EEGXXX10.1177/1550059417736444Clinical EEG and NeurosciencePrichep et al research-article7364442017

Original Article

Clinical EEG and 11–­1 Exploration of the Pathophysiology of © EEG and Society (ECNS) 2017 Reprints and permissions: Chronic Using Quantitative EEG sagepub.com/journalsPermissions.nav DOI:https://doi.org/10.1177/1550059417736444 10.1177/1550059417736444 Source Localization journals.sagepub.com/home/eeg

Leslie S. Prichep1,2, Jaini Shah3, Henry Merkin4, and Emile M. Hiesiger5

Abstract Chronic pain affects more than 35% of the US adult population representing a major public health imperative. Currently, there are no objective means for identifying the presence of pain, nor for quantifying pain severity. Through a better understanding of the pathophysiology of pain, objective indicators of pain might be forthcoming. Brain mechanisms mediating the painful state were imaged in this study, using source localization of the EEG. In a population of 77 chronic pain patients, significant overactivation of the “Pain Matrix” or pain network, was found in brain regions including, the anterior cingulate, anterior and posterior insula, parietal lobule, thalamus, S1, and dorsolateral prefrontal cortex (DLPFC), consistent with those reported with conventional functional imaging, and extended to include the mid and posterior cingulate, suggesting that the increased temporal resolution of electrophysiological measures may allow a more precise identification of the pain network. Significant differences between those who self-report high and low pain were reported for some of the regions of interest (ROIs), maximally on left hemisphere in the DLPFC, suggesting encoding of pain intensity occurs in a subset of pain network ROIs. Furthermore, a preliminary multivariate logistic regression analysis was used to select quantitative-EEG features which demonstrated a highly significant predictive relationship of self-reported pain scores. Findings support the potential to derive a quantitative measure of the severity of pain using information extracted from a multivariate descriptor of the abnormal overactivation. Furthermore, the frequency specific (theta/low alpha band) overactivation in the regions reported, while not providing direct evidence, are consistent with a model of thalamocortical dysrhythmia as the potential mechanism of the neuropathic painful condition.

Keywords EEG, chronic neuropathic pain, low-resolution electromagnetic tomographic analysis (LORETA), brain imaging, thalamocortical dysrhythmia (TCD), Pain Matrix, pain network

Received September 17, 2015; revised August 31, 2017; accepted September 4, 2017.

Introduction and/or the insula in intensity encoding of pain.6-8 Vogt9 describes a dynamic role played by the cingulate cortex in Chronic pain affects more than 35% of the US adult population pain perception with different functions of the anterior, mid, and as such represents a major public health imperative. and posterior regions. Using the significantly increased tem- Currently, there is no objectively verifiable, clinically useful poral resolution of electromagnetic method, Bromm10 com- means of identifying the presence of pain or quantifying the pared findings with that of positron emission tomography severity of pain. The standard of care relies on subjective rat- (PET) and functional magnetic resonance imaging (fMRI), ings, using a visual analog scale (VAS), frequently resulting in and demonstrated that the more posterior parts of the cingu- suboptimal or excessive treatment and inappropriate distribu- late play an important role in the painful state, occurring tion of resources. The ability to quantify pain objectively also prior to projection to the ACC, suggesting that the increased has therapeutic implications for patients who are obtunded, unconscious, or very young, in whom current subjective meth- ods cannot be used. 1Department of , NYU School of Medicine, New York, NY, USA 2 Following the seminal publication by Melzack and Wall,1 BrainScope Co, Inc, Bethesda, MD, USA 3Center for Neural Science, New York University, New York, NY, USA functional methods have been used to study 4Neurometric Evaluation Service–NY, New York, NY, USA the neuronal basis of pain and pain perception, and have 5Departments of and Radiology, NYU Medical Center, New demonstrated that painful stimulation activates extensive York, NY, USA 2-5 areas of cortical and subcortical brain regions which they Corresponding Author: referred to as the “Pain Matrix.” Neuroimaging studies fur- Leslie S. Prichep, 1115 Broadway, Suite 1082, New York, NY 10010, USA. ther suggest the specific role of the anterior cingulate (ACC) Email: [email protected] 2 Clinical EEG and Neuroscience 00(0) time resolution allows a more precise identification of the Patients were excluded who had prior histories of neurologi- pain pathway. cal or psychiatric illness, head injuries with loss of conscious- A number of pain studies using electromagnetic recordings, ness, or skull abnormalities. Patients on permanent disability report oscillatory shifts in the frequency spectrum in the pres- and those with a spinal cord stimulator or other implantable ence of pain,11 especially in theta and beta frequency bands.12-15 devices which precluded MRIs, were also excluded. Using source localization of scalp recorded EEG data (low- Seventy-seven age- and gender-matched controls were resolution electromagnetic tomographic analysis [LORETA]) selected from the Brain Research Laboratories NYU School of in patients with chronic neuropathic pain,16,17 sources consis- Medicine database and used as controls for this study. Subjects tent with those sources identified using conventional neuroim- in this normative database did not have any evidence of clini- aging methods were reported, including the insula, ACC, cally significant pain, depression, or anxiety. prefrontal, inferior posterior parietal, cortices. No differences in the frequency specific overactivation were found in those 17 EEG Data Acquisition taking analgesic medications. Furthermore, patients reevalu- ated following successful central lateral nucleus thalamotomy Twenty minutes of eyes closed resting EEG was recorded from (CLT), showed significant diminution in overactivation in the 19 electrodes, placed using the international 10/20 electrode insular and cingulate cortices. Such results support a model placement system, referenced to linked earlobes, using a suggesting the in chronic pain is based on a Deymed Tru-Scan EEG system. A differential eye channel 18,19 thalamocortical dysrhythmia (TCD). In addition, recent (diagonally above and below the eye orbit) was used for the studies have suggested the potential of EEG as a biomarker in detection of eye movement. All electrode impedances were less 20 the prediction of treatment response to opioids. than 5000 ohm. Amplifiers had a bandpass from 0.5 to 70 Hz (3 21 Using similar technology Prichep et al published a fea- dB points), with a 60 Hz notch filter. Data were sampled at a sability study demonstrating significant frequency specific rate of 200 Hz with 12-bit resolution. Patients were monitored (theta and beta EEG bands) overactivation of the "Pain Matrix" during EEG recording. or pain network in chronic neuropathic pain patients. All patients were evaluated in a high pain state and then reevalu- ated following pain relieving procedure/treatment that resulted Pain Scales and Clinical Data in a 50% or more decrease in their pain rating. Significant The following self-rating scales were used to provide data on reduction in overactivation of brain regions of the pain network severity of pain, pain related disability, presence of depression was found in the lower pain state with maximal change across and anxiety, presence of neuropathic pain, and relevant medical patients occurring in the insula, thalamus, and cingulate. history: This study was conducted in a large population of chronic pain patients, using quantitative EEG (QEEG) source localiza- Brief Pain Inventory (BPI) and Follow-up Pain Ratings. The BPI was tion to investigate the overactivation of the pain network, the used to rate overall pain, current pain and maximal pain over relationship between the level of activation and the severity of varied time periods, to indicate regions of pain, and to rate the self-rated pain, the evidence of TCD underlying the abnormali- influence of pain on activities of daily life and quality of life. ties observed, and the potential to use QEEG biomarkers as a predictor of the level of pain reported by the patient. Oswestry Disability Questionnaire. Oswestry Disability Ques- tionnaire22 was used to quantify information about how the Methods patient’s pain affects the ability to manage everyday life. Subjects Neuropathic Pain Questionnaire. The validated Neuropathic Pain Male and female chronic pain patients, aged 19 to 85 years, Questionnaire23 rates the presence or absence of neuropathic were candidates for study. This population was referred from pain. pain clinics and pain practitioners in the New York City area. Written informed consent was obtained in all cases. Beck Depression Inventory (BDI). Because of the high incidence All patients had a history of symptoms for at least 3 months’ of comorbid depression in pain patients we measured depres- duration (chronic pain as defined by the International sion using the BDI.24 This is the standard scale for self-rating Association for the Study of Pain [IASP] criteria), and a diag- depression which is correlated with the presence of clinical nosis from the referring physician, which supported the symp- depression. toms. The etiology of the pain included most commonly radicular (neuropathic) pain, involving lumbar or cervical root Beck Anxiety Inventory (BAI). The BAI25 is the standard scale for compression from benign degenerative causes (as determined self-rating anxiety which is correlated with the presence of by MRI and/or computed tomography [CT] scans), and/or clinical anxiety. myofacial/musculoskeletal pain, as the predominant source of their pain. Patients were permitted to continue their analgesic Clinical History and Treatment History. The referring physician medication during the study. filled out a form containing clinical history, pertinent radiological Prichep et al 3 and electro-diagnostic data and relevant clinical findings. Special realistic head model,33 using the Montreal Neurological attention was paid to details of the treatment history, the time Institute (MNI) 152 template34 with the 3-dimensional solution since onset of painful symptoms and medication history, includ- space restricted to cortical gray matter, as determined by the ing the present. probabilistic Talairach atlas.35 The scalp electrode positions are placed in spatial registration with the Montreal Probabilistic 36 EEG Data Analysis and Source Localization MRI Brain Atlas (PBA). Z-scores of deviations from expected normative values are computed for each voxel relative to age- The raw EEG data were visually edited off-line by trained EEG expected normative values. The significance levels of the technologists, to identify and eliminate artifacts (eg, signals images take into consideration the large number of measure- due to eye [electro-oculography, EOG] or muscle movement ments made, using the correction introduced by Worsley and [electromyography, EMG]), augmented by a computerized colleagues.37 The voxel norms are based on replicated, vali- artifact algorithm. The EEG recordings were not reviewed by a dated normative equations which describe the features of the neurologist as the analyses to be performed were statistical in broad band EEG resting state (“ground state” or default net- nature (taking advantage of advances in signal processing of work) as a set of mathematical equations as a function of the EEG signal) and performed only on the first 2 minutes of age30,38 and extended to the very narrow band spectra (done in artifact-free EEG data. The use of 2 minutes as an adequate collaboration with Pasqual-Marquis). Significant maxima in sample for the estimation of the stationary ground state of the bands of interest were selected for source imaging and the brain electrical activity is the standard in the field of quantita- inverse solution applied. tive electrophysiology.26-28 Quantitative analyses included fast In this work, a modification of the sLORETA algorithm was Fourier transform (FFT) to convert the data from the time to the used which extended the model to include regions to the level frequency domain. For every electrode, data were averaged of the thalamus. The neocortical volume in the sLORETA mask across artifact free segments to yield measures of absolute and is divided into 6896 voxels of dimension 5 mm3. In addition to relative power and mean frequency in the conventional wide individual voxels, a generalized solution is computed for the frequency bands between 1.5 and 35 Hz (delta, 1.5-3.5 Hz; specific ROIs of interest, to include, ACC, mid-CC, as well as theta, 3.5-7.5 Hz; alpha, 7.5-12.5 Hz; beta, 12.5-25 Hz; and all ROIs for regions thought to be part of the pain network (eg, beta2/gamma1, 25-35Hz bands), and total power for the entire left and right insula, parietal lobule, DLPFC, SI, etc). Three- frequency spectrum. Inter- and intrahemispheric relationships, dimensional color-coded tomographic images were generated, including power gradients and synchrony (coherence) were with source generator distributions superimposed upon trans- also computed for all electrode pairs. Following neurometric axial, coronal, and sagittal slices (4 mm3) of the PBA which QEEG methodology,26,29,30 all quantitative features were log correspond to the loci of the inverse solutions. These images transformed to obtain an approximately Gaussian distribution are not used for analysis, but rather for purposes of visualizing and z-transformed relative to age expected normal values. The the data for descriptive purposes only. Tables of z-values for age-regression equations used in the z-transform of the data in each ROI are constructed to aid in interpretation of the images. this study were those published and validated by John and col- Within each patient the largest z-score peak in the VNB leagues30 on an independent normal population. The z-score spectra was identified. A VNB maxima was found within the (standard score) was computed taking the difference between range 5.46 to 9.36 Hz (theta/low alpha) in the vast majority of the individual patient value for the feature and the mean value the patients. Source images were derived for each patient at for the feature in the normal population for that age and divid- their maxima within this band. The averages shown in the fig- ing by the standard deviation of the norm population for that ures represent the group average of the sLORETAs for each age. Computation of source localization was conducted using individual. This approach takes individual differences into sLORETA,31,32 a standardized, discrete, distributed, linear, account by requiring that sources were derived at a peak fre- minimum norm inverse solution for estimating the source gen- quency for each individual patient. In addition, this band cov- erators of the scalp recorded EEG. ered the frequency specific range reported in the scientific Similar to the wide band frequency spectra described above, literature for patients in the pain population.16-18 Group average this algorithm computes the very narrow band (VNB) fre- images were computed using sources derived from the indi- quency spectra or current density from each of the VNB from vidual maxima. There are many approaches used in the litera- 0.5 to 50 Hz in 0.39-Hz steps, as well as conventional broad ture to characterize each ROI, in this case we used the mean of bands, delta (1.5-3.5 Hz), theta (3.5-7.5 Hz), alpha (7.5-12.5 the 10% largest voxel z-scores in each ROI. Using this method Hz), beta (12.5-25 Hz), and beta2/gamma1 (25-35 Hz). takes advantage of the strength of using the most significant sLORETA (LORETA freeware programs: http://www.uzh.ch/ value, while eliminating a single outlier value and is not subject keyinst/NewLORETA/Software/Software.htm) requires no a to the regression to the mean which can cause blurring and priori assumptions of the number of active dipoles for each underestimates of the effects. component and implements a three concentric shells spherical Only the VNB source data are presented in this article as the head model. Current density is computed as the linear, weighted focus of this study was to demonstrate the use of EEG source sum of the scalp electrical potentials, squared for each voxel, localization to reflect activation in the pain network of the brain yielding power of current density. Computations are made in a described from conventional neuroimaging. 4 Clinical EEG and Neuroscience 00(0)

Statistical Analyses QEEG Source Localization Images were constructed for z-scores for all voxels and com- Figure 1 shows source localization images (sLORETA) for the puted z-scores for ROIs in the pain population, and for ROIs in group average z-scores for ROIs in the frequency band of inter- the high and low pain populations. The following statistical est, 5.46 to 9.36 Hz, for the pain population. As can be seen in analyses were conducted to test for the significance of the dif- this image, significant overactivation (orange/yellow) was ferences in ROIs between groups, including present for many ROIs, with the most significant abnormities seen in the parietal lobule (L > R, regardless of the side of 1. T tests for differences between pain patients and normal pain), the insula, the mid cingulate, the dorsolateral prefrontal matched controls for ROIs of the pain network; cor- cortex, S1. Overactivation is also observed in the ACC, rected for multiple comparisons using permutation tests although less than that seen in the mid cingulate. 2. T tests for differences between groups for ROIs of the Figure 2 shows t-scores for the significance of the differ- pain network comparing high and low pain popula- ences in the ROIs between the patients with pain and controls. tions; corrected for multiple comparisons using permu- Highly significant differences can be seen between these tation tests groups in the regions of the pain network, with the most signifi- 3. An exploratory stepwise logistic regression was con- cant differences being in the mid and posterior cingulate, the ducted to identify those features of the QEEG which insula (L > R), the DLPFC (L > R), the parietal lobule (L > R), optimized the relationship between the prediction based and S1. on QEEG and self-rated pain score. Table 1 shows the significance of the differences between mean z-scores for each ROIs of the pain network, shown in Results Figure 1. Highly significant differences were found in the mid and anterior cingulate, anterior and posterior insula, S1, Subjects DLPFC, supplementary motor area, hippocampus, and thalamus. Seventy-seven chronic pain male and female patients, with a When patients were divided into those who reported high mean age of 49.3 years (SD = 15.8, range 19.5-81.7 years), pain scores (5-10) and compared with those reporting low pain 53% females, were included for study. The mean pain score scores (1-4), similar patterns of overactivation in the pain net- reported on the 10-point VAS reported was 4.97 (SD = 2.32, work were found, as can be seen in Figure 3, top and middle range 1-10); which was consistent with the mean pain score panels, although more significant overactivation was seen in obtained on the BPI of 5.07 (SD = 2.04, range 0.28-9.14). the high pain group (top panel). The bottom panel of Figure 3 Seventy-nine percent of the patients were right handed, 16% shows the significance of the differences in ROIs between the left handed, and 5% ambidextrous. The mean number of years high and low pain populations. Highly significant differences in pain was 8.4 years (SD = 8.3). The mean BDI score was were obtained between groups, with the most significant differ- 13.27 (SD = 7.25 [1-30]) and the mean BAI score was 12.29 ences found on the left hemisphere in the parietal lobule, the (SD = 8.30 [0-35]). Using the Oswestry scale 33% of the insula (posterior > anterior), S1, and the DLPFC. patients reported moderate and 49% reported severe impair- Table 2 shows the means and standard deviations for the ment in functions of daily life. There were 40 patients who high and low pain populations for each ROI, and the signifi- reported moderate to severe pain (VAS ≥5) and 37 reported low cance of the difference between the two subpopulations. The pain scores (VAS <5). The high pain group had a mean age of regions where significance was obtained included the DLPFC, 50.9 years (SD = 16.2), with a VAS score of 6.8 (SD = 1.48) parietal lobule, S1, anterior and posterior insula, (posterior > and average years of pain of 7.63 years (SD = 8.09). The low anterior) and the medial frontal gyrus; all on the left hemi- pain group had a mean age of 47.5 years (SD = 15.4), with a sphere. It is noted that using a measure of functional impair- VAS score of 2.97 (SD = 1.01) and average years of pain of ment in the presence of pain (Oswestry), divided by mild, 9.32 years (SD = 8.65). Normal controls were matched for age moderate, or severe levels, significant correlations were found and gender, with a mean age of the normal controls was 50.25 with z-scores in medial frontal, ACC, DLPFC, and left anterior (SD = 16.6, range 19.4-81.8). insula ROIs. The correlation between Oswestry and VAS was The vast majority of the subjects suffered from neuropathic significant (r = 0.43, P < .01). pain and or myofascial pain. The most common location of A preliminary analysis was performed using a stepwise pain was back and or leg. The most common predominant logistic regression to test the feasibility of predicting the pain sources of pain were root compression, herniations, osteo- scores reported by the patient based on a small number of phytes of foraminal stenosis. Forty-five percent of patients’ QEEG features from a limited montage selected by the logistic pain was reported to be bilateral, 26% left side, and 29% right regression analysis (n = 67, a subset of the full population due side. All patients were on medication for their chronic pain, to missing information). These features included most impor- including oral and transcutaneous narcotics, tricyclic antide- tantly frontal and central scalp locations and features in the pressants, serotonin-norepinephrine reuptake inhibitors theta and beta frequency bands. Figure 4 shows the regression (SNRIs), nonsteroidals, and corticosteroids. curve. Highly significant prediction accuracy was obtained, Prichep et al 5

Figure 1. Source localization (sLORETA) images for group average z-scores for ROIs in the band 5.46 to 9.36 Hz for the pain population (n = 77). The images are color coded for significance of difference according to the color scale (below figure) where extremes of the color scale are P < .01, and the Δ represents the P < .05 point on the scale. Since all features are z-transformed and therefore expressed in standard deviation units of the normal population, there is an associated probability where z-1.96 = P < .01. It is noted that since these are group averages the signficance of the group z-score and associated probability are computed times the square root of the n of the group population, thus 8.77 times the group average z-score shown. (Left is on left in these images.)

R = 0.907, P < .00001. It can be seen that the tightest fit is on Clinical Features and ROI z-Scores the extremes of the scale and that the largest variance is in the middle of the scale, suggesting that it is the middle of the scale None of the ROIs were found to correlate significantly with where the patients have the most difficulty in assigning a score. depression scores on the BDI. There was only 1 ROI that With anchors at both ends of the scale, the estimates from the showed a significant correlation with anxiety on the BAI equation are potentially more reliable objective estimates. and that was the medial frontal region (R = 0.33, P < .04). 6 Clinical EEG and Neuroscience 00(0)

Figure 2. T-scores for the significance of the difference pain patients and matched normal controls, for sources of the frequency band 5.46 to 9.36 Hz.

However, although reaching significance, this low correla- side only. There were no significant relationships with z-scores tion was not considered to be clinically meaningful. There for ROIs. were no significant correlations with ROIs for chronicity (the number of years the patient reported to have been in Discussion pain). When considering if there was a relationship between side This study represents one of the largest chronic pain popula- of reported pain and the z-score for the ROIs. Patients were tions studied with electrophysiological imaging reported in the divided into those who reported bilateral pain, including bilat- literature. Using EEG source localization, the data demonstrate eral in combination with either side, left side only and right the significant overactivation of regions described as part of the Prichep et al 7

Table 1. Mean and Standard Deviation Voxel Values for Pain and Control Groups for Each Region of Interest.a It is important to note that the probability associated with group mean z-scores would be computed times the square root of the n of the group, thus 8.77 times the group z-score shown.

Region of Interest Pain, Mean (SD) Control, Mean (SD) t Permutation Test P Anterior cingulate 0.63 (1.26) 0.55 (0.90) 1.58 .11 Mid cingulate 0.95 (1.39) 0.65 (0.92) 5.34 <.0001** Posterior cingulate 0.79 (1.53) 0.37 (1.02) 6.22 <.0001** Anterior L insula 0.57 (1.39) 0.29 (0.88) 5.07 <.0001** Posterior L insula 0.91 (1.61) 0.40 (0.92) 7.90 <.0001** Anterior R Insula 0.50 (1.43) 0.29 (0.88) 3.62 .0008** Posterior R Insula 0.65 (1.54) 0.37 (0.96) 4.37 <.0001** L S1 1.20 (1.53) 0.51 (0.98) 11.17 <.0001** R S1 0.89 (1.48) 0.45 (1.00) 7.14 <.0001** L inferior parietal lobule 1.26 (1.59) 0.48 (0.98) 12.13 <.0001** R inferior parietal lobule 0.81 (1.50) 0.42 (0.99) 6.28 <.0001** L DLPFC 0.95 (1.41) 0.52 (0.88) 7.60 <.0001** R DLPFC 0.81 (1.31) 0.47 (0.85) 6.24 <.0001** SMA 0.73 (1.36) 0.46 (0.95) 4.66 <.0001** Medial frontal cortex 0.35 (1.25) 0.30 (0.94) 1.02 .30 L parahippocampal gyrus 0.63 (1.54) 0.28 (0.93) 5.74 <.0001** (hippocampus/amygdala) R parahippocampal gyrus 0.54 (1.50) 0.30 (0.95) 3.85 .0002** (hippocampus/amygdala) Thalamus 0.63 (1.54) 0.40 (0.97) 3.76 <.0001** L caudate 0.72 (1.46) 0.50 (0.92) 3.70 <.0001** R caudate 0.63 (1.43) 0.51 (0.93) 2.09 .03*

Abbreviations: L, left; R, right; DLPFC, dorsolateral prefrontal cortex; SMA, supplementary motor area. at value and corrected P values (using permutation tests) are shown for each region of interest. *.001 < P < .05. **P < .001. pain network using conventional functional neuroimaging the side of reported pain) and the largest significance was found methods. These regions include most significantly the mid and in the DLPFC. In a PET study of the relationship between posterior cingulate, the anterior and posterior insula, the pari- DLPFC and pain intensity, Lorenz and colleagues39 reported etal lobule, the thalamus, and the DLPFC. Since this study was that DLPFC activity correlated negatively with perceived conducted in pain patients in the absence of nociceptive stim- intensity and unpleasantness of pain. More specifically, during uli, results support continuous overactivation of regions of the high left DLPFC activity inter-regional correlation of midbrain pain network in the chronic pain state. It is noted that the mid and thalamic activity was reduced, which they interpreted as and posterior cingulate show highly significant differences reflecting modulation of effective connectivity between mid- between normal and pain patients, while significance was not brain-thalamic pathways. Further evidence of the left hemi- seen in the ACC ROI often associated with the painful state in sphere in the encoding of pain was reported by Favilla and conventional neuroimaging studies. The ROI for the anterior colleagues40 who rank-ordered fMRI ROIs based on the rela- cingulate in this study was not limited to the rostral ACC, tionship with perceived pain level and found the most signifi- where others have indicated the maximal differences in the cant relationship to be in the left mid cingulate cortex. Further ACC region, but included posterior and mid cingulate regions exploration of the functional significance of such asymmetric as well. The differences obtained may reflect the advantage of representation in the encoding of pain is needed. The fact that the temporal resolution of electrophysiological data which the differences between high and low pain states occur in a although averaged across FFT epochs, is based on high-resolu- subset of ROIs of the pain network suggests that these regions tion frequency encoded information captured by the EEG sig- are more involved in the encoding of the subjective intensity of nal and support the importance of these areas of the cingulate pain, while the others represent the presence of pain regardless cortex in chronic pain.9,10 of severity. It is also important to note that the evidence of sev- Support for the relationship between the magnitude of over- eral ROIs showing significant differences in the difference activation in the significant ROIs and encoding of the level of between high and low pain states supports distributed process- pain was also demonstrated in this study. Highly significant ing of pain intensity, also suggested by others.41 differences were obtained in comparisons between patients Differences obtained between high and low pain patients reporting high versus low pain scores. It was observed that suggest the potential for the objective quantification of pain most significances were on the left hemisphere (independent of derived from such measures. Support of the relationship 8 Clinical EEG and Neuroscience 00(0)

Figure 3. Group average z-scores for ROIs for sources at in the band 5.46 to 9.36 Hz. High pain population (top row), low pain population (middle row), and the t-test between high and low groups (bottom row). The scale for the top 2 panels shows ±z P < .001, and for the bottom panel t = ±2.5 for P < .001. between functional severity and ROI activation was also seen must be interpreted with caution until replicated and systemati- using the Oswestry disability scale, further supporting the cally optimized. potential scaling of impairment using these methods. The Since chronic pain is reportedly accompanied by depression in regression between subjective report of pain and the predicted the vast majority of cases, it was important to investigate the rela- level of pain using a small subset of QEEG features was found tionship between the magnitude of activation in pain related ROIs to be closest at the extremes of the scale, while self-report in and the self-report of depression. Depression ratings (BDI) showed the middle range (3-6 on VAS, for example) shows larger dif- no significant correlation with mean z-scores in any ROIs, sup- ferences between self-rated and predicted values. Since the mid porting the fact that pain state could be reliably measured in the region of the scale is where patients report the most difficulty presence of depression. Similarly, only one ROI, the medial fron- in assigning a number, coupled with the reported lack of test- tal cortex, was found to have a significant correlation with anxiety retest reliability of the VAS, suggests that it is in this range that self-ratings (BAI). This is consistent with reports that cerebral an objective measure derived from overactivation of the pain blood flow in this region was found to be inversely correlated with network might have the greatest clinical utility. Further studies anxiety self-rating in a study delivering painful stimuli.42 might incorporate the referring physicians’ independent esti- Based on electrophysiological studies, a model has been pro- mates of level of pain as related to their review of the history, posed to explain the underlying pathophysiology of pain,18,19,43 clinical presentation, and imaging in the patient. This would which shares common electrophysiological characteristics, based allow a comparison between the self-ratings and the predicted on activity in the thalamocortical system. In such a model, persis- ratings from the QEEG, especially when divergent. While this tent thalamic slow wave activity serves as a trigger for TCD, in regression supports the potential for a predictive relationship which a region of the cortex oscillates at low frequency (theta), between QEEG features and subjective report of pain level, it surrounded by beta or gamma activity (edge effect) and that this Prichep et al 9

Table 2. Mean and Standard Deviation Mean Voxel Values for High and Low Pain Groups for Each Region of Interest. It is important to note that the probability associated with group mean z-scores would be computed times the square root of the n of the group.

Region of Interest Mean (sd) High Mean (sd) Low t-Value Perm p-Value Anterior cingulate 0.67 (1.29) 0.60 (1.23) 0.82 0.41 Mid cingulate 1.00 (1.37) 0.90 (1.41) 1.09 0.29 Posterior Cingulate 0.86 (1.50) 0.72 (1.56) 1.38 0.17 L anterior insula 0.66 (1.36) 0.48 (1.42) 1.84 0.06* L posterior insula 1.03 (1.61) 0.77(1.60) 2.33 0.02* R anterior insula 0.43 (1.51) 0.57 (1.34) −1.34 0.17 R posterior insula 0.62 (1.65) 0.67 (1.42) −0.47 0.64 L S1 1.38 (1.44) 1.01 (1.59) 3.55 0.0002** R S1 0.88 (1.50) 0.89 (1.47) −0.10 0.93 L inferior parietal lobule 1.44 (1.55) 1.06 (1.62) 3.46 0.0002** R inferior parietal lobule 0.80 (1.59) 0.82 (1.40) −0.21 0.83 L DLPFC 1.14 (1.42) 0.75 (1.37) 4.05 <0.0001** R DLPFC 0.82 (1.36) 0.80 (1.25) 0.20 0.83 SMA 0.78 (1.30) 0.68 (1.42) 1.04 0.31 Medial frontal gyrus 0.44 (1.28) 0.26 (1.21) 2.17 0.03* L parahippocampal gyrus 0.68 (1.56) 0.57 (1.52) 1.04 0.29 (hippocampus/amygdala) R parahippocampal gyrus 0.51(1.57) 0.53 (1.43) −0.42 0.67 (hippocampus/amygdala) Thalamus (all regions) 0.63 (1.56) 0.64 (1.53) −0.09 0.93 L caudate 0.75 (1.46) 0.69 (1.46) 0.59 0.55 R caudate 0.61 (1.44) 0.660 (1.42) −0.50 0.63

Abbreviations: L, left; R, right; DLPFC, dorsolateral prefrontal cortex; SMA, supplementary motor area. a t-value and corrected P values are shown for each region of interest. *.001 < P < .05. **P < .001.

EEG sources of the frequency specific abnormalities (theta and beta) occur in regions of the pain network, including the thalamus, TCD has been hypothesized to be the pacing/oscillatory mecha- nism that perpetuates the painful condition.16,17,44,45 The peak of maximal abnormalities reported in this study were found to be fre- quency specific in the theta/low alpha band in regions of the pain network and were found to show more overactivation in high pain compared with low pain states, consistent with such a model. Limitations of this study include the use of 19-lead EEG as input to the source localization. While there is literature that demonstrates concordance between source findings based on 19 leads and other neuroimaging tools46,47 studies of source localization when used to localize epileptic sources report that LORETA can lead to oversmoothed solutions.48 Another limi- tation of this study is the focus on neuropathic chronic pain patients and suggests the need for further studies which include more types of pain (eg, musculoskeletal, visceral, migraine) to explore the robustness of the findings. However, it is noted that these findings are consistent with reports in the literature where Figure 4. Preliminary prediction equation based multivariate other types of pain patients were studied. Also, the heterogene- stepwise logistic regression using regression selected quantitative ity within this study population related to chronicity and side of EEG features from a limited montage. The x-axis is the reported pain score on the visual analog scale and the y-axis is the predicted pain would be expected to add variance to the population statis- value from the regression equation based on EEG features. Open tics, resulting in the reported findings being conservative. circles are individual patients (n = 67). Likewise, potential medication effects, although previously reported not be a factor in the frequency specific overactivation activity disrupts (disconnects) normal circuit function. In the con- reported, present concern as it would be considered medically text of this and other studies in neuropathic pain patients when the unethical to prohibit these patients from taking analgesics. 10 Clinical EEG and Neuroscience 00(0)

Caution should be used in interpreting the results of the pre- 2. Iannetti G, Mouraux A. From the neuromatrix to the pain matrix liminary data presented here on the relationship between self- (and back). Exp Brain Res. 2010;205:1-12. report and predicted pain values. These data remain to be 3. Bushnell MC, Apkarian AV. Representation of pain in the brain. expanded and replicated in future studies. In: McMahon S, Koltzenburg M, eds. Textbook of Pain. 5th ed. In summary, this study demonstrates significant overactiva- Philadelphia, PA: Churchill Livingstone; 2005:267-289. 4. Peyron R, Laurent B, Garcia-Larrea L. Functional imaging tion of the pain network using quantitative EEG in a large pop- of brain responses to pain. A review and meta-analysis. Clin ulation of chronic neuropathic pain patients lending support to Neurophysiol. 2000;30:263-288. a TCD model for the pathophysiology of chronic pain. The 5. Derbyshire SW, Jones AK. Cerebral responses to a continual concordance with conventional functional neuroimaging find- tonic pain stimulus measured using positron emission tomogra- ings suggests the possible clinical utility of such a methodol- phy. Pain. 1998;76:127-135. ogy, which is readily available, cost-effective, and noninvasive 6. deCharms RC, Maeda F, Glover GH, et al. Control over brain for the objective evaluation of pain at the point of care. These activation and pain learned by using realtime function. Proc Natl data also support the potential to derive an objective quantita- Acad Sci U S A. 2005;102:18626-18631. tive measure of the severity of pain using information extracted 7. Mohr C, Binkofski F, Erdmann C, Buchel C, Helmchen C. The from a multivariate descriptor of the abnormal brain activity anterior cingulate cortex contains distinct areas dissociating activation in the regions of the pain network. Such an index (or external from self-administered painful stimulation: a parametric fMRI study. Pain. 2005;114:347-357. biomarker) of pain could assist the clinician in better assess- 8. Davis KD, Taylor SJ, Crawley AP, Wood ML, Mikulis DJ. ment of subjective, self-report of pain and its intensity, helping Functional MRI of pain- and attention-related activations in the minimize over- and under- use of medications and help to opti- human cingulate cortex. J Neurophysiol. 1997;77:3370-3380. mize the care and evaluation of care in chronic pain patients. 9. Vogt BA. Pain and emotion interactions in subregions of the cin- gulate gyrus. Nat Rev Neurosci. 2005;6:533-544. Acknowledgments 10. Bromm B. Brain images of pain. News Physiol Sci. 2001;16: The authors acknowledge the patients who despite being in pain vol- 244-249. unteered to participate in this important research. They also acknowl- 11. Pinheiro ES, de Queirós FC, Montoya P, et al. Electgroen­ edge Robert Isenhart for his thoughtful input and E.R. John whose cephalographic patterns in chronic pain: a systematic review of the legacy has inspired this work. literature. PLoS One. 2016;11:e0149085. 12. Babiloni C, Brancucci A, Arendt-Nielsen L, et al. Cortical sen- sorimotor interactions during the expectancy of a go/no-go task: Author Contributions effects of painful stimuli. Behav Neurosci. 2004;118:925-935. LSP contributed to conception and design; contributed to analysis and 13. Chang P, Arendt-Nielsen L, Graven-Nielsen T, Svensson P, Chen interpretation; drafted manuscript; critically revised manuscript; gave A. Comparative EEG activation to skin pain and muscle pain final approval; agrees to be accountable for all aspects of work ensur- induced by capsaicin injection. Int J Psychophysiol. 2004;51: ing integrity and accuracy. JS contributed to data analysis. HM con- 117-126. tributed to data acquisition. EMH contributed to conception and 14. Chang JW, Choi JY, Lee BW, Kang UJ, Chung SS. Unilaterial design; contributed to interpretation; critically revised manuscript; globus pallidus internus stimulation improves delayed onset post- gave final approval. traumatic cervical dystonia with an ipsilateral focal basal ganglia lesion. J Neurol Neurosurg Psychiatry. 2002;73:588-590. Declaration of Conflicting Interests 15. Fallon N, Chiu YH, Li X, Nurmikko TJ, Stancak A. Ipsilateral cortical activation in fibromyalgia patients during brushing cor- The author(s) declared the following conflicts of interest with relates with symptom severity. Clin Neurophysiol. 2013;124: respect to the research, authorship, and/or publication of this article: 154-163. Drs Prichep and Hiesiger are inventors of a patent assigned to NYU 16. Stern J, Jeanmonod D, Sarnthein J. Persistent EEG overacti- that was licensed to the company (Advanced Neural Metrics) at the vation in the cortical pain matrix of neurogenic pain patients. time they funded this study (in part), this intellectual property was Neuroimage. 2006;31:721-731. not used in this work. The funding company had no input to the 17. Sarnthein J, Stern J, Aufenberg C, Roussen V, Jeanmonod D. study design, conduct of the study, analysis of the data, or writing of Increased EEG power and slowed dominant frequency inpatients the manuscript. with neurogenic pain. Brain. 2006;129:55-64. 18. Jeanmonod D, Schulman J, Ramirez R, et al. Neuropsychiatric Funding thalamocortical dysrhythmia: surgical implications. Neurosurg The author(s) disclosed receipt of the following financial support for Clin N Am. 2003;14:251-265. the research, authorship, and/or publication of this article: Funded in 19. Llinas R, Ribary U, Jeanmonod D, Kronberg E, Mitra P. part by Advanced Neural Metrics. Thalamocortical dysrhythmia: a neurological and neuropsychi- atric syndrome characterized by magnetoencephalography. Proc Natl Acad Sci U S A. 1999;96:15222-15227. References 20. Gram M, Erlenwein J, Petzke F, et al. Prediction of postopera- 1. Melzack R, Wall PD. Pain mechanisms: a new theory. Science. tive opioid analygesia using clinical-experimental parameters and 1965;150:971-979. . Eur J Pain. 2017;21:264-277. Prichep et al 11

21. Prichep LS, John ER, Howard B, Merkin H, Hiesiger E. 36. Evans AC, Collins DL, Neelin P, MacDonald D, Kamber M, Evaluation of the pain matrix using EEG source localization: a Marrett TS. Three-dimensional correlative imaging: appli- feasibility study. Pain Med. 2011;12:1241-1248. cations in mapping. In: Thatcher R, Hallet M, 22. Fairbank JCT, Pynsent PB. The Oswestry Disability Index. Spine. Zeffiro T, John ER, Huerta M, eds. Functional Neuroimaging: 2000;25:2940-2953. Technical Foundations. San Diego, CA: Academic Press; 23. Backonja MM, Krause SJ. Neuropathic Pain Questionnaire–short 1994:145-161. form. Clin J Pain. 2003;19:315-316. 37. Worsley KJ, Marrett S, Neelin P, Vandal AC, Friston KJ, Evans 24. Beck AT, War CH, Mendelson M, Mock JE, Erbaugh J. An inven- AC. A unified statistical approach for determining significant tory for measuring depression. Arch Gen Psychiatry. 1961;4: signals in images of cerebral activation. Hum Brain Mapp. 561-571. 1995;4:58-73. 25. Beck AT, Brown G, Epstein N, Steer RA. An inventory for mea- 38. John ER, Prichep LS, Easton P. Normative data banks and suring clinical anxiety: psychometric properties. J Consult Clin neurometrics: basic concepts, methods and results of norm Psychol. 1988;56:893-897. construction. In: Gevins AS, Remond A, eds. Handbook of 26. Prichep LS. Use of normative databases and statistical methods in Electroencephalography and Clinical , Vol. I. demonstrating clinical utility of QEEG: importance and cautions. Amsterdam, Netherlands: Elsevier; 1987:449-495. Clin EEG Neurosci. 2005;36:82-87. 39. Lorenz J, Minoshima S, Casey KL. Keeping pain out of mind: 27. Thatcher RW, North D, Biver C. Evaluation and validity of a LORETA the role of the dorsolateral prefrontal cortex in pain modulation. normative EEG database. Clin EEG Neurosci. 2005;36:116-122. Brain. 2003;126:1079-1091. 28. Jobert M, Wilson FJ, Ruigt GS, Brunovsky M, Prichep LS, 40. Favilla S, Huber A, Pagnoni G, et al. Ranking brain areas encod- Drinkenburg WH. Guidelines for the recording and evaluation ing the perceived level of pain from fMRI data. Neuroimage. of pharmaco-EEG data in man: the International Pharmaco-EEG 2014;90:153-162. Society (IPEG). Neuropsychobiology. 2012;66:201-220. 41. Coghill RC, Sang CN, Maisog JM, Iadarola MJ. Pain intensity 29. John ER, Prichep LS. Principles of neurometrics and neurometric processing within the human brain: a bilateral, distributed mecha- analysis of EEG and evoked potentials. In: Neidermeyer E, Lopes nism. J Neurophysiol. 1999;82:1934-1943. 42. Simpson J, Drevets W, Snyder A, Gusnard D, Raichle M. Da Silva F, eds. EEG: Basics Principles, Clinical Applications Emotion-induced changes in human medial prefrontal cor- and Related Fields. 3rd ed. Baltimore, MD: Williams & Wilkins; tex: II. During anticipatory anxiety. Proc Natl Acad Sci U S A. 1993:989-1003. 2001;98:688-693. 30. John ER, Prichep LS, Friedman J, Easton P. Neurometrics: 43. Walton KD, Dubois M, Llinas RR. Abnormal thalmocortical computer-assisted differential diagnosis of brain dysfunctions. activity in patients with complex regional pain syndrome (CRPS) Science. 1988;293:162-169. type I. Pain. 2010;150:41-51. 31. Pascual-Marqui RD, Lehmann D, Koukkou M, et al. Assessing 44. Fallon N, Chiu Y, Nurmillo T, Stancak A. Altered theta oscil- interactions in the brain with exact low-resolution electromag- lations in resting EEG of fibromyalgia syndrome patients netic tomography. Philos Trans R Soc A: Math Phys Eng Sci. [published online July 31, 2017]. Eur J Pain. doi:10.1002/ 2011;369:3768-3784. ejp.1076. 32. Pascual-Marqui RD, Michel CM, Lehmann D. Low resolution 45. Lim M, Kim JS, Kim D, Chung C. Increased low- and high-fre- electromagnetic tomography: a new method for localizing electri- quency oscillatory activity in the prefrontal cortex of fibromyal- cal activity in the brain. Int J Psychophysiol. 1994;18:49-65. gia patients. Front Hum Neurosci. 2016;10:111. 33. Fuchs M, Kastner J, Wagner M, Hawes S, Ebersole J. A standard- 46. Benar C, Grova C, Kobayashi E, et al. EEG-fMRI of epileptic ized boundary element method volume conductor model. Clin spikes: concordance with EEG source localization and intracra- Neurophysiol. 2002;113:702-712. nial EEG. Neuroimage. 2006;30:1161-1170. 34. Mazziotta J, Toga A, Evans A, et al. A probabilistic atlas and 47. Vitacco D, Brandis D, Pascual-Marqui R, Martin E. reference system for the human brain: International Consortium Correspondence of event related potential tomography and func- for (ICBM). Philos Trans R Soc Lond B Biol Sci. tional magnetic resonance imaging during language processing. 2001;356:1293-1322. Hum Brain Mapp. 2002;17:4-12. 35. Lancaster JL, Woldorff MG, Parsons LM, et al. Automated 48. Michel CM, Murray MM, Lantz G, Gonzalez S, Spinelli Talairach atlas labels for functional brain mapping. Hum Brain L, de Peralta RG. EEG source imaging. Clin Neurophysiol. Mapp. 2000;10:120-131. 2004;115:2195-2222.