
where BC(f1,f2) is the crossbispectrum for an electrode pair, and f1 and f2 designate limits of the frequency range over which the crossbiphase computation is carried out.

59. The system for noninvasively detecting cerebral phenomena of claim 57 further comprising:
means for computing for at least one electrode a real triple product of all data records acquired by said at least one electrode;
means for computing a crossbicoherence valve R(f1,f2) for said at least one electrode pair such that R(f1, f2) = BD (f1, f2) / [BR(f1, f2)] 1/2 where BD(f1,f2) is the crossbispectral density value for an electrode pair, BR(f1, f2) is the real triple product for the same electrode pair, and f1 and f2 designate limits of the frequency range over which the crossbicoherence computation is carried out.

60. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said means for comparing further comprises:
means for organizing said generated bispectral values in an array of bispectral variables;
means for selecting a physical phenomena to be diagnosed;
means for retrieving an appropriate bispectral reference array from a resident memory, said reference array containing frequency pairs that are most sensitive to the physical phenomena to be diagnosed;
means for adding data values in locations of each of said at least one array of bispectral values that are identified by the retrieved reference array as being locations containing data of significance to obtain a sum of said significant locations;
means for averaging the values stored in said significant locations to generate a diagnostic index relating to the cerebral phenomena to be detected.

61. The system for noninvasively detecting cerebral pheomena of claim 35 further comprising:
means for generating three arrays of bispectral data for each of three different states of the subject;
means for performing a paired Student's t test comparing the data in a first and a second array of said three arrays of bispectral data to produce a first t array and performing a paired Student's t test comparing the data in said second and a third array of said three arrays of bispectral data to produce a second t array;
means for comparing each data value in said first t array with data values in corresponding locations in said second t array;
means for identifying those corresponding locations in said first and second t arrays that differ by more than a pre-selected amount, said identified locations representing those locations that are significant for detecting the cerebral pheomena.

62. The system for noninvasively detecting cerebral phenomena of claim 35 further comprising means for displaying a representation of a subject's head, divided into a selected number of sections, said means for displaying including means for displaying a compressed continuous tracing of a computer diagnostic index determined from the signals acquired from an electrode positioned at a location represented by said section.

63. The system for noninvasively detecting cerebral phenomena of claim 61 wherein each displayed section includes a background of one of a plurality of colors, each of which colors is unique to a distinct selected range of possible values of a selected diagnostic index.

64. The method of noninvasively detecting cerebral phenomena of claim 1 further comprising the steps of:
generating three arrays of bispectral data for each of three different states of the subject;
performing statistical operations on said three arrays of bispectral data in order to identify those locations in said arrays that are significant for detecting the cerebral phenomena.

65. The system for noninvasively detecting cerebral phenomena of claim 1 further comprising:
means for generating three arrays of bispectral data for each of three different states of the subject;
means for statistically analyzing said arrays of bispectral data in order to identify those locations in said arrays that are significant for detecting the cerebral phenomena.

66. The method of noninvasively detecting cerebral phenomena of claim 1 wherein the cerebral phenomenon being detected is chronic ischemia or infarction in the subject being analyzed.

67. The method of noninvasively detecting cerebral phenomena of claim 1 wherein said bispectral values are generated by computing the Fourier transform of the third order autocorreltaiton function of said filtered signals.

68. The method of noninvasively detecting cerebral pheonemona of claim 1 wherein said bispectral values are generated by computing the Fourier transform of the third order crosscorrelation function of said filtered signals.

69. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said means for generating bispectral values comprises a means for computing the Fourier transform of the third order autocorrelation function of said filtered signals.

70. The system for noninvasively detecting cerebral phenomena of claim 35 wherein said means for generating bispectral values comprises a means for computing the Fourier transform of the third order crosscorrelation function of said filtered signals.