Identifying Montages that Best Detect Electrographic Seizure Activity During Polysomnography

ELECTROGRAPHIC SEIZURE ACTIVITY DURING POLYSOMNOGRAPHY Identifying Montages that Best Detect Electrographic Seizure Activity During Polysomnography Nancy Foldvary DO, 1 A.Cosmo Caruso MD, 1 Edward Mascha MS, 2 Michael Perry RPSG, R.EEG T, 1 George Klem R. EEG T, 1 Vincent McCarthy MD, 1 Farrukh Qureshi MD, 1 and Dudley Dinner MD 1 Departments of Neurology 1 and Biostatistics and Epidemiology, 2 The Cleveland Clinic Foundation, Cleveland, Ohio Study Objectives: Recognizing epileptic seizures during video polysomnography (VPSG) can be challenging, particularly when using standard, limited EEG montages and paper speed. Few sleep laboratories have PSG equipment that allows for the recording of 18 channels of EEG without compromising the ability to detect sleep apnea, periodic limb movements, and parasomnias. We studied the ability of sleep medicine- and EEG-trained polysomnographers to correctly identify epileptic seizures during PSG using 4, 7, and 18 channels of simultaneous EEG, recording at conventional PSG and EEG paper speeds. The purpose of this study was to determine the value of limited EEG montages viewed with EEG reformatting capability in the identification of seizures during PSG. Design: Blinded EEG analysis of seizures and arousals during VPSG. Setting: Tertiary care hospital with sleep laboratory and epilepsy monitoring unit. Patients: Subjects with focal (partial) epilepsy that underwent video-eeg monitoring. Interventions: We designed two 7-channel EEG montages that might facilitate the identification of seizures arising from the frontal and temporal lobes. Sleep medicine- and EEG-trained polysomnographers were asked to review tracings containing frontal or temporal lobe epileptic seizures and arousals from sleep. Utilizing the capability of our digital recording equipment to reformat EEG channels and change paper speeds, we asked the readers to classify events recorded with 4, 7, and 18 channels of simultaneous EEG, at paper speeds of 10 and 30 mm/sec. Measurements and Results: 6 readers viewed 32 sleep-related events (13 frontal lobe seizures, 11 temporal lobe seizures, and 8 arousals). The following factors were analyzed for their influence on accuracy of event detection: 1) the type of training of the reader (EEG vs. sleep medicine); 2) the number of EEG channels (4, 7, or 18); and 3) paper speed (10 vs. 30 mm/sec). Pair-wise comparisons and generalized estimating equations were used to identify factors leading to more accurate detection of seizures and arousals. 77% of events were correctly identified: 74% of seizures and 88% of arousals. Seizure detection was better using 7 and 18 channels (sensitivity of 82% and 86%, respectively) than 4 EEG channels (sensitivity of 67%) for temporal lobe seizures only. The number of EEG channels did not affect the accuracy of frontal lobe seizure detection. For EEG-trained readers, accuracy was greater using 30mm/sec than 10 mm/sec paper speed (85% vs. 78% correct, respectively). Conclusions: Adding EEG channels and EEG reformatting capabilities to PSG interpretation improves the detection of some types of epileptic seizures. Accuracy of temporal lobe seizure detection using an abbreviated 7-channel montage approximates that of an 18-channel EEG recording. However, the same is not true of frontal lobe seizures in which accuracy was similar regardless of the number of EEG channel available. Further studies are needed to identify specific EEG montages that would best detect epileptiform activity during VPSG. Key words: Video-EEG polysomnography; epileptic seizures; parasomnias; electroencephalography INTRODUCTION Accepted for publication August 1999 Address Correspondence to: Nancy Foldvary DO, The Cleveland Clinic Foundation, Department of Neurology, Section of Epilepsy and Sleep Disorders, 9500 Euclid Avenue, S-51, Cleveland, OH, 44195, Tel: (216) 445-2990, Fax: (216) 445-1022, E-mail: foldvan@.ccf.org SLEEP, Vol. 23, No. 2, 2000 1 ACCORDING TO CONVENTIONAL SLEEP STAGE SCORING, a minimum of one EEG channel is required for the identification of sleep stages and arousals during overnight polysomnography (PSG). 1,2 However, the differentiation of epileptic seizures, nonepileptic behavioral events in sleep, arousals, and artifacts is difficult, if not impossible, during standard PSG using one or even two EEG channels. The AASM Standards of Practice Indications for PSG Guidelines recommends the use of PSG with video and additional EEG channels to assist in the identification of events that are thought to represent seizures when the initial clinical evaluation and routine awake and sleep EEG are inconclusive. 3 Known as video- EEG polysomnography (VPSG), this technique combines simultaneous PSG and video-eeg to evaluate patients with

nocturnal behaviors or movements in whom parasomnias or epileptic seizures are suspected. Video-EEG PSG has several advantages over standard PSG including the ability to analyze behavior, correlate behavior with EEG, and more accurately detect seizure activity due to the additional scalp electrodes. Aldrich and Jahnke reviewed their experience of 122 patients with suspected parasomnias that underwent VPSG with 12 to 16 channels of EEG. 4 Video polysomnography provided a definite diagnosis of epilepsy or a sleep disorder in 35% of patients, supportive evidence of an epileptic or sleep disorder in 30%, and was inconclusive in the remainder. At our institution, patients with nocturnal behavioral events are scheduled for VPSG using 18-channel EEG recordings or inpatient video-eeg monitoring in the epilepsy monitoring unit (EMU) depending on the clinical situation and frequency of behavioral events. Patients are studied in the EMU when the clinical history and routine EEG are inconclusive or when events do not occur on a nightly basis. The technical time required for the placement of additional EEG electrodes in the sleep laboratory is twice that of standard PSG and the added time for physician analysis is substantial. Furthermore, few sleep laboratories have PSG equipment that allows for the recording of 18 channels of EEG without compromising the ability to detect sleep apnea, periodic limb movements, and parasomnias. We studied the ability of sleep medicine- and EEGtrained polysomnographers to correctly identify epileptic seizures during PSG using 4, 7, and 18 channels of simultaneous EEG, recording at conventional PSG and EEG paper speeds. The purpose of this study was to determine the value of limited EEG montages viewed with EEG reformatting capability for the identification of seizures during PSG. METHODS Patient Population SLEEP, Vol. 23, No. 2, 2000 2 Records from all patients with medically refractory frontal or temporal lobe epilepsy, who were admitted to the EMU for video EEG monitoring at the Cleveland Clinic Foundation in 1995, were reviewed. Thirty-two subjects were chosen for this study. All of these patients had brain MRI studies showing a single lesion in the frontal or temporal lobe and video EEG evaluations (including ictal and interictal recordings) supportive of epilepsy arising from the lobe harboring the lesion. A board-certified electroencephalographer selected a seizure from 13 patients with frontal lobe epilepsy (FLE) and 11 patients with temporal lobe epilepsy (TLE), and arousals from sleep from eight additional subjects. Only seizures with unambiguous EEG changes localized to the frontal and temporal region(s) were selected in order to test the validity of ictal localization using different montages. Files obscured by myogenic or movement artifact were excluded. All records were digitally acquired and reviewed on an 18-channel A-P bipolar montage using the 10/20 system of electrode placement and modified electrode position nomenclature and verified on referential recordings. 5,6 Filters were set at HFF 100 Hz and LFF 1.0 Hz. Montages The 32 events were randomly placed in five-minute digital files by one of the investigators (ACC) and displayed in five montages (below) using different numbers of EEG channels and paper speeds. Six readers reviewed all of the 32 files displayed in each of the five montages on five consecutive days in the EEG laboratory on a digital EEG review station. Two hours were allotted for each session. Once a file had been viewed, readers were not permitted to return to it at a later time in the session. Readers identified each file as to whether it contained either a seizure or an arousal. If a seizure was present, the reader was asked to localize it to a temporal or frontal lobe; if not possible, seizures were classified as 'undetermined.' Readers were blinded to all clinical information including medical records, videotapes, and technical notations during the recordings. Readers were not permitted to alter the number of channels or paper speed, but adjustments in sensitivity were permitted. Files in which the event was totally obscured by myogenic or movement artifact were excluded. The readers included two board-certified electroencephalographers and polysomnographers (NF, DSD), a board certified registered EEG technician (GK), two sleep medicine physicians (one pulmonologist and one neurologist) without specific training in EEG (VM, FQ), and a board-certified polysomnographic technician (MP). The board-certified electroencephalographers and EEG technician were considered EEG-trained, and the sleep medicine physicians and polysomnographic technician, sleep medicine-trained. The montages were as follows: Day 1/Montage 1: C3, C4, O1, and O2 linked to TP9 at paper speed of 10mm/sec (standard PSG montage and paper speed) Day 2/Montage 2: FZ, CZ, PZ, F7, F8, O1, and O2 linked to TP9 at paper speed of 10mm/sec (abbreviated EEG montage at PSG paper speed) Day 3/Montage 3: C3, C4, O1, and O2 linked to TP9 at paper speed of 30 mm/sec (standard PSG montage at EEG paper speed) Day 4/Montage 4: FZ, CZ, PZ, F7, F8, O1, and O2 linked to TP9 reference at paper speed of 30 mm/sec (abbreviated EEG montage at EEG paper speed)

Figure 1. The 10/20 system of electrode placement and TP9/10 Day 5/Montage 5: 18 channel A-P bipolar montage (shown below) at a paper speed of 30mm/sec (standard EEG montage and paper speed) FP1-F7 FP2-F8 FP1-F3 FP2-F4 FZ-CZ F7-T7 F8-T8 F3-C3 F4-C4 CZ-PZ T7-P7 T8-P8 C3-P3 C4-P4 P7-O1 P8-O2 P3-O1 P4-O2 The electrode placement TP9 was chosen as a reference because of their location in proximity to auricular and mastoid electrodes which are not used during video EEG monitoring in our laboratory. Unlike mastoid electrodes, TP9 and TP10 are recognized placements in the 10/10 system and therefore, their precise location can be consistently calculated (Figure 1). Electrodes TP9 and TP10 are placed on all patients admitted to the EMU as part of a standard bitemporal montage. 6 Similarly, auricular electrodes are not placed in patients admitted to the EMU for long-term monitoring because of their potential for skin irritation. For the identification of frontal lobe seizures, electrode placements FZ, CZ, and PZ were chosen for the abbreviated (7-channel) montage for two reasons. First, frontal lobe seizures tend to propagate rapidly to the contralateral hemisphere and the midline electrodes are most likely to detect these spread patterns. Second, seizures of frontal lobe origin are often obscured by muscle artifact due to tonic, clonic, and violent motor activity. This muscle activity tends to be minimal in the region of the vertex. For the identification of temporal lobe seizures, electrode placements F7 and F8 were chosen for the 7-channel montage. These electrodes record activity both from the anterior temporal and inferior frontal regions, although seizures arising from the former are significantly more common. More posterior temporal electrodes were specifically not chosen because recorded activity from the left temporal region would have been more susceptible to inphase cancellation when linked to the reference electrode, TP9. To reduce the detection of frontal lobe seizure activity by F7/8, we excluded patients with seizures arising from an orbitofrontal lesion on MRI. Table 1. Correct event designation by factor 1 Factor All Montage 4 channel, 7 channel, 4 channel, 7 channel, 18 channel, 10 mm/sec 10 mm/sec 30 mm/sec 30 mm/sec 30 mm/sec All 958 (77) 192 (70) 192 (74) 192 (80) 190 (81) 192 (81) Training Sleep 478 (73) 96 (70) 96 (67) 96 (76) 94 (78) 96 (73) EEG 480 (82) 2 96 (71) 96 (82) 96 (84) 96 (84) 96 (90) Event Seizure 718 (74) 144 (64) 144 (69) 144 (76) 142 (80) 144 (80) Temporal seizure 328 (77) 66 (67) 66 (67) 66 (79) 64 (86) 66 (86) Frontal seizure 390 (71) 78 (62) 78 (71) 78 (74) 78 (76) 78 (74) Arousal 240 (88) 3 48 (90) 48 (92) 48 (92) 48 (83) 48 (85) 1 Total number of readings (% correct; for Event rows, the percent = sensitivity) 2 EEG readers correctly identified a higher percentage of events than sleep medicine-trained readers (p= 0.007) 3 The sensitivity for arousal dectection was higher than seizures overall (p< 0.001) and frontal seizures (p= 0.001) SLEEP, Vol. 23, No. 2, 2000 3

Statistical Analysis We assessed the effect of various factors on the accuracy of seizure and arousal detection using the generalized estimating equation (GEE) method. 7 Specifically, we modeled the probability of correct designation (as either seizure or arousal) for each of the 958 collected observations. The factors of interest were training (EEG vs. sleep medicine), number of EEG channels (4, 7 or 18), paper speed (30 vs. 10 mm/sec), and event type (seizure vs. arousal, and frontal seizure vs. temporal seizure vs. arousal). The GEE method allowed us to assess each factor while adjusting for withinsubject correlation inherent in the repeated measures design. We also assessed the two- and three-way interactions among factors. An interaction between two factors is said to exist if the effect of a factor is not constant over the levels of another factor. A significance criterion of 0.10 was used for the interactions instead of 0.05 since it is more difficult statistically to detect important interactions. A multivariable model including significant interactions and main effects was fit, and factors were assessed by making pairwise comparisons within the interactions. The significance level for each hypothesis was 0.05 (i.e., effect of training, speed, number of channels or seizure type). Bonferroni-like corrections were made to the significance criterion when assessing factors within an interaction by dividing 0.05 by the number of levels we were comparing within. For example, when assessing any factor within each of the two levels of training (EEG and Sleep Medicine) the criterion was 0.05/2, or 0.025, and when assessing any factor within levels of event type or number of channels the criterion was 0.05/3, or 0.017. RESULTS The six readers reviewed 960 events. Results were available for 958 events; one reader labeled two files as uninterpretable due to the presence of myogenic artifact. The accuracy of event detection by factor is shown in Table 1. The six readers identified 77% of the 958 events correctly, including 74% of seizures (77% temporal, 71% frontal) and 88% of arousals. The readers correctly identified 70% of events when only provided with 4 channels of EEG using a paper speed of 10 mm/sec (montage 1). Accuracy increased to only 74% when 7 channels of EEG were available (montage 2). Accuracy increased to 80-81% when reading at a paper speed of 30 mm/sec, independent of the number of EEG channels available. As shown in Table 1, accuracy improved significantly with the number of EEG channels for EEG-trained readers, from 71% recording 4 channels of EEG at 10 mm/sec to 90% viewing 18 channels at 30 mm/sec. Increasing the number of EEG channels did not improve the diagnostic accuracy of the sleep medicine-trained readers, who correctly identified 70% of events using 4 EEG channels at 10 mm/sec and 73% of events viewing 18 channels at 30 mm/sec. In testing the effect of training alone, the accuracy of event detection was significantly greater for EEGthan sleep medicine-trained readers (82% vs. 73%). Arousals were more often correctly identified (88%) than seizures (74%) and temporal lobe seizures more often correctly identified than frontal lobe seizures (77% versus 71%). Neither paper speed, number of channels, nor specific montage affected the accuracy of correctly identifying arousals when interactions among factors were not considered. However, analysis of the interactions among factors revealed that the effect of any of the four factors of interest (paper speed, training, number of channels, and event type) is best assessed within levels of one or more of the other factors, and not overall as above. Specifically, the final GEE model included three significant interactions: training and paper speed (p=0.011), type of event and training (p=0.004), and number of EEG channels and type of event (p=0.037). This model allowed us to make pair-wise comparisons of the levels of each factor within levels of the other factor in each interaction, all the while adjusting for Table 2. Interaction between event type, training and channels on correct event designation 1 Event Type Training Number Channels ALL Sleep EEG 4 7 18 Temporal 163 (74) 165 (79) 132 (67) 130 (82) 4 66 (86) 4 328 (77) Frontal 195 (68) 195 (74) 156 (66) 156 (75) 78 (74) 390 (71) Arousal 120 (78) 120 (99) 2,3 96 (91) 5 96 (88) 48 (85) 240 (88) Interaction Event type x training P= 0.004 Event type x channels P= 0.037 1 Total number of readings (% correct or sensitivity) 2 Correct designation (sensitivity) is higher for EEG- than sleep-medicine-trained readers for arousals only (p= 0.008) 3 The sensitivity for arousal detection was higher than for frontal seizures for EEG readers (p= 0.013) 4 For temporal lobe seizure detection, the sensitivity is higher with 7 (p= 0.011) or 18 (p< 0.001) EEG channels than 4 channels 5 The sensitivity of arousal detection was higher than for frontal seizures using 4 EEG channels (p=0.007) SLEEP, Vol. 23, No. 2, 2000 4

the factors not involved in that interaction. Results of these within-factor assessments are summarized in Tables 2 and 3. Comparisons of event types within training are summarized in Table 2, where the percents correspond to the sensitivity for temporal seizures, frontal seizures, and arousals. This analysis revealed that EEG readers had higher sensitivity in identification of arousals than their sleep-trained counterparts (99% vs. 78%). No differences in sensitivity between groups of readers were found for temporal or frontal seizures. The percentage of arousals correctly identified by EEG-trained readers was significantly greater than for frontal lobe seizures (sensitivity of 99% vs. 74%), but not for temporal lobe seizures. No differences in sensitivity between event types were detected for sleep medicinetrained readers. Also shown in Table 2 is the interaction between number of EEG channels and event type. Pairwise comparisons revealed that the 7- and 18-channel montages provided significantly greater sensitivity than the 4-channel montages for temporal seizures only (82% and 86%, respectively vs. 67%), while no difference was observed between 7- and 18-channel montages for any event type (Figure 2). The added channels did not improve the sensitivity of frontal seizure or arousal detection (Figure 3,4). No differences were found among event types within a given number of channels, except that arousals were identified significantly more accurately than frontal and temporal lobe seizures when using 4 EEG channels (sensitivity of 91% vs. 66%, and 67%, respectively). The interaction of paper speed and reader training is shown in Table 3. No significant difference in diagnostic accuracy was found between the two paper speeds for either group of readers. However, when using a paper speed of 30 mm/sec, the percentage of events correctly identified by EEG- and sleep medicine-trained readers using a paper speed of 30 mm/sec was 85% and 72%, respectively, a difference that was statistically significant. No significant interaction between paper speed and number of channels was found (not shown). DISCUSSION The purpose of this study was to compare the accuracy of seizure detection during VPSG using abbreviated EEG montages and a standard 18-channel A-P bipolar montage. The advantages of limited EEG montages include reduced technical time for electrode placement, physician time for data analysis, and cost of data storage. Due to the additional technical time required, in our laboratory, one VPSG study is scheduled in the place of two standard PSGs. The AASM Standards of Practice Indications for PSG Guidelines recommends the use of PSG with video and additional EEG channels in an extended bilateral montage to assist in the diagnosis of SLEEP, Vol. 23, No. 2, 2000 5 nocturnal behaviors or movements that are thought to represent seizures. However, the guidelines do not provide specific recommendations for the optimal number and placement of EEG electrodes. The American Electroencephalographic Society recommends the use of at least six EEG channels in the evaluation of patients with suspected seizures. Electrode placements FP1, FP2 (or other frontal placements), C3, C4, T3, T4, O1, and O2 are suggested. 2 However, the accuracy of seizure detection using this montage has not been previously studied. Our goal in devising the modified montages was to use a minimum number of electrodes strategically placed to maximize the identification of focal seizures. We chose to add electrodes FZ, CZ, and PZ of the 10/20 System to optimize the identification of frontal lobe seizure patterns. Since seizures of frontal lobe origin tend to propagate rapidly to the contralateral hemisphere, we thought these midline electrodes would best detect this activity. We excluded the frontopolar (FP1, FP2) and supraorbital (SO1, SO2) electrode placements because they are so prone to contamination from ocular movements and muscle artifact. The F7 and F8 electrode placements may be considered anterior temporal or inferior frontal electrodes depending on the distribution of the field of the recorded activity. We incorporated F7 and F8 into our abbreviated montage as the temporal electrodes for two reasons. First, most temporal lobe seizures arise from the mesial temporal structures and propagate anteriorly, appearing maximal at the F7/8 or T7/T8 (formerly T3/4) electrode placements. The F7 and F8 electrodes also record activity from the orbitofrontal and cingulate areas, but seizures arising from these regions are exceedingly rare and were not included in the present study. In addition, since we chose electrode TP9 as reference, it was important to avoid the use of temporal electrode placements posterior to F7/F8, which, when linked to TP9 would produce low voltage activity due to in-phase Table 3. Interaction between training and paper speed in correct event designation 1 Training Paper Speed ALL 10 mm/sec 30 mm/sec Sleep 192 (73) 286 (72) 478 (73) EEG 192 (78) 288 (85) 2 480 (82) Interaction Training x speed P= 0.011 All 384 (75) 574 (79) 958 (77) 1 Total number of readings (% correct designation) 2 Correct designation was greater for EEG- than sleep-medicinetrained readers at 30 mm/sec (p<0.001). No difference was found between 30 and 10 mm/sec speeds for EEG- or sleep Medicine-trained readers.

cancellation and short inter-electrode distances. As with auricular electrodes, the use of TP9 as reference poses a problem for seizure localization since this electrode might be involved (active) in seizures arising from the left temporal region. Therefore, the differentiation of temporal and frontal seizures required some knowledge of the rules used in the localization of epileptiform activity and the effects of an active reference. For this reason, readers were not required to specify the side of seizure origin. The accuracy of temporal lobe seizure detection was significantly greater using 7- and 18- channel montages compared to 4 channels. More importantly, no significant difference was found between 7- and 18-channel recordings in the detection of temporal seizures. However, the additional midline electrode placements did not facilitate the identification of frontal lobe seizures. The reason for this is A B C D E F Figure 2. Right temporal lobe seizure in a 46-year old male with epilepsy due to encephalitis. The EEG onset (arrow) consisted of rhythmic alpha activity in the right temporal region preceding the clinical onset by 26 seconds as shown in consecutive 10-second epochs on the 18-channel A-P bipolar montage (A,B). The evolution of rhythmic activity in the right temporal region (FT10) can be appreciated on the 7-channel montage at 10 and 30 mm/sec paper speed (C,D). However, the pattern is not clearly evident on the 4-channel montage even viewed at a paper speed of 30 mm/sec (E,F). SLEEP, Vol. 23, No. 2, 2000 6

unclear. Explanations for the reduced yield of surface EEG in FLE include the inaccessibility of the mesial and basal aspects of the frontal lobes to surface electrodes, short seizure duration, and the tendency for the EEG to become contaminated by myogenic artifact due to the presence of prominent motor manifestations. 8 Since the seizures chosen for this study were preselected for their localized EEG changes, these effects were minimized. We found that the identification of arousals (i.e., the absence of an electrographic seizure) was slightly better when only 4 channels were used, although the difference between 4 channels and 7 or 18 channels was not statistically significant. One possible explanation for this finding is that availability of additional EEG provided a greater opportunity for readers to over-interpret an arousal as a seizure. Readers with EEG experience were most accurate using a paper speed of 30 mm/sec, the speed in which routine EEG and video-eeg monitoring records are reviewed. Not unexpectedly, sleep medicine-trained readers did not perform to the same degree using standard EEG paper speed. Also, not surprising is the fact that accuracy improved with the number of EEG channels for EEGtrained readers only, since sleep medicine-trained polysomnographers have limited, if any, exposure to conventional EEG. We chose to limit this study to subjects with clearly defined temporal or frontal lobe epilepsy based on the concordance of a lesion on MRI, ictal symptomatology, and ictal and interictal EEG findings. We did this for two reasons. First, temporal lobe epilepsy represents the most common symptomatic focal epilepsy, constituting over 80% of cases. 9 Second, frontal lobe seizures commonly occur in sleep and are often characterized by violent motor activity and vocalizations that are difficult to differentiate from parasomnias. Further, the EEG in FLE is often normal, obscured due to excessive motor activity, or abnormal showing widespread epileptiform activity, making the localization and diagnosis of FLE challenging. 8 Unlike our study population, many patients with focal epilepsy do not have lesions on MRI or interictal and ictal electrographic features limited to a single region of the brain. Consequently, our results are not strictly comparable A B C D Figure 3. Generalized motor seizure in a 13-year old girl with left frontal lobe epilepsy. Seizures were characterized by generalized clonic activity sometimes beginning in the right hand that obscured the EEG. The ictal pattern (arrow) seen on consecutive 10-second epochs using the 18-channel montage (A, B) cannot be differentiated from muscle artifact on the 4-channel (C) and 7-channel (D) montages. SLEEP, Vol. 23, No. 2, 2000 7

A B to results of patients with nocturnal seizures referred to sleep laboratories for VPSG. In addition, our patients had seizures recorded in the EMU over a period of three to seven days during antiepileptic drug withdrawal. Events may not be recorded on one or two nights of VPSG in the sleep laboratory. The ictal recordings used in this study were chosen for their localizing features to the frontal or temporal lobe when viewed on a variety of 18-channel montages used in video EEG monitoring. The montages most commonly used include an A-P bipolar (the double banana, and montage 5 in the present study), a transverse (coronal) bipolar montage, and a unipolar montage using PZ as reference. However, digitally acquired EEG can be reformatted and viewed in an infinite number of ways, making it possible to design montages that optimize the recording of ictal and interictal activity from suspected epileptogenic regions. During video EEG monitoring, it is not uncommon for an EEG seizure pattern to be apparent on one montage and not another. The bipolar tranverse montage, which was not used in this study, provides the most extensive midline coverage and is often useful in the evaluation of midline or parasagittal foci including those in the frontal lobes. Since ictal patterns typically evolve to reach frequencies in the alpha and beta ranges, it is not surprising that seizures were missed when viewed at a paper speed of 10 mm/sec. Conventional PSG paper speed effectively condenses these frequencies, rendering the underlying rhythmicity undetectable. Presumably for some of these reasons, 26% of seizures in the present study were not identified as such, including 20% of seizures viewed on the 18-channel A-P bipolar montage. In many instances, ictal EEG alone is insufficient for the identification and localization of focal seizures. To improve one's ability to diagnose and localize epileptic seizures, results of ictal and interictal EEG, ictal semiology and neuroimaging studies must be considered together. The identification of seizures during overnight PSG is a challenging issue for polysomnographers. A variety of factors, including the number and placement of EEG electrodes, experience of the polysomnographer, size and location of the ictal generator, and paper speed, likely affect the yield of seizure detection during VPSG. Given the cost and labor-intensive nature of VPSG, further studies are needed to identify whether abbreviated EEG montages are useful in the evaluation of patients with nocturnal seizures and nonepileptic events in the sleep laboratory. ACKNOWLEDGMENTS The authors would like to thank Madeleine Grigg- Damberger for her review of the manuscript. REFERENCES C Figure 4. Spontaneous arousal from sleep (arrow) shown on a 4-channel montage and 10 mm/sec paper speed (A), the abbreviated montage (B) and an 18- channel A-P bipolar montage (C). Accuracy for arousal detection was similar for all montages. SLEEP, Vol. 23, No. 2, 2000 8 1. Rechtschaffen A, Kales A., eds. A manual for standardized terminology: techniques and scoring system for sleep stages of human subjects. Washington, DC: U.S. Government Printing Office, 1968. 2. Guideline fifteen: guidelines for polygraphic assessment of sleeprelated disorders (polysomnography). J Clin Neurophysiol 1994;11(1):116-124. 3. Practice parameters for the indications for polysomnography and related procedures. Sleep 1987;20(6):406-422. 4. Aldrich MS, Jahnke B. Diagnostic value of video-eeg polysomnography. Neurology 1991;41:1060-1066.

5. Jasper HH. The 10-20 electrode system of the International Federation. Electroencephalogr Clin Neurophysiol. 1958;10:370-375. 6. American Electroencephalographic Society. Guideline thirteen: guidelines for standard electrode position nomenclature. J Clin Neurophysiol 1994;11:111-113. 7. Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika 1986;73:13-22. 8. Quesney LF. Preoperative electroencephalographic investigation in frontal lobe epilepsy: electroencephalographic and electrocorticographic recordings. Can J Neurol Sci 1991;18:559-563. 9. Hauser WA. The natural history of temporal lobe epilepsy. In: Lhders H, ed. Epilepsy surgery. New York: Raven Press, 1991:133-141. SLEEP, Vol. 23, No. 2, 2000 9