- Email: [email protected]
Computerized EEG monitoring
Computerized EEG Monitoring Bernard Rosenblatt and Jean Gotman Monitoring of central nervous system function in the intensive care unit is becoming more widely accepted as an integral part of critical care, The history of developments in electroencephalogram (EEG) technology is reviewed to better appreciate the rate of technological developments and their application to clinical practice. Basic concepts of digital EEG are reviewed. Principals of intensive care unit monitoring as they apply to clinical neurophysiological techniques are examined to better understand the goals for an "ideal central nervous system monitor." Some current advances and directions for future development in computerized EEG monitoring are discussed.
Copyright9 1999by W.B. Saunders Company
HE ABILITY TO monitor physiological functions has become a mainstay of intensive care unit (ICU) activity. However, until recently, the central nervous system (CNS) has been monitored principally by multiple clinical examinations. This technique, of course, has several serious limitations: (1) it is not applicable in the anesthetized, paralyzed patient on a ventilator; (2) it often does not detect deterioration in function before irreversible damage has occurred; (3) it requires a degree of expertise in eliciting and interpreting physical findings; and (4) it can only be performed intermittently. The electroencephalogram (EEG) has been proposed as a useful monitor of CNS function. 1,2 Recent advances in computer hardware and software technology allow us to consider the possibility of using this noninvasive technique for the collection, storage, and analysis of continuously recorded EEG data. 3-5 This review (1) gives a brief historical perspective on the development of EEG technology; (2) reviews some basic concepts of digital EEG; (3) discusses principles of ICU monitoring particularly as they apply to clinical neurophysiological techniques; and (4) reviews some of the current ad-
T
From the Division of Neurology and Department of Clinical Neurophysiology, Montreal Children's Hospital; and the Departments of Neurology/Neurosurgery and Pediatrics, McGill University, Montreal, Quebec, Canada. Address reprint requests to Bernard Rosenblatt, MD, Montreal Children's Hospital, Department of Clinical Neurophysiology, 2300 Tupper St, Room A522, Montreal, Quebec, Canada H3H 1P3. This work was supported in part by a grant from the Medical Research Council of Canada and Stellate Systems through the University-Industry program of the MRC (Grant MRC UI11514). Copyright 9 1999 by W.B. Saunders Company 1071-9091/99/0602-0007510.00/0
120
vances and directions for future developments in computerized EEG monitoring. HISTORICAL PERSPECTIVE
To have a better perspective on the current state of EEG technology and to better understand the potential for future developments, a brief historical review follows. The earliest electrical detector was a frog muscle preparation developed by the Italian Luigi Galvani (1737 to 1798). This was used to detect static electricity.6 After 11 years of experiments, Galvani in 1791 published a treatise 7 wherein he demonstrated that electrical stimulation of a peripheral nerve would cause an impulse to travel to and throughout the spinal cord and subsequently exit and travel distally with the resultant contraction of a frog muscle. He named this discovery "animal electricity." This discovery was not generally accepted for the next 50 years in part because of the influence of fellow Italian Allessandro Volta. Notwithstanding this delay, the contribution of the Italians Galvani (1737 to 1798) and Volta (1755 to 1832) and two Englishmen, George Ohm (1787 to 1854) and Michael Farraday (1791 to 1867), was such that by the mid 1800s, there was a general understanding of electrical potential and current and the recognition that living tissue had important electrical properties. 6 The latter part of the 19th century saw the development of nonpolarizable electrodes, the mirror, and then the string galvanometer. This allowed for further amplification of low-voltage biological activity and minimization of artifact. After the discovery of the electrical properties of neural tissue, the major efforts over the next 120 years was amplification of the signal, noise reduction, and data display. In the 1870s Caton was able to demonstrate EEG and even evoked potentials from the surface of animal brains. However, he Seminars in Pediatric Neurology, Vol 6, No 2 (June), 1999: pp 120-127
EEG MONITORING
could not make a permanent record and had to demonstrate his findings to an audience. Over the next 50 years, technological advances allowed adequate amplification and noise reduction such that by 1929 Berger could record the EEG from the scalp in humans. By the late 1930s the first commercial EEG machines became available. In 1947, Dawson 8 made the first successful effort at signal processing to record human somatosensoryevoked potentials. He developed an analog averaging device to enhance the low-voltage evoked potentials. The advent of desktop computers some 25 years later would make this a widely applicable clinical technique. Ives and Gloor made important refinements to EEG video telemetry. By 1973, they described a form of computerized telemetry where the computer provided a 2-minute delay in the EEG, allowing 24-hour unattended monitoring of seizure activity.6 With the development of computer technology and digitization of the EEG, it became possible to transform the EEG signal into mathematically derived parameters. This was done to reduce data and simplify interpretation. In long-term monitoring, the EEG could be presented in a variety of formats, such as compressed spectral array. 9 A number of methods of spike and seizure detection, which have been widely accepted in EEG/video monitoring of epileptic patients, were described, l~ The computer's identification of seizures and spikes allowed for more selective review of the EEG and video, making this a practical clinical technique. Attempts at monitoring for potential cerebral injury have primarily involved the use of intraoperative limited channel EEG instruments, such as the cerebral function monitor (CFM) and the cerebral function analyzing monitor (CFAM), which have shown some prognostic value in adults undergoing cardiopulmonary bypass. 11,12 As we can see, Caton was able, more than 100 years ago, to study many of the features of brain electrical activity with which we are familiar. Within 50 years the first human EEG recordings had been carried out, and not long after the first commercially available EEG machines began to appear. As Collura 6 notes in his review of EEG history, those who studied the EEG over the years were always at the forefront of technology and devel-
121
oped the technology themselves when it was not available to carry on their work. With the advent of increasingly more powerful and less expensive computers, we are entering the era of digital EEG. What is the role of EEG in monitoring patients? Does digital EEG offer us advantages particularly in neurological monitoring? What are the developments in software and hardware that will make widespread neurological monitoring both reliable and clinically practical? We try to deal with some of these issues in this review. PRINCIPLES OF NEUROLOGICAL MONITORING
Chiappa and Hoch I indicated the importance of monitoring the CNS in the ICU. Jordan 2 has presented five criteria for a successful ICU monitoring system. These set an appropriate standard by which to evaluate new technologies as they are introduced into ICU monitoring. The system must be (1) more sensitive and specific than clinical observations, (2) noninvasive, (3) easily operated and interpreted by nonexperts, (4) usable at the patient bedside, and (5) not interfere with medical or nursing care of the patient. Analog EEG with large machines, reams of paper, and the requirement of constant technician oversight could not meet a number of the abovementioned criteria. With increasingly more powerful computers and the development of appropriate software, digital EEG monitoring has become feasible. In 1992, Lesser et aP presented some design principles for computerized EEG monitoring. Since then, increasing numbers of digital EEG systems have become commercially available. With the use of digital EEG, we can now record continuously for long periods of time, unattended by a technician. The referentially recorded EEG can then be reformatted to any desired montage with various filters applied. Though this may be adequate for the EEG laboratory or epilepsy monitoring, what about the situation in the ICU? Are we able to meet Jordan's criteria for a system that will supply clinically relevant information in a timely fashion while being acceptable to ICU staff having to take care of critically ill children? Current EEG data, whether analog or digital, is clinically useful only after skilled interpretation. Can computers further analyze and present the acquired data so as to effect clinical decisions?
122
ROSENBLATT AND GOTMAN
BASIC PRINCIPLES
Analog to Digital Conversion The EEG signal must be amplified to the volt range, then the continuous signal (analog) is sampled at discrete time intervals. The voltage can then be represented as integer values (digital). Smaller steps mean greater accuracy of digitization. This step size should be less than the noise level of the amplifiers so all EEG fluctuations exceeding noise level can be represented. The step size is determined by the number of bits of A-D conversion. Current systems use 10- to 12-bit A-D conversion (eg, 1 jaV step size - 500 pV to 500 ~tV = 1,000 steps. This requires a 10-bit A-D converter (1,024 steps). Horizontal resolution (time) is determined by the sampling rate. Because a digital system only acquires voltage data at fixed intervals, the rate of data acquisition or sampling rate will determine the horizontal resolution. If sample points are too widely spaced, rapidly changing activity, such as a spike, may be distorted or entirely missed, t3-16 The sampling rate must be at least twice the frequency of the analog signal of highest frequency and non-negligible energy (Nyquist theorem). A sampling rate that is too low or failure to filter the EEG input before digitization can result in aliasing. Aliasing occurs when a high-frequency wave form on digitization occupies the same sample points as a lower frequency wave form and will appear in the digital representation under the lower frequency "alias." 13 In addition, not only must the EEG be sampled at an adequate rate, but the data must also be displayed with adequate resolution. 14The increasingly higher resolution of monitors is eliminating some of the sources of distortion of the displayed EEG. Slew errors can occur in multichannel A-D conversion. In sampling one channel at a time on a 16-channel machine, a 4.7 msec delay may occur from the last to first channel. "Sample and hold" technology allows simultaneous sampling of all channels.
Comparison of Digital and Analog EEG In Converting the EEG from analog to digital, one must be aware of advantages in addition to possible pitfalls of this technology. 17 Digital EEGs are most often recorded referentially. We can therefore have mathematical reconstruction of any montage at the time of interpretation. Analog
recordings require that montages be recorded sequentially such that no individual event can be examined by more than one montage. Digital recordings can be read with sensitivity, filter, and time adjustments made at the time of interpretation. One of the major advantages of digital EEG is in the area of data storage and recovery. Progressively larger volumes of data can be stored digitally at lower prices (eg, CD Rom). Paper is more expensive at high volumes with greater difficulty in handling and storage. Unfortunately, digitally stored EEGs have no universal standard such that records from different manufacturers are not at this time completely interchangeable. One of the great potential advantages for digital EEG systems is the ability to make them significantly smaller and lighter than analog recorders. This will be particularly important in the ICU setting.
Spike and Seizure Detection What are the advantages of automatic seizure detection? Although seizures are behavioral events, they are not always easy to detect because (1) the behavioral manifestations may be subtle and not always easily identified, (2) observers may not be constantly available, (3) in children, the patient may be unable to communicate the occurrence of an event or may be unaware of what happened and (4) in newborns, behavioral manifestations can be particularly difficult to recognize. Gotman 18 notes that seizure detection is a difficult problem because seizures are primarily behavioral events and not EEG events of specific morphology. Seizure morphology is highly variable. Ictal activity may consist of patterns, such as lowamplitude desynchronization, polyspike activity, rhythmic waves at a variety of frequencies, and spike-wave discharges. In addition, scalp recordings can be easily obscured by EMG, movement, and eyeblink artifact occurring during a seizure. Detection. Detection methods function by extracting features, such as amplitude, slope, sharpness, and peak to peak wave duration. To detect an event, comparisons can be made against fixed thresholds or against previously observed background patterns. Spikes. Spikes can be detected when sharpness and amplitude both exceed several standard deviations above the average for those parameters over the past 5 to 10 seconds. This method has high
EEG MONITORING
sensitivity but low specificity, that is, there are many false-positives because of the numerous transients in the EEG that are not epileptiform spikes. More sophisticated methods analyze in more detail the detected transients to eliminate most of those that are not epileptiform. Seizures. Seizures can be detected by their rhythmicity. The EEG can be analyzed for rhythmicity by taking the average and standard deviation of wave deviation and amplitude. Rhythmic events have low standard deviation. False-positives can result from rhythmic physiological activity, for example, rhythmic theta with drowsiness. The threshold can be dynamically set with respect to ongoing background. In adults, such methods result in the detection of 70% to 80% of seizures. It is therefore obvious that close clinical observation remains of critical importance during long-term monitoring. 18 PROGRESS IN THE DEVELOPMENT OF A NEUROPHYSIOLOGICAL MONITOR FOR THE PEDIATRIC ICU
Over the last several years, we have been involved in the development of a computerized EEG monitoring system at the Montreal Children's Hospital. This has involved a number of steps: (1) the introduction of digital EEG recording equipment in the pediatric ICU to monitor children after cardiac surgery; (2) the development of a seizure detection program applicable to neonates and infants; (3) the development of an expert system to act as an alarm in the recognition of background abnormalities; and (4) the automation of EEG analysis making the large amounts of data generated by monitoring more amenable to quick review by the electroencephalographer.
Continuous EEG Monitoring Following Congenital Cardiac Surgery in a Pediatric Critical Care Unit This study 4,5 allowed us to assess the relationship between EEG abnormalities in the first 24 hours after open cardiac surgery and short-term neurological outcome. It also proved the technical feasibility of obtaining high-quality EEGs in an electronically hostile environment without interfering in the care of critically ill children. Participants were 77 infants and children undergoing open cardiac surgical repair of congenital cardiac lesions with normal preoperative EEGs (39
123
acyanotic, 38 cyanotic). Subjects underwent portable continuous 8-channel digital EEG recording for 1 hour preoperatively and for the first 20 to 24 hours postoperatively. EEGs were visually analyzed for abnormalities. Neurological outcomes were determined by clinical examination before discharge from the ICU. Thirty-two patients had normal EEGs (26, normal outcome; 6, abnormal outcome); 45 patients had abnormal EEGs (14, normal outcome; 31, abnormal outcome). EEG versus outcome was significant at P < .0000003 by Fishers Exact Test with sensitivity 0.84, specificity 0.65, negative predictive value 0.81, and positive predictive value 0.68. The relationship of EEG abnormalities to neurological outcome was highly significant. These findings demonstrated that continuous EEG monitoring could be a useful tool to predict short-term neurological outcome following congenital cardiac surgery. To further automate EEG analysis in the pediatric ICU, two elements that must be available are (1) seizure detection and (2) ability to evaluate background abnormalities.
Automatic Seizure Detection in the Newbom One of the most important aspects of ICU EEG monitoring is the ability to recognize electrographic seizures, which may not be clinically evident. Although these seizures have been recognized as a serious problem in the pediatric ICU, prolonged monitoring without some form of automated seizure detection is not clinically practical. Automated seizure detection methods are now widely available for long-term epilepsy monitoring in adults. The electrographic features of neonatal seizures and background activity are very different from those in adults. In light of these differences, Gotman et a119,20 have developed a method for seizure detection in the neonate. Though most seizures in the newborn are characteristically rhythmic, 2x some newborn seizures do not fit into this category. Some neonatal seizures have a very gradual onset in comparison to seizures in adults. Therefore detection methods requiring sudden onset of ictal activity may not detect these events. Rhythmic seizure activity can also be much slower in the newborn (down to 0.5 Hz) compared with older children or adults. The proposed method for seizure detection had to detect a wide range of patterns, including
ROSENBLATTAND GOTMAN
124
rhythmic paroxysmal discharges at a wide range of frequencies, as well as repetitive spikes even when they were not very rhythmic. Three detection methods were applied. (1) Detection of rhythmic discharges (0.5 to 10 Hz). This method used spectral analysis of sequential 10second epochs of the EEG in a sliding window. This window moved along the EEG in 2.5-second steps. The algorithm was designed to extract features from each epoch and compare them with features from two previous epochs (background) 60 seconds behind. The long gap between background and current epoch allowed for detection of seizures that start very gradually (Fig 1). (2) A method for detection of multiple spikes based on the spike detection procedure of Gotman. l~ This required modification for the newborn because spikes in neonates have a much longer duration than in adults. A high pass digital filter was therefore applied (Fig 2), which allowed easier detection by the standard spike detection method. (3) Method 1 often failed to detect very slow activity with complex wave forms because detection of the dominant spectral peak was difficult.
The EEG was therefore filtered with a low-pass digital filter (Fig 3). The seizure detection method of Gotman could then be applied. These methods were developed using EEGs obtained from 55 newborns recorded at three hospitals. Two hundred and eighty-one hours of recordings containing 679 seizures were analyzed. An initial evaluation indicated that 71% of seizures and 78% of seizure clusters (groups of seizures separated by <90 seconds) were detected with a false detection rate of 1.7/hour. These methods were developed so as to operate in real time. To this point, we have shown the feasibility of carrying out prolonged digital recordings in the ICU, unattended by a technician, on critically ill children. Though these data were clinically relevant showing both significant background abnormalities or electrographic seizures, there was a significant time delay between the recording and the very labor-intensive job of reading these prolonged records. The neonatal seizure detection method would greatly simplify the identification and interpretation of electrographic seizures. However, it would not help deal with background
A
C
Fp2-T4
T3-C3
D
c3-cz
O. 0
1
2
3
4
5
6
7
'~-"--" 0
Frecl (Hz)
1
2
3
4
5
6
Freq (Hz)
E I
I Background epochs 20 sec
I Gap 60 sec
I Current epoch 10 sec
Fig 1. Schematic representation of the spectral analysis method. (A) EEG background for comparative analysis. (B) Spectrum of channel T3-C3 for the section shown in A. (C) EEG in the current epoch, (D) Spectrum of channel T3-C3 for the section shown in C, (E) Representation of the relative timing of current epoch and comparative background. The background consists of 2 epochs ending 60 seconds before the current epoch. Dominant frequency and width of dominant spectral peak are calculated on spectrum in D. The power ratio is then defined as the ratio of the area hatched in D to the area hatched in B; the location on the frequency axis of the area in B is the same as that in D, (Reprinted with permission20)
EEG MONITORING
125
A C3-P3 . . . . , , ' . " - , . . . . . , . ~ . , . , - - . . . . . . . , - . ~ ~ i P3-O1
~.,-,,--'-x......._ _ . _ ~
..,--v.~-.,,~,,,.~,.,.-...--.-,-~ . , , - ~ - . ~ - . , ~ j ~ r ~ / ~ ~ ' x . _ , _ . ~ , - ~ _ _ ~ - . _
?C3-P3
-
P3-O1 ~
~
'
v - ~ -
~
-
-
~
....
Cap 21 Mar 21:34
100/~v 1 sec
Fig 2. Example of newborn seizure consisting of runs of arrhythmic spikes (sometimes not detected by the spectral analysis method). However, spikes in the newborn EEG are relatively slow and are not detected with the normal spike detection method. (A) Original seizure with relatively long duration of spikes. (B) Seizure filtered with a high pass filter: infinite impulse response (IIR), third order, cutoff 2 Hz. After filtering, the spikes are briefer, sharper, and easier to detect by standard spike detection method. (Reprinted with permission. 20)
A Fp2-F4 F4-P4
P4-O2
B Fp2-F4 F4-P4 P4-O2
1200v
Sam 24 Mar 10:10
2 sec Fig 3. Example of newborn seizure consisting of a very slow frequency discharge. The waveforms are often complex, and such seizures are sometimes not detected by the spectral analysis method. (A) Original seizure. (B) Seizure after application of a low pass filter: IIR, 2nd order, cutoff I Hz. (Reprinted with permission. 20)
126
abnormalities either (1) identifying their presence and particularly their evolution over many hours or days or (2) facilitate their interpretation. The following two studies were meant to achieve these goals.
An Expert System for Monitoring in the Pediatric Intensive Care Unit The objective of this project 22 was to develop a system for automated neurophysiological monitoring in the pediatric ICU. The system was meant to indicate reliably the presence of EEG abnormalities and the need for expert review. This was not meant to provide the ICU with a level of EEG abnormality but act simply as an alarm indicating the need for review. A fully automated system, interpreting the EEG, was considered an aim unlikely to be reached with reliability, given the complexity of the factors that can affect an ICU recording. A total of 188 EEG sections lasting 6 hours each were obtained from 74 patients in the ICU. The method of EEG analysis involved many steps grouped in two parts: (1) feature extraction and (2) knowledge-based expert system. For feature extractions a 6-hour digital EEG recording had spectral band powers computed for every 30-second epoch. Artifact rejection was then carried out. Three main background features were extracted: amplitude, symmetry, front/back differentiation. The three main features were computed for each 5-minute section of EEG and their distribution compared with the distribution of the EEGs of a control group. This provided statistical values representing the level of abnormality of the three features. The second part of this process requires fuzzy logic to establish a correspondence between statistical levels of abnormality and the human readers' judgment of each of the three features. A neural network then allowed the system to learn to combine the judgments of abnormality for each of the features into an overall judgment applying to the complete recording. Such a complex system is required because there is no single correspondence between, for instance, a quantitative measure of asymmetry and the clinical judgment of the significance of this asymmetry. The EEGer and the expert system classified the EEG in seven levels of abnormality. There was concordance between the two in 45% of the cases. The expert system was within one abnormality level of the EEGer in 91% of cases and within two
ROSENBLAT'T AND GOTMAN
levels in 97%. This type of system could act as an alarm during ICU monitoring indicating that the EEG segment requires review by a neurophysiologist. According to the results, it would very rarely miss the presence of a seriously abnormal EEG. The implementation of such a system would greatly reduce the cost of monitoring since only a fraction of the EEGs required interpretation.
Automatic EEG Analysis During Long-Term Monitoring in the ICU To assist in reviewing prolonged EEGs, an automatic EEG analysis method was developed that can be used to compress the prolonged EEGs into two pages. 23 The proposed approach of automatic analysis of segmented EEG (AAS-EEG) consists of four basic steps: (1) segmentation, (2) feature extraction, (3) classification, and (4) presentation. The idea is to break down the EEG into stationary segments and extract features that can be used to classify the segments into groups of like patterns. The final step involves the presentation of the processed data in a compressed form. This is done by providing the EEGer with a representative sample from each group of EEG patterns and a compressed time profile of the complete EEG. To verify the above approach, 41 6-hour EEG records were assessed for normality via the AAS-EEG and conventional EEG approaches. The difference between the overall assessment of compressed and conventional EEG was within one abnormality level 100% of the time and within on-half level for 73.6% of the records. In this study, the feasibility and reliability of automatically segmenting and clustering the EEG was demonstrated. This allowed the reduction of a 6-hour tracing to a few representative segments and their time sequence. This should facilitate the review of long recordings during ICU monitoring. CONCLUSIONS
The technical and cost advantages of digital EEG make it evident that analog systems will gradually be replaced in the EEG laboratory. The ultimate role of computerized EEG monitoring in the ICU remains to be defined. Although hardware and software developments have facilitated monitoring and the review of data, the clinical utility of this
EEG MONITORING
127
t e c h n i q u e i n d e t e r m i n i n g p a t i e n t care r e q u i r e s f u r t h e r study. N u w e r 24 h a s s t r e s s e d t h e i m p o r t a n c e o f s y s t e m a t i c a s s e s s m e n t o f this n e w t e c h n o l o g y a n d to " t r y to define the role o f t h e t e c h n i q u e in the evaluation of individual patients." C o m p u t e r i z e d E E G m o n i t o r i n g h a s the p o t e n t i a l
to h e l p in the e a r l y r e c o g n i t i o n o f p o s s i b l e n e u r o l o g i c a l s e q u e l a e in critically ill p a t i e n t s a n d a l l o w a p p r o p r i a t e i n t e r v e n t i o n . H o w e v e r , its u s e f u l n e s s will b e directly p r o p o r t i o n a l to o u r u n d e r s t a n d i n g o f the s t r e n g t h s a n d pitfalls a s s o c i a t e d w i t h this technique.
REFERENCES
1. Chiappa KH, Hoch DB: Electrophysiologic Monitoring. Neurological and Neurosurgical IntensiVe Care (ed 3). Raven Press, New York, 1993; pp 147-183 2. Jordan KG: Continuous EEG and evoked potential monitoring in the neuroscience intensive care unit. J Clin Neurophysiol 10:445-475, 1993 3. Lesser RP, Webber WRS, Risher RS: Design principles for computerized EEG monitoring. Electroenceph Clin Neurophysiol 82:239-247, 1992 4. Rosenblatt B, Gottesman RD, Gotman J, et al: Continuous electroencephalogram recording following cardiac surgery: Progress in the development of a neurophysiological monitor for the pediatric intensive care unit. Ann Neurol 38:512, 1995 (abstr) 5. Rosenblatt B: Monitoring the CNS in children with congenital heart defects: Clinical neurophysiological techniques. Semin Pediatr Neurol 6:27, 1999 6. Collura TF: History and evolution of electroencephalographic instruments and techniques. J Clin Neurophysio110:476504, 1993 7. Goldensohn ES: Animal electricity from Bologna to Boston. Electroenceph Clin Neurophysiol 106:94-100, 1998 8. Dawson GD: Cerebral responses to electrical stimulation of peripheral nerve in man. J Neurol Neurosurg Psychiatry 10:137-140, 1947 9. Bickford RG, Billinger TW, Fleming NI: The Compressed Spectral Array (SA): A pictoral EEG Pro. San Diego Biomed Symp 11:365-370, 1972 10. Gotman J: Automatic seizure detection: Improvements and evaluation. Electroenceph Clin Neurophysiol 76:317-324, 1990 11. Nevin J, Colchester AC, Adam S, et al: Prediction of neurological damage after cardiopulmonary bypass surgery. Lancet 44:725-729, 1989
12. Maynard DE, Jenkinson JL: The cerebral function analysing monitor. Anaesthesia 39:678-690, 1984 13. Gotman J: The Use of Computers in Analysis and Display of EEG and Evoked Response. Current Practice of Clinical Electroencephalography (ed 2). New York, NY: Raven Press, 1990 14. Blum DE: Computer based electroencephalopgraphy: Technical basics: Basis for new application and potential pitfalls. Electroenceph Clin Neurophysiol 106:118-126, 1998 15. Chiappa K: Evoked potentials in clinical medicine (ed 3). Philadelphia, Lippincott-Raven, 1997 16. Wong PK: Digital EEG in Clinical Practice. Philadelphia, Lippincott-Raven, 1996 17. Swartz BE: The advantages of digital over analog recording techniques. Electroenceph Clin Neurophysio1106:113117, 1998 18. Gotman J: Automatic Detection of Seizures and Spikes in the EEG. Epilepsy Surgery, Hans Luders, Raven Press, New York, 1991, pp 307-316 19. Gotman J, Flanagan D, Rosenblatt B, et al: Evaluation of an automatic seizure detection method for the newborn EEG. Electroenceph Clin Neurophysiol 103:363-369, 1997 20. Gotman J, Flanagan D, Zhang J, et al: Automatic seizure detection in the newborn: Methods and initial evaluation. Electroenceph Clin Neurophysiol 103:356-362, 1997 21. Lui A, Hahn JS, Heldt GR et al: Detection of neonatal seizures through computerized EEG analysis. Electroenceph Clin Neurophysiol 82:30-37, 1992 22. Si Y, Gotman J, Pasupathy A, et al: An expert system for EEG monitoring in the pediatric intensive care unit. Electroenceph Clin Neurophysiol 106:488-500, 1998 23. Agarwal R, Gotman J, Flanagan D, et al: Automatic EEG analysis during long-term monitoring in the ICU. Electroenceph Clin Neurophysiol 107:44-58, 1998 24. Nuwer MR: Assessing digital and quantitative EEG in clinical settings. J Clin Neurophysiol 15:458-463, 1998
Copyright9 1999by W.B. Saunders Company
HE ABILITY TO monitor physiological functions has become a mainstay of intensive care unit (ICU) activity. However, until recently, the central nervous system (CNS) has been monitored principally by multiple clinical examinations. This technique, of course, has several serious limitations: (1) it is not applicable in the anesthetized, paralyzed patient on a ventilator; (2) it often does not detect deterioration in function before irreversible damage has occurred; (3) it requires a degree of expertise in eliciting and interpreting physical findings; and (4) it can only be performed intermittently. The electroencephalogram (EEG) has been proposed as a useful monitor of CNS function. 1,2 Recent advances in computer hardware and software technology allow us to consider the possibility of using this noninvasive technique for the collection, storage, and analysis of continuously recorded EEG data. 3-5 This review (1) gives a brief historical perspective on the development of EEG technology; (2) reviews some basic concepts of digital EEG; (3) discusses principles of ICU monitoring particularly as they apply to clinical neurophysiological techniques; and (4) reviews some of the current ad-
T
From the Division of Neurology and Department of Clinical Neurophysiology, Montreal Children's Hospital; and the Departments of Neurology/Neurosurgery and Pediatrics, McGill University, Montreal, Quebec, Canada. Address reprint requests to Bernard Rosenblatt, MD, Montreal Children's Hospital, Department of Clinical Neurophysiology, 2300 Tupper St, Room A522, Montreal, Quebec, Canada H3H 1P3. This work was supported in part by a grant from the Medical Research Council of Canada and Stellate Systems through the University-Industry program of the MRC (Grant MRC UI11514). Copyright 9 1999 by W.B. Saunders Company 1071-9091/99/0602-0007510.00/0
120
vances and directions for future developments in computerized EEG monitoring. HISTORICAL PERSPECTIVE
To have a better perspective on the current state of EEG technology and to better understand the potential for future developments, a brief historical review follows. The earliest electrical detector was a frog muscle preparation developed by the Italian Luigi Galvani (1737 to 1798). This was used to detect static electricity.6 After 11 years of experiments, Galvani in 1791 published a treatise 7 wherein he demonstrated that electrical stimulation of a peripheral nerve would cause an impulse to travel to and throughout the spinal cord and subsequently exit and travel distally with the resultant contraction of a frog muscle. He named this discovery "animal electricity." This discovery was not generally accepted for the next 50 years in part because of the influence of fellow Italian Allessandro Volta. Notwithstanding this delay, the contribution of the Italians Galvani (1737 to 1798) and Volta (1755 to 1832) and two Englishmen, George Ohm (1787 to 1854) and Michael Farraday (1791 to 1867), was such that by the mid 1800s, there was a general understanding of electrical potential and current and the recognition that living tissue had important electrical properties. 6 The latter part of the 19th century saw the development of nonpolarizable electrodes, the mirror, and then the string galvanometer. This allowed for further amplification of low-voltage biological activity and minimization of artifact. After the discovery of the electrical properties of neural tissue, the major efforts over the next 120 years was amplification of the signal, noise reduction, and data display. In the 1870s Caton was able to demonstrate EEG and even evoked potentials from the surface of animal brains. However, he Seminars in Pediatric Neurology, Vol 6, No 2 (June), 1999: pp 120-127
EEG MONITORING
could not make a permanent record and had to demonstrate his findings to an audience. Over the next 50 years, technological advances allowed adequate amplification and noise reduction such that by 1929 Berger could record the EEG from the scalp in humans. By the late 1930s the first commercial EEG machines became available. In 1947, Dawson 8 made the first successful effort at signal processing to record human somatosensoryevoked potentials. He developed an analog averaging device to enhance the low-voltage evoked potentials. The advent of desktop computers some 25 years later would make this a widely applicable clinical technique. Ives and Gloor made important refinements to EEG video telemetry. By 1973, they described a form of computerized telemetry where the computer provided a 2-minute delay in the EEG, allowing 24-hour unattended monitoring of seizure activity.6 With the development of computer technology and digitization of the EEG, it became possible to transform the EEG signal into mathematically derived parameters. This was done to reduce data and simplify interpretation. In long-term monitoring, the EEG could be presented in a variety of formats, such as compressed spectral array. 9 A number of methods of spike and seizure detection, which have been widely accepted in EEG/video monitoring of epileptic patients, were described, l~ The computer's identification of seizures and spikes allowed for more selective review of the EEG and video, making this a practical clinical technique. Attempts at monitoring for potential cerebral injury have primarily involved the use of intraoperative limited channel EEG instruments, such as the cerebral function monitor (CFM) and the cerebral function analyzing monitor (CFAM), which have shown some prognostic value in adults undergoing cardiopulmonary bypass. 11,12 As we can see, Caton was able, more than 100 years ago, to study many of the features of brain electrical activity with which we are familiar. Within 50 years the first human EEG recordings had been carried out, and not long after the first commercially available EEG machines began to appear. As Collura 6 notes in his review of EEG history, those who studied the EEG over the years were always at the forefront of technology and devel-
121
oped the technology themselves when it was not available to carry on their work. With the advent of increasingly more powerful and less expensive computers, we are entering the era of digital EEG. What is the role of EEG in monitoring patients? Does digital EEG offer us advantages particularly in neurological monitoring? What are the developments in software and hardware that will make widespread neurological monitoring both reliable and clinically practical? We try to deal with some of these issues in this review. PRINCIPLES OF NEUROLOGICAL MONITORING
Chiappa and Hoch I indicated the importance of monitoring the CNS in the ICU. Jordan 2 has presented five criteria for a successful ICU monitoring system. These set an appropriate standard by which to evaluate new technologies as they are introduced into ICU monitoring. The system must be (1) more sensitive and specific than clinical observations, (2) noninvasive, (3) easily operated and interpreted by nonexperts, (4) usable at the patient bedside, and (5) not interfere with medical or nursing care of the patient. Analog EEG with large machines, reams of paper, and the requirement of constant technician oversight could not meet a number of the abovementioned criteria. With increasingly more powerful computers and the development of appropriate software, digital EEG monitoring has become feasible. In 1992, Lesser et aP presented some design principles for computerized EEG monitoring. Since then, increasing numbers of digital EEG systems have become commercially available. With the use of digital EEG, we can now record continuously for long periods of time, unattended by a technician. The referentially recorded EEG can then be reformatted to any desired montage with various filters applied. Though this may be adequate for the EEG laboratory or epilepsy monitoring, what about the situation in the ICU? Are we able to meet Jordan's criteria for a system that will supply clinically relevant information in a timely fashion while being acceptable to ICU staff having to take care of critically ill children? Current EEG data, whether analog or digital, is clinically useful only after skilled interpretation. Can computers further analyze and present the acquired data so as to effect clinical decisions?
122
ROSENBLATT AND GOTMAN
BASIC PRINCIPLES
Analog to Digital Conversion The EEG signal must be amplified to the volt range, then the continuous signal (analog) is sampled at discrete time intervals. The voltage can then be represented as integer values (digital). Smaller steps mean greater accuracy of digitization. This step size should be less than the noise level of the amplifiers so all EEG fluctuations exceeding noise level can be represented. The step size is determined by the number of bits of A-D conversion. Current systems use 10- to 12-bit A-D conversion (eg, 1 jaV step size - 500 pV to 500 ~tV = 1,000 steps. This requires a 10-bit A-D converter (1,024 steps). Horizontal resolution (time) is determined by the sampling rate. Because a digital system only acquires voltage data at fixed intervals, the rate of data acquisition or sampling rate will determine the horizontal resolution. If sample points are too widely spaced, rapidly changing activity, such as a spike, may be distorted or entirely missed, t3-16 The sampling rate must be at least twice the frequency of the analog signal of highest frequency and non-negligible energy (Nyquist theorem). A sampling rate that is too low or failure to filter the EEG input before digitization can result in aliasing. Aliasing occurs when a high-frequency wave form on digitization occupies the same sample points as a lower frequency wave form and will appear in the digital representation under the lower frequency "alias." 13 In addition, not only must the EEG be sampled at an adequate rate, but the data must also be displayed with adequate resolution. 14The increasingly higher resolution of monitors is eliminating some of the sources of distortion of the displayed EEG. Slew errors can occur in multichannel A-D conversion. In sampling one channel at a time on a 16-channel machine, a 4.7 msec delay may occur from the last to first channel. "Sample and hold" technology allows simultaneous sampling of all channels.
Comparison of Digital and Analog EEG In Converting the EEG from analog to digital, one must be aware of advantages in addition to possible pitfalls of this technology. 17 Digital EEGs are most often recorded referentially. We can therefore have mathematical reconstruction of any montage at the time of interpretation. Analog
recordings require that montages be recorded sequentially such that no individual event can be examined by more than one montage. Digital recordings can be read with sensitivity, filter, and time adjustments made at the time of interpretation. One of the major advantages of digital EEG is in the area of data storage and recovery. Progressively larger volumes of data can be stored digitally at lower prices (eg, CD Rom). Paper is more expensive at high volumes with greater difficulty in handling and storage. Unfortunately, digitally stored EEGs have no universal standard such that records from different manufacturers are not at this time completely interchangeable. One of the great potential advantages for digital EEG systems is the ability to make them significantly smaller and lighter than analog recorders. This will be particularly important in the ICU setting.
Spike and Seizure Detection What are the advantages of automatic seizure detection? Although seizures are behavioral events, they are not always easy to detect because (1) the behavioral manifestations may be subtle and not always easily identified, (2) observers may not be constantly available, (3) in children, the patient may be unable to communicate the occurrence of an event or may be unaware of what happened and (4) in newborns, behavioral manifestations can be particularly difficult to recognize. Gotman 18 notes that seizure detection is a difficult problem because seizures are primarily behavioral events and not EEG events of specific morphology. Seizure morphology is highly variable. Ictal activity may consist of patterns, such as lowamplitude desynchronization, polyspike activity, rhythmic waves at a variety of frequencies, and spike-wave discharges. In addition, scalp recordings can be easily obscured by EMG, movement, and eyeblink artifact occurring during a seizure. Detection. Detection methods function by extracting features, such as amplitude, slope, sharpness, and peak to peak wave duration. To detect an event, comparisons can be made against fixed thresholds or against previously observed background patterns. Spikes. Spikes can be detected when sharpness and amplitude both exceed several standard deviations above the average for those parameters over the past 5 to 10 seconds. This method has high
EEG MONITORING
sensitivity but low specificity, that is, there are many false-positives because of the numerous transients in the EEG that are not epileptiform spikes. More sophisticated methods analyze in more detail the detected transients to eliminate most of those that are not epileptiform. Seizures. Seizures can be detected by their rhythmicity. The EEG can be analyzed for rhythmicity by taking the average and standard deviation of wave deviation and amplitude. Rhythmic events have low standard deviation. False-positives can result from rhythmic physiological activity, for example, rhythmic theta with drowsiness. The threshold can be dynamically set with respect to ongoing background. In adults, such methods result in the detection of 70% to 80% of seizures. It is therefore obvious that close clinical observation remains of critical importance during long-term monitoring. 18 PROGRESS IN THE DEVELOPMENT OF A NEUROPHYSIOLOGICAL MONITOR FOR THE PEDIATRIC ICU
Over the last several years, we have been involved in the development of a computerized EEG monitoring system at the Montreal Children's Hospital. This has involved a number of steps: (1) the introduction of digital EEG recording equipment in the pediatric ICU to monitor children after cardiac surgery; (2) the development of a seizure detection program applicable to neonates and infants; (3) the development of an expert system to act as an alarm in the recognition of background abnormalities; and (4) the automation of EEG analysis making the large amounts of data generated by monitoring more amenable to quick review by the electroencephalographer.
Continuous EEG Monitoring Following Congenital Cardiac Surgery in a Pediatric Critical Care Unit This study 4,5 allowed us to assess the relationship between EEG abnormalities in the first 24 hours after open cardiac surgery and short-term neurological outcome. It also proved the technical feasibility of obtaining high-quality EEGs in an electronically hostile environment without interfering in the care of critically ill children. Participants were 77 infants and children undergoing open cardiac surgical repair of congenital cardiac lesions with normal preoperative EEGs (39
123
acyanotic, 38 cyanotic). Subjects underwent portable continuous 8-channel digital EEG recording for 1 hour preoperatively and for the first 20 to 24 hours postoperatively. EEGs were visually analyzed for abnormalities. Neurological outcomes were determined by clinical examination before discharge from the ICU. Thirty-two patients had normal EEGs (26, normal outcome; 6, abnormal outcome); 45 patients had abnormal EEGs (14, normal outcome; 31, abnormal outcome). EEG versus outcome was significant at P < .0000003 by Fishers Exact Test with sensitivity 0.84, specificity 0.65, negative predictive value 0.81, and positive predictive value 0.68. The relationship of EEG abnormalities to neurological outcome was highly significant. These findings demonstrated that continuous EEG monitoring could be a useful tool to predict short-term neurological outcome following congenital cardiac surgery. To further automate EEG analysis in the pediatric ICU, two elements that must be available are (1) seizure detection and (2) ability to evaluate background abnormalities.
Automatic Seizure Detection in the Newbom One of the most important aspects of ICU EEG monitoring is the ability to recognize electrographic seizures, which may not be clinically evident. Although these seizures have been recognized as a serious problem in the pediatric ICU, prolonged monitoring without some form of automated seizure detection is not clinically practical. Automated seizure detection methods are now widely available for long-term epilepsy monitoring in adults. The electrographic features of neonatal seizures and background activity are very different from those in adults. In light of these differences, Gotman et a119,20 have developed a method for seizure detection in the neonate. Though most seizures in the newborn are characteristically rhythmic, 2x some newborn seizures do not fit into this category. Some neonatal seizures have a very gradual onset in comparison to seizures in adults. Therefore detection methods requiring sudden onset of ictal activity may not detect these events. Rhythmic seizure activity can also be much slower in the newborn (down to 0.5 Hz) compared with older children or adults. The proposed method for seizure detection had to detect a wide range of patterns, including
ROSENBLATTAND GOTMAN
124
rhythmic paroxysmal discharges at a wide range of frequencies, as well as repetitive spikes even when they were not very rhythmic. Three detection methods were applied. (1) Detection of rhythmic discharges (0.5 to 10 Hz). This method used spectral analysis of sequential 10second epochs of the EEG in a sliding window. This window moved along the EEG in 2.5-second steps. The algorithm was designed to extract features from each epoch and compare them with features from two previous epochs (background) 60 seconds behind. The long gap between background and current epoch allowed for detection of seizures that start very gradually (Fig 1). (2) A method for detection of multiple spikes based on the spike detection procedure of Gotman. l~ This required modification for the newborn because spikes in neonates have a much longer duration than in adults. A high pass digital filter was therefore applied (Fig 2), which allowed easier detection by the standard spike detection method. (3) Method 1 often failed to detect very slow activity with complex wave forms because detection of the dominant spectral peak was difficult.
The EEG was therefore filtered with a low-pass digital filter (Fig 3). The seizure detection method of Gotman could then be applied. These methods were developed using EEGs obtained from 55 newborns recorded at three hospitals. Two hundred and eighty-one hours of recordings containing 679 seizures were analyzed. An initial evaluation indicated that 71% of seizures and 78% of seizure clusters (groups of seizures separated by <90 seconds) were detected with a false detection rate of 1.7/hour. These methods were developed so as to operate in real time. To this point, we have shown the feasibility of carrying out prolonged digital recordings in the ICU, unattended by a technician, on critically ill children. Though these data were clinically relevant showing both significant background abnormalities or electrographic seizures, there was a significant time delay between the recording and the very labor-intensive job of reading these prolonged records. The neonatal seizure detection method would greatly simplify the identification and interpretation of electrographic seizures. However, it would not help deal with background
A
C
Fp2-T4
T3-C3
D
c3-cz
O. 0
1
2
3
4
5
6
7
'~-"--" 0
Frecl (Hz)
1
2
3
4
5
6
Freq (Hz)
E I
I Background epochs 20 sec
I Gap 60 sec
I Current epoch 10 sec
Fig 1. Schematic representation of the spectral analysis method. (A) EEG background for comparative analysis. (B) Spectrum of channel T3-C3 for the section shown in A. (C) EEG in the current epoch, (D) Spectrum of channel T3-C3 for the section shown in C, (E) Representation of the relative timing of current epoch and comparative background. The background consists of 2 epochs ending 60 seconds before the current epoch. Dominant frequency and width of dominant spectral peak are calculated on spectrum in D. The power ratio is then defined as the ratio of the area hatched in D to the area hatched in B; the location on the frequency axis of the area in B is the same as that in D, (Reprinted with permission20)
EEG MONITORING
125
A C3-P3 . . . . , , ' . " - , . . . . . , . ~ . , . , - - . . . . . . . , - . ~ ~ i P3-O1
~.,-,,--'-x......._ _ . _ ~
..,--v.~-.,,~,,,.~,.,.-...--.-,-~ . , , - ~ - . ~ - . , ~ j ~ r ~ / ~ ~ ' x . _ , _ . ~ , - ~ _ _ ~ - . _
?C3-P3
-
P3-O1 ~
~
'
v - ~ -
~
-
-
~
....
Cap 21 Mar 21:34
100/~v 1 sec
Fig 2. Example of newborn seizure consisting of runs of arrhythmic spikes (sometimes not detected by the spectral analysis method). However, spikes in the newborn EEG are relatively slow and are not detected with the normal spike detection method. (A) Original seizure with relatively long duration of spikes. (B) Seizure filtered with a high pass filter: infinite impulse response (IIR), third order, cutoff 2 Hz. After filtering, the spikes are briefer, sharper, and easier to detect by standard spike detection method. (Reprinted with permission. 20)
A Fp2-F4 F4-P4
P4-O2
B Fp2-F4 F4-P4 P4-O2
1200v
Sam 24 Mar 10:10
2 sec Fig 3. Example of newborn seizure consisting of a very slow frequency discharge. The waveforms are often complex, and such seizures are sometimes not detected by the spectral analysis method. (A) Original seizure. (B) Seizure after application of a low pass filter: IIR, 2nd order, cutoff I Hz. (Reprinted with permission. 20)
126
abnormalities either (1) identifying their presence and particularly their evolution over many hours or days or (2) facilitate their interpretation. The following two studies were meant to achieve these goals.
An Expert System for Monitoring in the Pediatric Intensive Care Unit The objective of this project 22 was to develop a system for automated neurophysiological monitoring in the pediatric ICU. The system was meant to indicate reliably the presence of EEG abnormalities and the need for expert review. This was not meant to provide the ICU with a level of EEG abnormality but act simply as an alarm indicating the need for review. A fully automated system, interpreting the EEG, was considered an aim unlikely to be reached with reliability, given the complexity of the factors that can affect an ICU recording. A total of 188 EEG sections lasting 6 hours each were obtained from 74 patients in the ICU. The method of EEG analysis involved many steps grouped in two parts: (1) feature extraction and (2) knowledge-based expert system. For feature extractions a 6-hour digital EEG recording had spectral band powers computed for every 30-second epoch. Artifact rejection was then carried out. Three main background features were extracted: amplitude, symmetry, front/back differentiation. The three main features were computed for each 5-minute section of EEG and their distribution compared with the distribution of the EEGs of a control group. This provided statistical values representing the level of abnormality of the three features. The second part of this process requires fuzzy logic to establish a correspondence between statistical levels of abnormality and the human readers' judgment of each of the three features. A neural network then allowed the system to learn to combine the judgments of abnormality for each of the features into an overall judgment applying to the complete recording. Such a complex system is required because there is no single correspondence between, for instance, a quantitative measure of asymmetry and the clinical judgment of the significance of this asymmetry. The EEGer and the expert system classified the EEG in seven levels of abnormality. There was concordance between the two in 45% of the cases. The expert system was within one abnormality level of the EEGer in 91% of cases and within two
ROSENBLAT'T AND GOTMAN
levels in 97%. This type of system could act as an alarm during ICU monitoring indicating that the EEG segment requires review by a neurophysiologist. According to the results, it would very rarely miss the presence of a seriously abnormal EEG. The implementation of such a system would greatly reduce the cost of monitoring since only a fraction of the EEGs required interpretation.
Automatic EEG Analysis During Long-Term Monitoring in the ICU To assist in reviewing prolonged EEGs, an automatic EEG analysis method was developed that can be used to compress the prolonged EEGs into two pages. 23 The proposed approach of automatic analysis of segmented EEG (AAS-EEG) consists of four basic steps: (1) segmentation, (2) feature extraction, (3) classification, and (4) presentation. The idea is to break down the EEG into stationary segments and extract features that can be used to classify the segments into groups of like patterns. The final step involves the presentation of the processed data in a compressed form. This is done by providing the EEGer with a representative sample from each group of EEG patterns and a compressed time profile of the complete EEG. To verify the above approach, 41 6-hour EEG records were assessed for normality via the AAS-EEG and conventional EEG approaches. The difference between the overall assessment of compressed and conventional EEG was within one abnormality level 100% of the time and within on-half level for 73.6% of the records. In this study, the feasibility and reliability of automatically segmenting and clustering the EEG was demonstrated. This allowed the reduction of a 6-hour tracing to a few representative segments and their time sequence. This should facilitate the review of long recordings during ICU monitoring. CONCLUSIONS
The technical and cost advantages of digital EEG make it evident that analog systems will gradually be replaced in the EEG laboratory. The ultimate role of computerized EEG monitoring in the ICU remains to be defined. Although hardware and software developments have facilitated monitoring and the review of data, the clinical utility of this
EEG MONITORING
127
t e c h n i q u e i n d e t e r m i n i n g p a t i e n t care r e q u i r e s f u r t h e r study. N u w e r 24 h a s s t r e s s e d t h e i m p o r t a n c e o f s y s t e m a t i c a s s e s s m e n t o f this n e w t e c h n o l o g y a n d to " t r y to define the role o f t h e t e c h n i q u e in the evaluation of individual patients." C o m p u t e r i z e d E E G m o n i t o r i n g h a s the p o t e n t i a l
to h e l p in the e a r l y r e c o g n i t i o n o f p o s s i b l e n e u r o l o g i c a l s e q u e l a e in critically ill p a t i e n t s a n d a l l o w a p p r o p r i a t e i n t e r v e n t i o n . H o w e v e r , its u s e f u l n e s s will b e directly p r o p o r t i o n a l to o u r u n d e r s t a n d i n g o f the s t r e n g t h s a n d pitfalls a s s o c i a t e d w i t h this technique.
REFERENCES
1. Chiappa KH, Hoch DB: Electrophysiologic Monitoring. Neurological and Neurosurgical IntensiVe Care (ed 3). Raven Press, New York, 1993; pp 147-183 2. Jordan KG: Continuous EEG and evoked potential monitoring in the neuroscience intensive care unit. J Clin Neurophysiol 10:445-475, 1993 3. Lesser RP, Webber WRS, Risher RS: Design principles for computerized EEG monitoring. Electroenceph Clin Neurophysiol 82:239-247, 1992 4. Rosenblatt B, Gottesman RD, Gotman J, et al: Continuous electroencephalogram recording following cardiac surgery: Progress in the development of a neurophysiological monitor for the pediatric intensive care unit. Ann Neurol 38:512, 1995 (abstr) 5. Rosenblatt B: Monitoring the CNS in children with congenital heart defects: Clinical neurophysiological techniques. Semin Pediatr Neurol 6:27, 1999 6. Collura TF: History and evolution of electroencephalographic instruments and techniques. J Clin Neurophysio110:476504, 1993 7. Goldensohn ES: Animal electricity from Bologna to Boston. Electroenceph Clin Neurophysiol 106:94-100, 1998 8. Dawson GD: Cerebral responses to electrical stimulation of peripheral nerve in man. J Neurol Neurosurg Psychiatry 10:137-140, 1947 9. Bickford RG, Billinger TW, Fleming NI: The Compressed Spectral Array (SA): A pictoral EEG Pro. San Diego Biomed Symp 11:365-370, 1972 10. Gotman J: Automatic seizure detection: Improvements and evaluation. Electroenceph Clin Neurophysiol 76:317-324, 1990 11. Nevin J, Colchester AC, Adam S, et al: Prediction of neurological damage after cardiopulmonary bypass surgery. Lancet 44:725-729, 1989
12. Maynard DE, Jenkinson JL: The cerebral function analysing monitor. Anaesthesia 39:678-690, 1984 13. Gotman J: The Use of Computers in Analysis and Display of EEG and Evoked Response. Current Practice of Clinical Electroencephalography (ed 2). New York, NY: Raven Press, 1990 14. Blum DE: Computer based electroencephalopgraphy: Technical basics: Basis for new application and potential pitfalls. Electroenceph Clin Neurophysiol 106:118-126, 1998 15. Chiappa K: Evoked potentials in clinical medicine (ed 3). Philadelphia, Lippincott-Raven, 1997 16. Wong PK: Digital EEG in Clinical Practice. Philadelphia, Lippincott-Raven, 1996 17. Swartz BE: The advantages of digital over analog recording techniques. Electroenceph Clin Neurophysio1106:113117, 1998 18. Gotman J: Automatic Detection of Seizures and Spikes in the EEG. Epilepsy Surgery, Hans Luders, Raven Press, New York, 1991, pp 307-316 19. Gotman J, Flanagan D, Rosenblatt B, et al: Evaluation of an automatic seizure detection method for the newborn EEG. Electroenceph Clin Neurophysiol 103:363-369, 1997 20. Gotman J, Flanagan D, Zhang J, et al: Automatic seizure detection in the newborn: Methods and initial evaluation. Electroenceph Clin Neurophysiol 103:356-362, 1997 21. Lui A, Hahn JS, Heldt GR et al: Detection of neonatal seizures through computerized EEG analysis. Electroenceph Clin Neurophysiol 82:30-37, 1992 22. Si Y, Gotman J, Pasupathy A, et al: An expert system for EEG monitoring in the pediatric intensive care unit. Electroenceph Clin Neurophysiol 106:488-500, 1998 23. Agarwal R, Gotman J, Flanagan D, et al: Automatic EEG analysis during long-term monitoring in the ICU. Electroenceph Clin Neurophysiol 107:44-58, 1998 24. Nuwer MR: Assessing digital and quantitative EEG in clinical settings. J Clin Neurophysiol 15:458-463, 1998