Research Article Novel Burst Suppression Segmentation in...

Research ArticleNovel Burst Suppression Segmentation inthe Joint Time-Frequency Domain for EEG inTreatment of Status Epilepticus

Jaeyun Lee1 Woo-Jin Song1 Hyang Woon Lee2 and Hyun-Chool Shin3

1Department of Electrical Engineering POSTECH Pohang Gyeongbuk Republic of Korea2Department of Neurology Ewha Womans University School of Medicine and Ewha Medical Research InstituteSeoul Republic of Korea3Department of Electronic Engineering Soongsil University Seoul Republic of Korea

Correspondence should be addressed to Hyun-Chool Shin shinhcssuackr

Received 9 June 2016 Revised 10 September 2016 Accepted 5 October 2016

Academic Editor Valeri Makarov

Copyright copy 2016 Jaeyun Lee et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

We developed amethod to distinguish bursts and suppressions for EEG burst suppression from the treatments of status epilepticusemploying the joint time-frequency domain We obtained the feature used in the proposed method from the joint use of the timeand frequency domains and we estimated the decision as to whether the measured EEG was a burst segment or suppressionsegment by the maximum likelihood estimation We evaluated the performance of the proposed method in terms of its accordancewith the visual scores and estimation of the burst suppression ratio The accuracy was higher than the sole use of the time orfrequency domains as well as conventional methods conducted in the time domain In addition probabilistic modeling provided amore simplified optimization than conventional methods Burst suppression quantification necessitated precise burst suppressionsegmentation with an easy optimization therefore the excellent discrimination and the easy optimization of burst suppression bythe proposed method appear to be beneficial

1 Introduction

Electroencephalogram (EEG) burst suppression representsan inactivated EEG pattern in which the aperiodic alter-nation of an isoelectric pattern (suppression) and a highvoltage pattern (bursts) appears The pattern found from ananesthetized catrsquos brain for the first time [1] would accom-pany the occurrence of a serious reduction in the brainrsquosactivity and metabolic rate [2] frequently seen from patientswith postanoxic encephalopathy or status epilepticus underparenteral benzodiazepine treatments such asmidazolam [3]those under general anesthesia [4] those with hypothermia[5] or those in a coma [6] or from neonates [7] In the caseof a burst suppression caused by anesthesia the duration ofthe burst or suppression varies depending on the level ofanesthetic concentration with high levels identifying theirrelevance to a long duration of suppression [6 8 9] Inaddition a long duration of suppression has also identified

its relevance with a worsening prognosis in certain cases (egbrain injuries caused by asphyxia) and further the progres-sion of burst suppression has provided important prognosticinformation in previously conducted studies [10ndash12] Accord-ingly researchers have developed methods for quantificationof burst suppression through calculations of the occupancyratio of suppression in burst suppression (BSR) [13] analysesusing the duration of suppression (interbursts interval IBI)[14] and quantitation of a burst suppression probability [15]

The first step in quantifying the depth of burst suppres-sion involves detecting the burst and suppression to distin-guish them thereby called burst suppression segmentation[16] The current practice employed for this purpose involvesamethod based on visual detection being a rough assessmentand time-consuming task subjective to the viewersrsquo interpre-tation Thus researchers initiated an algorithmic approachto burst suppression segmentation from the method used tomanually set the threshold in an EEG pattern [13 17 18]

Hindawi Publishing CorporationComputational and Mathematical Methods in MedicineVolume 2016 Article ID 2684731 12 pageshttpdxdoiorg10115520162684731

2 Computational and Mathematical Methods in Medicine

Table 1 Clinical manifestations of the patients

Patientnumber Age Gender Etiology Diagnosis EEG findings Antiepileptic

medications

1 39 M Viralmeningoencephalitis Status epilepticus Burst suppression Midazolam IVlowast

2 72 M Cardiac arrest Postanoxicencephalopathy Burst suppression None

3 82 F Prolonged hypoxemiadue to pneumonia

Postanoxicencephalopathy Burst suppression Midazolam IVlowast

4 53 M Viral encephalitis Status epilepticus Burst suppression Midazolam IVlowast

M male F female IVlowast intravenous continuous infusion

evolving into the current methods in use The methods ofburst suppression segmentation developed thus far mainlyinvolve detecting burst events by using certain features suchas Shannon entropy [19 20] a nonlinear energy operator[21] line length [14] a voltage envelope [22] and varianceusing recursive-variance estimation [16] These all representfeatures employed in the time domain as well as occasionallythe basic features of the frequency domain like 3 or 10Hzpower or mean power spectral density (PSD) [19 20 23]Previous studies also compared the performances by the basicfeatures and segmentation methods [19 20 24]

In this study we applied the joint use of features extractedfrom the time and frequency domains to burst suppressionsegmentation to enhance the accuracy of segmentation con-ventionally conducted solely in the time domain We thus setthe EEG data and PSD thereof as the time-frequency vectorto extract features to be used in the time-frequency domainWe then applied the Shannon entropy the Tsallis entropy andthe regularity taken as features in broadly used conventionalquantitative EEG (qEEG) analysis [25ndash28] to the time-frequency vector After modeling of the distributions of burstand suppression of each feature into a Gaussian distributionwe employed the method of maximum likelihood estimation(MLE) to conduct the burst suppression segmentation Weassessed the accuracy of the proposed method by comparingthe accordance to visually scored burst suppressions Inaddition we employed the BSR as an indicator most widelyused for quantification of burst suppression and evaluatedthe closeness of estimatedBSR to the true BSRs obtained fromvisual scores In this study the method presented revealedimproved accuracy and precise BSR estimation Besides wesolved the problems that reside in the optimization processof conventionalmethods through the application of Gaussianmodeling and the MLE

2 Methods

21 Data Acquisition In this study we used the 11 multi-channel EEG data sets recorded from the Mokdong Hospitalof Ewha Womans University The patients were recruitedbased on the discharge database of the patients who hadbeen hospitalized as status epilepticus and EEG monitoringwas performed at the intensive care unit in Ewha WomansUniversity Mokdong Hospital from July 2012 to June 2015

Time (min)

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

EEG

(V

)

3252151050

Figure 1 The different burst suppression characteristics in eightEEGs from patient number 1 (Table 1)

A total of 122 patients who met the criteria were enrolledso far and their medical records for the age at onsetetiology comorbidity consciousness level before treatmentoverall duration of status epilepticus brain imaging andvisual interpretation of EEG were reviewed Among themfour patients whose EEG clearly showed a burst suppressionpattern were selected for the qEEG analysis in this studyusing artifact-free EEG segments for at least 20 minuteschosen by an expert neurologist (H W Lee) Among thosefour patients with a burst suppression EEG pattern threepatients were male while one was female (male female =75 25) and age ranged between 39 and 82 (mean 615 plusmn192) years (Table 1) Specifically eight EEGs were recordedfrom patient number 1 (Table 1) on different days whilewe obtained the rest of the three EEGs from three differentpatients Figure 1 shows the initial 3 minutes of the eightEEGs from patient number 1 We can observe the differentburst suppression characteristics in the EEG evolution The11 EEGs were recorded from 21 electrode locations basedon the international 10ndash20 system with 200Hz of samplingfrequency The 21 channels corresponding to each electrodewere Fp1 Fp2 Fz F3 F4 F7 F8 T3 T4 T5 T6 Cz C3 C4Pz P3 P4 O1 O2 A1 and A2

Computational and Mathematical Methods in Medicine 3

100500

minus50minus100

100806040200

EEG

(V

)

(a)

100806040200

5040302010

Freq

uenc

y (H

z)

Time (s)

20100minus10minus20minus30 Po

wer

(dB)

(b)

Figure 2 (a) Example of 120 s EEG burst suppression (median over all the channels) and (b) its PSD by spectrogram plot

22 Time-Frequency Representation of Burst Suppression Let119909119899(119894) 119894 = 1 2 119871 denote the raw sampled EEG signalof the 119899th channel To remove artifacts in the EEG signalswe selected 119909(119894) 119894 = 1 2 119871 with the median valueover all channelsThemedian filtering over channelsmay losetimefrequency characteristics of the original EEG but burstsuppression patterns generally have synchrony over channelsThus the median value does not affect the burst suppressioncharacteristics seriously as shown in Figure 2(a) Then weobtained the power spectral density (PSD) of the EEG signal119909(119894) To accomplish this task we defined the119898th block of 119909(119894)as follows

x119898 = 119909 (119894) 119894 = 1 + 119898Δ 2 + 119898Δ 119873 + 119898Δ (1)

where 119873 indicates the block width and Δ means the slidingstep We calculated PSD P119898 by the short-time Fouriertransform (STFT) as follows

X119898 = STFT x119898 P119898 = 1003816100381610038161003816X11989810038161003816100381610038162 (2)

Figure 2 shows an example of EEG burst suppressionpatterns in both the time and the frequency domains Notethat we observed clear peaks corresponding to burst patternsin the frequency domain This observation implied that wecould detect the burst suppression patterns in the frequencydomain as well as in the time domain Thus we can expectthat a joint analysis in both the time and the frequencydomain may improve the accuracy of the burst suppressionsegmentation

However researchers executemost conventional segmen-tation or burst detection instances in the time domain Todo burst suppression segmentation in the time-frequencydomain we newly defined a joint time-frequency vector f119898as follows

f119898 = [x119898P119898] (3)

Then we extracted entropy [19 20 25ndash27] and regularity [28]features (widely used in the time domain detection of qEEG)from f119898

23 Entropy and Regularity Features in the Time-FrequencyDomain Entropy serves as a method to quantify theorderdisorder in signals typically used for measuring thevariety existing in burst suppression patterns Most entropy-based EEG analyses use Shannon entropy [19 20] or Tsallisentropy [25ndash27] To calculate Shannon and Tsallis entropywe first estimated the probability mass function (PMF) 119901119905and 119901119891 of x119898 and P119898 respectively in (3) To estimate PMFwe introduced disjoint amplitude intervals 119868119905(119896) and 119868119891(119896)such that x119898 = ⋃119896=119872119905

119896=1119868119905(119896) and P119898 = ⋃119896=119872119891

119896=1119868119891(119896) Then

the probability that the signal belonged to the 119896th interval isthe ratio between the number of the samples in the intervalsand the total number of samples that is

[119901119905 (119896)119901119891 (119896)] = 1119873 [The number of samples in 119868119905 (119896)The number of samples in 119868119891 (119896)] (4)

Then we calculated Shannon entropy 119878(119898) and Tsallisentropy 119879(119898) as

119878 (119898) = [[[[[[[

minus119872119905sum119897=1

119901119905 (119897) ln119901119905 (119897)minus119872119891sum119897=1

119901119891 (119897) ln119901119891 (119897)]]]]]]]

119879 (119898) = 1119902 minus 1[[[[[[[

1 minus 119872119905sum119897=1

119901119905 (119897)1199021 minus 119872119891sum119897=1

119901119891 (119897)119902]]]]]]]

(5)

where 119902 is a positive real valueIn addition regularity could measure how smoothly or

consistently the signals change For the descending-ordered


3

25

2

15

1

05

Time domain

Freq

uenc

y do

mai

n3 5

3252151

BurstSuppression

(a)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

10

8

6

4

2

0765432

(b)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

035

03

025

02

015

01

0050605040302010

(c)

Figure 3 The example features of burst suppression in joint time-frequency domain The features used are (a) Shannon entropy (b) Tsallisentropy and (c) regularity Dotted line a decision boundary of the sole use of time domain Dashed line a decision boundary of the sole useof frequency domain The examined burst and suppression segments are identified by the experts

data 119889119905(119894) and 119889119891(119894) of x1198982 and P119898 respectively we obtainedthe regularity of f119898 in (3) as

119877 (119898) =[[[[[[[[[

radic sum119873119894=1 1198942119889119905 (119894)(11987323)sum119873119894=1 119889119905 (119894)radic sum1198731198941015840=1 1198942119889119891 (119894)(11987323)sum119873119894=1 119889119891 (119894)

]]]]]]]]] (6)

The feature distributions of 119878(119898) 119879(119898) and 119877(119898) forthe burst suppression pattern in the time-frequency domainare represented in Figures 3(a)ndash3(c) respectively in whichthe patterns of burst and suppression corresponding to all

three features are distinguished clearly However since somebursts and suppressions deviated from each pattern theperformance of the segmentation appeared to potentiallybe degraded thereby By the single-domain approach thatis the sole use of the time or frequency domain thedecision boundary would be a straight line The vertical linesrepresented in Figures 3(a)ndash3(c) correspond to the decisionboundary in the time domain and in this case the errorsof either a falsely alarmed burst or a missing burst couldoccur The horizontal lines represented in Figures 3(a)ndash3(c)imply the decision boundary in the frequency domain andin this case could also be accompanied by misses and falsealarmsTherefore we took into account the use of a nonlineardecision boundary in the joint time-frequency domain to


improve the accuracy of the burst suppression segmentationThere exist many nonlinear classifiers for example artificialneural networks or support vector machines [19 20] Amongthese nonlinear classifiers we employed the MLE being theprobabilistically optimal classifier

24 Burst Suppression Segmentation To conduct the MLEwe needed the corresponding probability distributionmodelThus for this purpose we modeled the distributions of fea-tures corresponding to the respective burst and suppressioninto the Gaussian distribution Let the mean and covarianceof a certain feature (ie one of the Shannon entropy Tsallisentropy and regularity) corresponding to burst patterns be120583119861 and C119861 respectively Similarly let 120583119878 and C119878 be the meanand covariance of the features corresponding to suppressionpatterns Then we could express the Gaussian distributionmodel for bursts and suppressions as

119901119861 (z) = (2120587)minus1 1003816100381610038161003816C1198611003816100381610038161003816minus12 exp(minus12 (z minus 120583119861)119905 Cminus1119861 (z minus 120583119861)) (7)


respectively where | sdot | is the determinant of a matrix and (sdot)119905is the transpose of a vector

Equations (7) and (8) denote the likelihood of the burstand suppression of feature z and in this study we supposedthe decision to maximize the likelihood Let 120579 isin 119861 119878 thenwe expressed the decision rule as

= argmax120579

119901120579 (z) 120579 isin 119861 119878 (9)

Figure 4(a) represents Gaussian distributions of the Shan-non entropy generated by (7) and (8) as the two concentriccircles labeled as burst and suppression respectively Forthe Gaussian modeling we used the data in Figure 3Then we represented the Shannon entropies of the burstand suppression obtained from the burst suppression inFigure 4(a) as upward-pointing triangles and downward-pointing triangles respectively all identified and distin-guished clearly by the solid nonlinear line the decisionboundary determined by (9) In this case we generated thefalse alarms and misses marked with filled triangle by thevertical and horizontal lines fixed at the optimal thresholdSpecifically the two filled downward-pointing triangles onthe right side of Figure 4(a) represent errors generated bythe misclassification of a suppression into a burst with thesole application of the frequency domain while the filleddownward-pointing triangle in the center represents an errorof the misclassification of a suppression into a burst withthe sole use of the time domain The filled upward-pointingtriangle on the left side of Figure 4(a) also represents an errorowing to the misclassification of a burst into a suppressionwith the sole application of the frequency domain Figures4(b) and 4(c) represent the plots of cases using the Tsallisentropy and regularity respectively For these cases (9) alsorendered a clearly distinguished classification however we

generated the miss or false alarm in cases of the sole use ofeither the time or the frequency domains

25 Performance Evaluation We derived the results of theburst suppression segmentation from the 11 consecutive EEGsrecorded from 4 patients suffering from status epilepticusWe set the values of block width 119873 and sliding step Δ as140 and 40 respectively implying the number of samplescorresponding to 07 s and 02 s respectively The value 119873 ischosen to include at most one burst segment and the value Δis chosen to be smaller than 1198732 The frequency resolutioninvolved in STFT calculation is chosen to be 119873 and thisresolution was enough to identify the PSD distribution overthe frequency axis We set the values of 119872119905 and 119872119891 (theparameters used for the calculation of entropies) as 20 and 40respectively with 119902 = 05 for Tsallis entropy The parametersrelated to the features (ie 119872119905 119872119891 and 119902) were chosen tohave enough divisibility of burst clusters and suppressionclusters From the EEGs of the total of at least 20min each(mean 2155 plusmn 061min) we used 10min duration to modelthe Gaussian distributions (7) and (8) Depending on thedata the number of bursts in 10min duration varies Forsparse and dense bursting about 80 and 220 burstings wereobserved in 10min respectively Then we used the latter halfafter the initial duration of sim10min for the identification ofthe performance of the burst suppression segmentation

To evaluate the overall accuracy of the burst suppressionsegmentation for the 11 EEG data sets we employed thesensitivity specificity and accuracy based on the accordanceto the visual scores [14 21] Sensitivity means the ratio ofsamples detected as true bursts among burst samples by thealgorithm and the specificity denotes the ratio of samplesdetected as true suppressions among suppression samplesby the algorithm The accuracy means the ratio of properlydetected samples among whole samples determined bytaking the sensitivity and specificity into account Thereforethe values for sensitivity specificity and accuracy increasedalong with the improved accuracy of the burst suppressionsegmentation

As a result of the burst suppression segmentation wecould obtain the binarized burst suppression (BBS) patterncomprising sample units denoted as either 1 (burst sample)or 0 (suppression sample) and by using the BBS pattern wecould calculate the BSR known most widely as the measureof burst suppression [15 16 19] BSR is calculated as a ratioof the number of zeros in BBS in a certain interval to thenumber of samples in the interval that is a suppressionratio Known also to be correlated with cerebral metabolism[9] the BSR can be used for various patient monitoringapplications including treatment of status epilepticus [29]and monitoring the depth of anesthesia [30] We computedBSR at every 15-second interval with one sample slidingwindow Then we exploited the difference with visuallyscored BSR (true BSR) to identify the performance of burstsuppression segmentation

To evaluate the similarity between the true BSR andestimated BSR and to statistically analyze the results for all the


Freq

uenc

y do

mai

n

Time domain3252151

Burst

Suppression

3

25

2

15

1

05

35

(a)

Freq

uenc

y do

mai

n

Time domain765432

9

8

7

6

5

4

3

2

1

Burst

Suppression

(b)

Freq

uenc

y do

mai

n

035

03

025

02

015

01

005

Time domain060504

04

0302010

Burst

Suppression

(c)

Figure 4 The example Gaussian models in the joint time-frequency domain The features used are (a) Shannon entropy (b) Tsallis entropyand (c) regularity The feature segments (triangles) are different segments from Figure 3 Solid line the optimal decision boundary by MLEDotted line the optimal decision boundary in the time domain solely Dashed line the optimal decision boundary in the frequency domainsolely Upward-pointing and downward-pointing empty triangles detected true bursts and suppressions by all the boundaries Upward-pointing filled triangle a burst missed in time or frequency domain solely Downward-pointing filled triangles falsely alarmed suppressionsin time or frequency domain solely

EEG sets and all the features we calculated the root-mean-square error (RMSE) between the true BSR and estimatedBSR as in the following

RMSE = radic 111987110158401198711015840sum119894=1

(true BSR (119894) minus estimated BSR (119894))2 (10)

where 1198711015840 is the number of BSR samples being sim10minduration for performance evaluation A low RMSE for afeaturemeans a good BSR estimation of the feature thereforestatistically analyzed RMSE values can provide the usefulnessof the segmentation methods

By one visual score the results can be meaningless Thisis because there is no gold standard for burst suppressionsegmentation and the results from a certain visual score

are subjective Therefore to evaluate the performance ofthe proposed method independent visual scores from morethan one expert are needed The visual scores in the resultswere from two clinical experts and we exhibited agreement(ie sensitivity specificity and accuracy) and BSR estimationresults for the two visual scores rater 1 and rater 2 in Tables2 and 4 In addition we evaluated and exhibited interrateragreements in Table 3

3 Results

31 Comparison with Time Domain Detection We representthe results of the burst suppression segmentation obtainedfrom the sole use of the time domain for various featuresin Figure 5(a) We depicted the burst suppression pattern


Table 2 Mean (standard deviation) of sensitivity specificity and accuracy by different features and two raters over the 11 EEGs

Features Sensitivity () Specificity () Accuracy ()Rater 1 Rater 2 Avg Rater 1 Rater 2 Avg Rater 1 Rater 2 Avg

Proposedtime-frequencydomain

Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201

Regularity 7442(1327)

9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712

Note Avg average between two raters boldface the two highest means in Avg columns

Table 3 Interrater sensitivity specificity and accuracy (mean plusmnstandard deviation over the 11 EEGs)

Sensitivity () Specificity () Accuracy ()Rater 1 versusRater 2 (true) 8126 plusmn 767 9930 plusmn 080

9534 plusmn 326Rater 2 versusRater 1 (true) 9774 plusmn 120 9456 plusmn 429

with a duration of 33 s on the top and the plots placedthereunder and labeled as 119878 119879 and 119877 represent intervals ofburst detected as blocks by using Shannon entropy Tsallisentropy and regularity respectively The intervals expressedas boxes in each plot represent all intervals identified as falsealarms Therein we show one false alarm two false alarmsand one false alarm corresponding tomethods employing theShannon entropy Tsallis entropy and regularity respectivelyFigure 5(b) shows the results of the burst suppression seg-mentation obtained by the sole use of the frequency domainin the same burst suppression interval as that representedin Figure 5(a) The plot placed on the top represents thePSD portrayed as a spectrogram plot and the plots placedthereunder represent the intervals of detected bursts as in thecase in Figure 5(a) In these plots we expressed all the falsealarms andmisses as boxes and there evoked the onemiss andtwo false alarms the two false alarms and the one false alarmrespectively from the methods using the features of Shannonentropy Tsallis entropy and regularity In Figure 5(c) werepresent the results of the burst suppression segmentationobtained by the use of the proposed time-frequency domainin the same burst suppression intervalThemethods using theShannon entropy and regularity did not generate errors in thesame interval however the method using the Tsallis entropy

rendered one false alarm Thus as expected beforehand themethod using the proposed time-frequency domain resultedin reduced numbers of false alarms and misses generated

In Figure 6 we presented the interval of the burstsdetected by conventional methods on the same burst sup-pression pattern presented in Figure 5 The conventionalmethods compared are those based on the line length (LL)[14] envelope (EV) [22] and nonlinear energy operator(NLEO) [21] all employed in detecting burstsTheplot placedon the top of Figure 6 represents the burst suppression andthe other three plots present bursts detected by conventionalmethods In cases using the EV andNLEOmethods the falsealarms evoked more than the case of the method using LL Incomparison to the proposedmethod in Figure 5 we identifiedthat the burst suppression segmentation by the proposedmethod using the time-frequency domain could provide animproved accuracy

Table 2 represents the mean and standard deviation (std)of the assessment indicators (ie sensitivity specificity andaccuracy) for the whole 11 data sets in correspondencewith each feature We represented the two features corre-sponding to the two respective highest assessment indicatorsin boldface In general the values for sensitivity tendedto become higher with the method employing the time-frequency domain in contrast to the values for specificity thatappeared to remain relatively even For rater 1 the highestaccuracy appeared in the method employing the regularitywith the time-frequency domain andwe obtained the secondhighest accuracy from the method based on LL For rater2 the two highest accuracies appeared in the proposedmethods For all three assessment indicators the methodusing the time-frequency domain tended to show superiorperformance to that of the methods employing only the timedomain


Table 4 RMSE between true BSR and estimated BSR by burst suppression segmentation methods and two raters

FeaturesRMSE between true BSR and estimated BSR (mean plusmn

standard deviation over the 11 EEGs)Rater 1 Rater 2

Proposed time-frequencydomain

Shannon entropy 0111 plusmn 0074 0121 plusmn 0070Tsallis entropy 0121 plusmn 0068 0125 plusmn 0068Regularity 0094 plusmn 0051 0104 plusmn 0063

Time domainShannon entropy 0118 plusmn 0067 0134 plusmn 0064Tsallis entropy 0131 plusmn 0058 0142 plusmn 0054Regularity 0098 plusmn 0049 0113 plusmn 0049

Conventional methodsLine length [14] 0175 plusmn 0053 0152 plusmn 0095Envelope [22] 0210 plusmn 0136 0198 plusmn 0148NLEO [21] 0214 plusmn 0130 0196 plusmn 0161

Note Boldface the lowest means in column

10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)

Figure 5 Example of segmentation results by various features (a) a 33 s burst suppression (median value over all the channels) and detectedbursts in the time domain using Shannon entropy (119878) Tsallis entropy (119879) and regularity (119877) (b) PSD of (a) using a spectrogram plot (top)and detected bursts in the frequency domain using Shannon entropy (119878) Tsallis entropy (119879) and regularity (119877) (c) detected bursts in thejoint time-frequency domain using Shannon entropy (119878) Tsallis entropy (119879) and regularity (119877) Blocks detected bursts boxes false alarmsand misses

Table 3 shows mean and std of sensitivity specificityand accuracy between two visual scores (rater 1 and 2)The sensitivity and specificity labeled by rater 1 versus rater2 (true) were from the assumption that the visual scorefrom rater 2 is truly segmented and vice versa The raters

focused on picking out burst segments in burst suppressionpattern so the sensitivities between rater 1 and rater 2 canbe significantly different In Table 3 the sensitivity labeledas rater 1 versus rater 2 (true) is much smaller than theother sensitivity and this means rater 2 was more sensitive


100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL

Figure 6 Example of 33 s EEG (median value over all the channels)and segmentation results of conventional methods line-length-(LL-) based method envelope- (EV-) based method and nonlinearenergy operator- (NLEO-) based method Boxes detected falsealarms

to choose burst durations Consequently this difference be-tween two visual scores implies that the performance evalua-tion by multiple raters is strongly recommended

32 Burst Suppression Ratio (BSR) The first and the secondplots in Figure 7(a) show the burst suppression pattern andthe true BSRs from the visual scores of rater 1 and rater2 respectively The third plot therein represents the BBSsobtained respectively by the methods using the regularityin the time-frequency domain (Proposed119877) employing theregularity in the time domain (Time119877) and based on LLThethree features selected represent the ones evaluated as beingexcellent in terms of the accuracy We represented the BBSplot with a marker on the points of 1 of the value of BBSgenerally corresponding to the points of the appearance ofthe burst

To exhibit the exactitude of the estimated BSRs for eachfeature we employed the absolute difference ΔBSR betweenthe true BSR and estimated BSR (ie ΔBSR = |(true BSR) minus(estimated BSR)|) The subfigures labeled as ΔBSR (rater 1)andΔBSR (rater 2) of Figure 7(a) exhibitedΔBSRs regardingtheir true BSRs as of rater 1 and rater 2 respectivelyWith the distribution of low values for ΔBSR of Proposed119877in Figure 7(a) we can conclude that it represents the closeestimation of the true BSRs Figures 7(b)-7(c) represent theresults of BSRs corresponding to different burst suppressionpatternsThe trend of the smaller value of true BSR from rater2 than from rater 1 reflects the notion that rater 2wasmoresensitive to choose burst durations Through the three burstsuppression patterns represented the method employing theregularity with the proposed time-frequency domain showsthe small ΔBSR

Table 4 represents the mean RMSE and std over 11 EEGsets by each method of burst suppression segmentation Thelowest mean RMSE appeared from the regularity featureusing the time-frequency analysis and the next lowest oneappeared from the regularity feature solely using the timedomain from both raters Thus the proposed burst sup-pression segmentation methods using the time-frequencyanalysis also rendered excellent results in terms of BSRestimation

4 Conclusion

In this study we exhibited the usefulness of exploitingtime-frequency domain for burst suppression segmentationExisting qEEG features are usually conducted in the timedomain but we newly redefined the features (ie Shannonentropy Tsallis entropy and regularity) in the time-frequencydomain These redefined two-dimensional features formedclusters of bursts and suppressions as in Figure 3 Because ofthe distribution of the clusters the optimal classificationmustnot be a sole use of the time or frequency domainThismeansthat the joint use of the time and frequency domain wasexpected to improve segmentation performance To conductthe segmentation considering the distributions we assumedthat the clusters are modeled as Gaussians Finally usingthe Gaussian distributions burst suppression segmentationis conducted by exploiting MLE which is probabilisticallyoptimal

We compared the method developed in the time-fre-quency domain with the method employing the time domainonly and with those of existing ones defined in the timedomain to derive and analyze the results of the burstsuppression segmentation Seeking precision we used threeassessment indicators comprising sensitivity specificity andaccuracy for the comparison and we improved the accuracyof the method proposed more than the sole use of the timedomain In addition we also identified the accuracy of theproposed method similar to or better than that of existingmethods defined in the time domain To verify the usefulnessof the proposed method we employed BSR which is broadlyused as ameasure of burst suppression and directly calculatedby burst suppression segmentation and evaluated the BSR forall compared methods We evaluated the RMSEs of the trueand estimated BSRs as indicators and the regularity using thetime-frequency domain revealed the best performance withthe lowest mean RMSE among all methods compared witheach other

The method proposed in this study not only appearsbetter than existing methods with respect to burst suppres-sion segmentation but also resolves the problems residingin the process of optimization of the conventional methodsThe burst suppression pattern may vary according to thepatientsrsquo condition or the environment of the measurementso some conventional approaches with a fixed threshold cangenerate severe errors However in our study we employedsufficient instances of burst and suppression segments inthe burst suppression pattern for the Gaussian probabilisticmodeling to solve such issues In contrast to cases of existingalgorithms using user-defined parameters optimized usingROC curves [14 21 22] for which the optimization processcan be time-consuming as a result of redundant calculationsof the sensitivityspecificity for slightly changing user param-eters the probabilistic modeling simplified and replacedthis process In conclusion the improved accuracy and BSRestimation realized by the method proposed in this studydemonstrated that burst suppression segmentation usingthe time-frequency domain appears to be more accurateand useful than that of conventional methods However interms of data used this study possesses a few shortages


BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)

Figure 7Three examples of burst suppression segmentation results and BSR estimates Each subfigure shows burst suppression pattern trueBSR by visual segmentation of two raters BBS which indicates detected bursts as bar plots andΔBSR progress for each true BSRThe featuresused are the proposed method using regularity (Proposed119877) the time domain method using regularity (Time119877) and LL-based method

which are the similar application (ie the treatment of statusepilepticus) and the limited number of patients The burstsuppression patternrsquos occurrence has been reported in manyapplications and compared methods actually have differentapplications (premature neonatal EEGs) from this study(treatments of status epilepticus) In fact our data includedmultiple recordings in different days from one patient sincethe patientrsquos medical condition had been changed everyday during early hospital days which could be a possiblelimitation of this study By usingmore various applications ofburst suppression and increased number of patients we willbe a little more confident about the improved performance ofthe proposedmethod Besides an application or introductionof new or different features to the method proposed in thisstudy appears to be capable of bringing about even morepromising performance

Competing Interests

The authors declare that there are no competing interestsregarding the publication of this paper

Acknowledgments

This research was supported by the Industrial ConvergenceCore Technology Development Program (no 10048474)funded by the Ministry of Trade Industry amp Energy(MOTIE)

References

[1] A J Derbyshire B Rempel A Forbes and E F Lambert ldquoTheeffects of anesthetics on action potentials in the cerebral cortexof the catrdquoAmerican Journal of PhysiologymdashLegacy Content vol116 no 3 pp 577ndash596 1936

[2] F Amzica ldquoBasic physiology of burst-suppressionrdquo Epilepsiavol 50 pp 38ndash39 2009

[3] S Shorvon ldquoThe treatment of status epilepticusrdquo CurrentOpinion in Neurology vol 24 no 2 pp 165ndash170 2011

[4] E N Brown R Lydic and N D Schiff ldquoGeneral anesthesiasleep and comardquoTheNewEngland Journal ofMedicine vol 363no 27 pp 2638ndash2650 2010

[5] D W Marion L E Penrod S F Kelsey et al ldquoTreatment oftraumatic brain injury with moderate hypothermiardquo The NewEngland Journal of Medicine vol 336 no 8 pp 540ndash546 1997


[6] G B Young ldquoThe EEG in comardquo Journal of Clinical Neurophys-iology vol 17 no 5 pp 473ndash485 2000

[7] E Niedermeyer D L Sherman R J Geocadin H C Hansenand D F Hanley ldquoThe burst-suppression electroencephalo-gramrdquo Clinical EEG Electroencephalography vol 30 no 3 pp99ndash105 1999

[8] P C M Vijn and J R Sneyd ldquoIV Anaesthesia and EEG burstsuppression in rats bolus injections and closed-loop infusionsrdquoBritish Journal of Anaesthesia vol 81 no 3 pp 415ndash421 1998

[9] S Ching P L Purdon S Vijayan N J Kopell and E N BrownldquoA neurophysiological-metabolic model for burst suppressionrdquoProceedings of the National Academy of Sciences of the UnitedStates of America vol 109 no 8 pp 3095ndash3100 2012

[10] G B Young J T Wang and J F Connolly ldquoPrognostic deter-mination in anoxic-ischemic and traumatic encephalopathiesrdquoJournal of Clinical Neurophysiology vol 21 no 5 pp 379ndash3902004

[11] B Hallberg K Grossmann M Bartocci andM Blennow ldquoTheprognostic value of early aEEG in asphyxiated infants undergo-ing systemic hypothermia treatmentrdquo Acta Paediatrica vol 99no 4 pp 531ndash536 2010

[12] D Zhang X Jia H Ding D Ye and N V Thakor ldquoAppli-cation of tsallis entropy to EEG quantifying the presence ofburst suppression after asphyxial cardiac arrest in ratsrdquo IEEETransactions on Biomedical Engineering vol 57 no 4 pp 867ndash874 2010

[13] I J Rampil R B Weiskopf J G Brown et al ldquoI653 andisoflurane produce similar dose-related changes in the elec-troencephalogram of pigsrdquo Anesthesiology vol 69 no 3 pp298ndash302 1988

[14] N Koolen K Jansen J Vervisch et al ldquoLine length as a robustmethod to detect high-activity events automated burst detec-tion in premature EEG recordingsrdquo Clinical Neurophysiologyvol 125 no 10 pp 1985ndash1994 2014

[15] J Chemali S Ching P L Purdon K Solt and E N BrownldquoBurst suppression probability algorithms state-space methodsfor tracking EEGburst suppressionrdquo Journal of Neural Engineer-ing vol 10 no 5 Article ID 056017 2013

[16] M B Westover M M Shafi S Ching et al ldquoReal-timesegmentation of burst suppression patterns in critical care EEGmonitoringrdquo Journal of NeuroscienceMethods vol 219 no 1 pp131ndash141 2013

[17] I J Rampil and M J Laster ldquoNo correlation between quanti-tative electroencephalographic measurements and movementresponse to noxious stimuli during isoflurane anesthesia inratsrdquo Anesthesiology vol 77 no 5 pp 920ndash925 1992

[18] T Lipping V Jantti A Yli-Hankala and K HartikainenldquoAdaptive segmentation of burst-suppression pattern in isoflu-rane and enflurane anesthesiardquo International Journal of ClinicalMonitoring and Computing vol 12 no 3 pp 161ndash167 1995

[19] J Lofhede N Lofgren M Thordstein A Flisberg I Kjellmerand K Lindecrantz ldquoClassification of burst and suppressionin the neonatal electroencephalogramrdquo Journal of Neural Engi-neering vol 5 no 4 pp 402ndash410 2008

[20] S Bhattacharyya A Biswas J Mukherjee et al ldquoFeature selec-tion for automatic burst detection in neonatal electroen-cephalogramrdquo IEEE Journal on Emerging and Selected Topics inCircuits and Systems vol 1 no 4 pp 469ndash479 2011

[21] K Palmu N Stevenson S Wikstrom L Hellstrom-Westas SVanhatalo and J Matias Palva ldquoOptimization of an NLEO-based algorithm for automated detection of spontaneous activ-ity transients in early pretermEEGrdquo PhysiologicalMeasurementvol 31 no 11 pp N85ndashN93 2010

[22] W Jennekens L S Ruijs C M L Lommen et al ldquoAutomaticburst detection for the EEG of the preterm infantrdquo PhysiologicalMeasurement vol 32 no 10 pp 1623ndash1637 2011

[23] J Lee W J Song and H C Shin ldquoBurst suppression patternrecognition using time-frequency analysisrdquo inProceedings of theThe 31st International Technical Conference on CircuitsSystemsComputers and Communications (ITC-CSCC rsquo16) pp 799ndash802Tsukuba Japan July 2016

[24] K Murphy N J Stevenson R M Goulding et al ldquoAutomatedanalysis of multi-channel EEG in preterm infantsrdquo ClinicalNeurophysiology vol 126 no 9 pp 1692ndash1702 2015

[25] N V Thakor H-C Shin S Tong and R G GeocadinldquoQuantitative EEG assessmentrdquo IEEE Engineering in Medicineand Biology Magazine vol 25 no 4 pp 20ndash25 2006

[26] H-C Shin S Tong S Yamashita X Jia R G Geocadin andN V Thakor ldquoQuantitative EEG and effect of hypothermiaon brain recovery after cardiac arrestrdquo IEEE Transactions onBiomedical Engineering vol 53 no 6 pp 1016ndash1023 2006

[27] H-C Shin X Jia R Nickl R G Geocadin and N V ThakorldquoA subband-based information measure of EEG during braininjury and recovery after cardiac arrestrdquo IEEE Transactions onBiomedical Engineering vol 55 no 8 pp 1985ndash1990 2008

[28] M C Tjepkema-Cloostermans F B van Meulen G Meinsmaand M J A M van Putten ldquoA Cerebral Recovery Index (CRI)for early prognosis in patients after cardiac arrestrdquoCritical Carevol 17 no 5 article R252 2013

[29] A O Rossetti T A Milligan S Vulliemoz C Michaelides MBertschi and J W Lee ldquoA randomized trial for the treatmentof refractory status epilepticusrdquo Neurocritical Care vol 14 no1 pp 4ndash10 2011

[30] J Bruhn TW Bouillon and S L Shafer ldquoBispectral index (BIS)and burst suppression revealing a part of the BIS algorithmrdquoJournal of Clinical Monitoring and Computing vol 16 no 8 pp593ndash596 2000

Submit your manuscripts athttpwwwhindawicom

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinsonrsquos Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom


Table 1 Clinical manifestations of the patients

Patientnumber Age Gender Etiology Diagnosis EEG findings Antiepileptic

medications

1 39 M Viralmeningoencephalitis Status epilepticus Burst suppression Midazolam IVlowast

2 72 M Cardiac arrest Postanoxicencephalopathy Burst suppression None

3 82 F Prolonged hypoxemiadue to pneumonia

Postanoxicencephalopathy Burst suppression Midazolam IVlowast

4 53 M Viral encephalitis Status epilepticus Burst suppression Midazolam IVlowast

M male F female IVlowast intravenous continuous infusion

evolving into the current methods in use The methods ofburst suppression segmentation developed thus far mainlyinvolve detecting burst events by using certain features suchas Shannon entropy [19 20] a nonlinear energy operator[21] line length [14] a voltage envelope [22] and varianceusing recursive-variance estimation [16] These all representfeatures employed in the time domain as well as occasionallythe basic features of the frequency domain like 3 or 10Hzpower or mean power spectral density (PSD) [19 20 23]Previous studies also compared the performances by the basicfeatures and segmentation methods [19 20 24]

In this study we applied the joint use of features extractedfrom the time and frequency domains to burst suppressionsegmentation to enhance the accuracy of segmentation con-ventionally conducted solely in the time domain We thus setthe EEG data and PSD thereof as the time-frequency vectorto extract features to be used in the time-frequency domainWe then applied the Shannon entropy the Tsallis entropy andthe regularity taken as features in broadly used conventionalquantitative EEG (qEEG) analysis [25ndash28] to the time-frequency vector After modeling of the distributions of burstand suppression of each feature into a Gaussian distributionwe employed the method of maximum likelihood estimation(MLE) to conduct the burst suppression segmentation Weassessed the accuracy of the proposed method by comparingthe accordance to visually scored burst suppressions Inaddition we employed the BSR as an indicator most widelyused for quantification of burst suppression and evaluatedthe closeness of estimatedBSR to the true BSRs obtained fromvisual scores In this study the method presented revealedimproved accuracy and precise BSR estimation Besides wesolved the problems that reside in the optimization processof conventionalmethods through the application of Gaussianmodeling and the MLE

2 Methods

21 Data Acquisition In this study we used the 11 multi-channel EEG data sets recorded from the Mokdong Hospitalof Ewha Womans University The patients were recruitedbased on the discharge database of the patients who hadbeen hospitalized as status epilepticus and EEG monitoringwas performed at the intensive care unit in Ewha WomansUniversity Mokdong Hospital from July 2012 to June 2015

Time (min)

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

100

minus100

EEG

(V

)

3252151050

Figure 1 The different burst suppression characteristics in eightEEGs from patient number 1 (Table 1)

A total of 122 patients who met the criteria were enrolledso far and their medical records for the age at onsetetiology comorbidity consciousness level before treatmentoverall duration of status epilepticus brain imaging andvisual interpretation of EEG were reviewed Among themfour patients whose EEG clearly showed a burst suppressionpattern were selected for the qEEG analysis in this studyusing artifact-free EEG segments for at least 20 minuteschosen by an expert neurologist (H W Lee) Among thosefour patients with a burst suppression EEG pattern threepatients were male while one was female (male female =75 25) and age ranged between 39 and 82 (mean 615 plusmn192) years (Table 1) Specifically eight EEGs were recordedfrom patient number 1 (Table 1) on different days whilewe obtained the rest of the three EEGs from three differentpatients Figure 1 shows the initial 3 minutes of the eightEEGs from patient number 1 We can observe the differentburst suppression characteristics in the EEG evolution The11 EEGs were recorded from 21 electrode locations basedon the international 10ndash20 system with 200Hz of samplingfrequency The 21 channels corresponding to each electrodewere Fp1 Fp2 Fz F3 F4 F7 F8 T3 T4 T5 T6 Cz C3 C4Pz P3 P4 O1 O2 A1 and A2


100500

minus50minus100

100806040200

EEG

(V

)

(a)

100806040200

5040302010

Freq

uenc

y (H

z)

Time (s)


wer

(dB)

(b)



x119898 = 119909 (119894) 119894 = 1 + 119898Δ 2 + 119898Δ 119873 + 119898Δ (1)


X119898 = STFT x119898 P119898 = 1003816100381610038161003816X11989810038161003816100381610038162 (2)



f119898 = [x119898P119898] (3)



119896=1119868119905(119896) and P119898 = ⋃119896=119872119891

119896=1119868119891(119896) Then




119878 (119898) = [[[[[[[

minus119872119905sum119897=1

119901119905 (119897) ln119901119905 (119897)minus119872119891sum119897=1

119901119891 (119897) ln119901119891 (119897)]]]]]]]

119879 (119898) = 1119902 minus 1[[[[[[[

1 minus 119872119905sum119897=1

119901119905 (119897)1199021 minus 119872119891sum119897=1

119901119891 (119897)119902]]]]]]]

(5)




3

25

2

15

1

05

Time domain

Freq

uenc

y do

mai

n3 5

3252151

BurstSuppression

(a)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

10

8

6

4

2

0765432

(b)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

035

03

025

02

015

01

0050605040302010

(c)



119877 (119898) =[[[[[[[[[


]]]]]]]]] (6)










= argmax120579

119901120579 (z) 120579 isin 119861 119878 (9)








Freq

uenc

y do

mai

n

Time domain3252151

Burst

Suppression

3

25

2

15

1

05

35

(a)

Freq

uenc

y do

mai

n

Time domain765432

9

8

7

6

5

4

3

2

1

Burst

Suppression

(b)

Freq

uenc

y do

mai

n

035

03

025

02

015

01

005

Time domain060504

04

0302010

Burst

Suppression

(c)



RMSE = radic 111987110158401198711015840sum119894=1





3 Results






Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201


9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712


















10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















100500

minus50minus100

100806040200

EEG

(V

)

(a)

100806040200

5040302010

Freq

uenc

y (H

z)

Time (s)


wer

(dB)

(b)



x119898 = 119909 (119894) 119894 = 1 + 119898Δ 2 + 119898Δ 119873 + 119898Δ (1)


X119898 = STFT x119898 P119898 = 1003816100381610038161003816X11989810038161003816100381610038162 (2)



f119898 = [x119898P119898] (3)



119896=1119868119905(119896) and P119898 = ⋃119896=119872119891

119896=1119868119891(119896) Then




119878 (119898) = [[[[[[[

minus119872119905sum119897=1

119901119905 (119897) ln119901119905 (119897)minus119872119891sum119897=1

119901119891 (119897) ln119901119891 (119897)]]]]]]]

119879 (119898) = 1119902 minus 1[[[[[[[

1 minus 119872119905sum119897=1

119901119905 (119897)1199021 minus 119872119891sum119897=1

119901119891 (119897)119902]]]]]]]

(5)




3

25

2

15

1

05

Time domain

Freq

uenc

y do

mai

n3 5

3252151

BurstSuppression

(a)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

10

8

6

4

2

0765432

(b)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

035

03

025

02

015

01

0050605040302010

(c)



119877 (119898) =[[[[[[[[[


]]]]]]]]] (6)










= argmax120579

119901120579 (z) 120579 isin 119861 119878 (9)








Freq

uenc

y do

mai

n

Time domain3252151

Burst

Suppression

3

25

2

15

1

05

35

(a)

Freq

uenc

y do

mai

n

Time domain765432

9

8

7

6

5

4

3

2

1

Burst

Suppression

(b)

Freq

uenc

y do

mai

n

035

03

025

02

015

01

005

Time domain060504

04

0302010

Burst

Suppression

(c)



RMSE = radic 111987110158401198711015840sum119894=1





3 Results






Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201


9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712


















10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















3

25

2

15

1

05

Time domain

Freq

uenc

y do

mai

n3 5

3252151

BurstSuppression

(a)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

10

8

6

4

2

0765432

(b)

Time domain

Freq

uenc

y do

mai

n

BurstSuppression

035

03

025

02

015

01

0050605040302010

(c)



119877 (119898) =[[[[[[[[[


]]]]]]]]] (6)










= argmax120579

119901120579 (z) 120579 isin 119861 119878 (9)








Freq

uenc

y do

mai

n

Time domain3252151

Burst

Suppression

3

25

2

15

1

05

35

(a)

Freq

uenc

y do

mai

n

Time domain765432

9

8

7

6

5

4

3

2

1

Burst

Suppression

(b)

Freq

uenc

y do

mai

n

035

03

025

02

015

01

005

Time domain060504

04

0302010

Burst

Suppression

(c)



RMSE = radic 111987110158401198711015840sum119894=1





3 Results






Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201


9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712


















10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of























= argmax120579

119901120579 (z) 120579 isin 119861 119878 (9)








Freq

uenc

y do

mai

n

Time domain3252151

Burst

Suppression

3

25

2

15

1

05

35

(a)

Freq

uenc

y do

mai

n

Time domain765432

9

8

7

6

5

4

3

2

1

Burst

Suppression

(b)

Freq

uenc

y do

mai

n

035

03

025

02

015

01

005

Time domain060504

04

0302010

Burst

Suppression

(c)



RMSE = radic 111987110158401198711015840sum119894=1





3 Results






Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201


9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712


















10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















Freq

uenc

y do

mai

n

Time domain3252151

Burst

Suppression

3

25

2

15

1

05

35

(a)

Freq

uenc

y do

mai

n

Time domain765432

9

8

7

6

5

4

3

2

1

Burst

Suppression

(b)

Freq

uenc

y do

mai

n

035

03

025

02

015

01

005

Time domain060504

04

0302010

Burst

Suppression

(c)



RMSE = radic 111987110158401198711015840sum119894=1





3 Results






Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201


9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712


















10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




















Shannonentropy

6855(1740)

7822(1193) 7339 8866

(2189)8278(1229) 8572 8451

(1777)8244(884) 8348

Tsallisentropy

6981(1630)

7403(1552) 7192 8831

(1791)8120(1330) 8476 8329

(1476)8073(964) 8201


9181(807) 8312 8860

(2304)8250(1410) 8555 8482

(1932)8520(1014) 8501

Time domain

Shannonentropy

5190(1510)

7743(1129) 6467 8791

(703)7407(1494) 8099 8372

(901)7697(914) 8035

Tsallisentropy

5099(1435)

7536(1305) 6318 8392

(602)6718(1328) 7555 8097

(831)7096(809) 7597


8702(742) 7489 8820

(797)7838(1671) 8329 7771

(808)8220(1041) 7996

Conventionalmethods

Line length[14]

6812(3130)

8603(566) 7708 9258

(1289)8096(1272) 8677 8467

(1342)8151(945) 8309

Envelope[22]

5713(2087)

7562(1428) 6638 8283

(2333)7722(2328) 8003 7732

(1916)7648(1838) 7690

NLEO [21] 5640(2190)

7289(1475) 6465 8360

(2217)7775(2470) 8068 7783

(1870)7641(1887) 7712


















10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

























10050

minus50minus100EE

G (

V)

0

302520151050

302520151050

302520151050

302520151050

T

R

S

Time (s)

(a)


wer

(dB)

302520151050

302520151050

Time (s)

0

5040302010

Freq

uenc

y (H

z)

T

R

S

302520151050

302520151050

(b)

Time (s)

T

R

S

302520151050

302520151050

302520151050

(c)





100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















100

minus100EEG

(V

)

302520151050

0

Time (s)

NLE

OEV

LL






4 Conclusion





BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















BBS ProposedR

TimeRLL

1451413513125121151110510

Rater 1Rater 2

ProposedRTimeR

LL

1000

minus100EE

G (

V)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

1451413513125121151110510

05

0

ΔBS

R(r

ater

1)

1451413513125121151110510

05

0

ΔBS

R(r

ater

2)

Time (min)

(a)

Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100

EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

04

02

01451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(b)

Figure 7 Continued


Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















Rater 1Rater 2

BBS ProposedR

TimeRLL

1451413513125121151110510

ProposedRTimeR

LL

1000

minus100EEG

(V

)

1451413513125121151110510

1451413513125121151110510

1

05

0True

BSR

04

02

01451413513125121151110510

ΔBS

R(r

ater

1)

0

04

02

1451413513125121151110510

ΔBS

R(r

ater

2)

Time (min)

(c)



Competing Interests


Acknowledgments


References





































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Research Article Novel Burst Suppression Segmentation in...

Documents

Transcript of Research Article Novel Burst Suppression Segmentation in...