A Study on Analysis of Correlation Characteristics of EEG Channels by Five Senses stimulation

Transcription

1 A Study o Aalysis of Correlatio Characteristics of EEG Chaels by Five Seses Dog-Gyu Kim 1, Byug-Hu Oh 1, Hyeog-Joo Kwo 1, Kwag-Woo Chug 2, Kwag-Seok Hog 1 1 Huma Computer Iteractio Laboratory, School of Iformatio & Commuicatio Egieerig, Sugkyukwa Uiversity, , 300 CheoCheo-Dog, Jaga-Gu, Suwo, Gyeoggi-Do, Korea 2 Departmet of Railway Operatio System Egieerig, Korea atioal Uiversity of Trasportatio, Uiwag-S Korea Abstract - This paper is for the study of feature aalysis of the correlatio of brai waves betwee chaels i the of the 5 seses (sight, hearig, touch, smell, taste). I the existig studies, feature aalysis of brai waves i respose to of the huma seses has bee made by sigle sese or double sese, but, study with of all 5 seses is still ot yet satisfactory. I this paper, the brai waves i respose to of the 5 seses i the behavior of eatig food are collected ad aalyzed for the aalysis of the features of brai waves. From time-based data ad frequecy elemets, features of correlatio i brai waves betwee chaels are extracted. The overlappig rates betwee the sectios, betwee experimeters, ad betwee frames cosistig of chaels with high correlatio coefficiets are calculated. I the experimet, 65 sectios (32 sectios, 33 break sectios) of a brai wave database (DB) about the of the 5 seses were collected ad aalyzed for 9 me ad 2 wome. I the results of the experimet, the is distiguishable for each sese. Keywords: EEG, Five seses stimulus, Aalysis of correlatio characteristics 1 Itroductio Recetly, the study of Brai Computer Iterfaces (BCI) has bee active. Users itesios ad emotioal coditios ca be detected i brai waves measured through the scalp, ad the electrical activity of brai cells show differeces accordig to the metal coditios of a idividual, which ca be used for BCI. Huma Iterfaces based o brai waves have bee used for medical treatmet, games, traiig for metal cocetratio, robot cotrol, ad mid recogitio [1, 2]. Huma beigs recogize situatios through 5 seses (sight, hearig, touch, smell, taste). Humas become aware of their situatio through the seses. I additio, iteractio betwee humas ad obects as well as with other humas takes place by meas of the seses. Thus, aalyses of brai wave characteristics i respose to huma sesory of have bee coducted. I the existig studies, the feature aalysis of brai waves i respose to the of huma seses has bee doe with the of a sigle sese [3-9] or two seses [10], but study with the of 5 seses has bee dissatisfactory. Expadig o these studies, the features of brai waves i respose to the of all 5 seses are aalyzed by extractig correlatio features i the brai waves. A DB of brai waves measured i respose to the of all 5 seses durig the behavior of eatig food was collected ad aalyzed. I Chapter 2, we will examie existig research o sese. I Chapter 3, we explai our proposed correlatio feature aalysis method. Chapter 4 describes our experimets. Chapter 5 discusses the coclusio of this study. 2 Existig Studies The feature aalysis of brai waves for sigle sese [3-9] ad two seses at the same time [10] has bee coducted i previous studies. I additio, brai waves with visual usig correlatio coefficiets have bee aalyzed [4, 5]. The method for feature aalysis of brai waves acquires brai wave data whe seses are stimulated, ad extracts time-based data ad frequecy elemets i each chael (each area of the scalp) from the acquired brai wave data ad aalyzes them. Brai wave data is iduced from the scalp, passes through a amplified circuit ad filter. The the brai wave data is trasmitted to a aalog-to-digital (AD) coverter ad set to a smart phoe or PC for aalysis. I existig studies, frequecy elemet aalyses of brai waves are commoly classified as Delta (1~3Hz), Theta (4~7Hz), Alpha (8~13Hz), Beta (14~30Hz), ad Gamma (31~50Hz), ad are aalyzed for features. 3 Correlatio Feature Aalysis Method of Brai Waves betwee Chaels The feature aalysis of brai waves i respose to the of 5 seses was made by correlatio feature extractio from brai waves betwee chaels. A portable brai wave acquisitio device was developed i a previous study [11]. Aalog brai wave voltage sigals i the rage of 0.017uV~1.101uV are amplified by a factor of 2,225 to trasform them ito digital values ad are trasmitted to a smart phoe. This device acquires the brai wave data for the 14 chaels ad takes 128 samples per secod. ADtrasformed brai wave data is chaged to voltage values (0-

2 2.5V) by deductig a offset average value from each chael voltage value, ad removig the DC elemet. A correlatio coefficiet is used to calculate the time-based data ad 7 kids of frequecy elemets. The frequecy data are classified as Delta (1~3Hz), Theta (4~7Hz), Alpha (8~13Hz), Beta (14~30Hz), Gamma (31~50Hz), 51~59Hz, ad 1~59Hz. There are 91 x 8 kids of correlatio coefficiets betwee the 14 chaels for a total of 728. The correlatio coefficiet is treated as a frame uit (1 uit is 128 samples). If the quatity is less tha 128 i the last frame of each sectio, it is disregarded. Equatio (1) is used to calculate the correlatio coefficiet. r = M : The ( X umber i = 1,2,, M 1 = i + 1, i + 2,, M X i, X )( X i, = 1 2 ( X X i ) = 1 = 1 of ( X X ), X ) EEG 2 Chaels (1) is data at time sample of sectio i (chael), ad X i is average value over samples of a sectio (chael) of data. X ad i X are differet chaels. r i, is a ormalized covariace value divided ito the multiply stadard deviatio of the two variables X i, X by covariace. The rage has the value 1 r + 1. I case of r = + 1, there is a perfect positive correlatio betwee the two variables. I case of r = 1, there is a perfect egative correlatio betwee the two variables. I case of r = 0, the two variables are perfectly idepedet. I the calculatio of the overlappig rate of chaels with high correlatio i each sectio, the overlappig rate of all 65 sectios i the brai wave DB is calculated oce per test. The 240 chaels which have high correlatio i each sectio are udged i regard to whether they are the same or ot. A block diagram of correlatio feature aalysis of brai waves betwee chaels is show i figure 1. The method of correlatio aalysis is as follows. 1) Brai wave data is iput: 14 chael brai wave data. 2) Pre-processig: AD coversio data are chaged to voltage values. 3) Correlatio coefficiet calculatio: The correlatio coefficiets of time-based data ad 7 kids of frequecy elemets are calculated (91 correlatio coefficiet betwee 14 chaels, 91 x 8 kids of correlatio coefficiets for a total of 728 coefficiets.) 4) Statistical feature calculatio: 5 kids (mea, variatio, stadard deviatio, maximum value, miimum value) of feature values about 8 kids (time-based data ad 7 kids of frequecy elemets), (8 kids x 91 correlatio coefficiet betwee chaels x 5 kids feature value, calculate 3,640 whole feature value betwee chaels). 5) Extractig egative umber ad positive umber chaels which have high correlatio: Extract 3 positive umber ad egative umber chaels which have high correlatio from 5 kids of feature values about 8 kids of correlatio coefficiets betwee chaels. Extract 48 from a whole feature value betwee 3,640 chaels (8 x 91 x 5 matrix). 6) Calculatio of the overlappig of the high chaels: Judge whether the chaels with high correlatio are the same or ot (compare 48 extracted features betwee chaels). 4 Experimets ad Results 4.1 Usig the device We used a portable electroecephalography (EEG) acquisitio system developed i a previous study [11]. Figure 2 shows the portable EEG sigal acquisitio system iterfaced with a Adroid device. This EEG device amplifies brai wave sigals 2,225 times. It has a HPF cut-off frequecy of 0.58Hz, a LPF cut-off frequecy of about 58,94Hz, ad 16-bit/128 per secod samplig activities i AD coversio. Figure 1. Block diagram of correlatio feature aalysis of brai waves betwee chaels. The statistical feature calculatio calculates 5 kids of feature values (mea, variatio, stadard deviatio, maximum value, miimum value) of 8 kids of correlatio coefficiets. There are a total of 3,640 feature values betwee the chaels. (a) (b) Figure 2. The portable EEG sigal acquisitio system usig a Adroid device: (a) portable EEG device, (b) EEG demostratio (iterfacig betwee the portable EEG device ad smart phoes).

3 4.2 EEG Measuremet Poits The lower regios behid the ears are where the EEG measuremets are take. The 14 measuremet poits are deoted as F7, AF3, FC5, F3, T7, P7, O1, F8, AF4, FC6, F4, T8, P8 ad O2, ad they are placed accordig to the iteratioal electrode system. The gyro X-axis ad Y- axis are assiged to chaels 8 ad 16, respectively. Figure 3 shows the poits of EEG measuremet. was collected by dividig the total 65 sectios ito 4 sets. Each was about 10sec ~ 30sec, ad the breaks were about 30sec ~ 40sec. The breaks were doe with ope eyes ad restful sittig. The brai wave DB collectio sceario is show i Figure 4. Table 1. The type of 5 seses i the behavior of eatig food. o type of Sight Hearig Touch Smell Taste The umber of 1 x x x x x o 2 o x x x x Figure 3. The poits of EEG measuremet. 4.3 The Experimetal Coditios ad Results I this study, brai waves i respose to of the 5 seses durig the behavior of eatig food were collected ad aalyzed. Correlatio features betwee chaels regardig the time base eergy ad frequecy elemets were also extracted ad aalyzed. The subects of the experimet were 9 me ad 2 wome betwee the ages of 23~36, ad the experimet was executed i 65 sectios (32 s, 33 break ad rest periods). The experimetal coditios were as follows: 1) The experimetal place was a quiet laboratory, ad the experimet was doe with the subect sittig i a chair i frot of a table, ad the laboratory was well vetilated with a ope widow. 2) The food was a beef hamburger (Lot***** brad). The diameter was about 4 iches ad the materials icluded bus, beef patty, mayoaise, ad beef sauce, weighig 154g ad cotaiig 390 kcal. 3) The experimet was composed of oe subect, oe experimeter, ad aother experimeter to give verbal istructios (e.g. close your eyes ) to the subect, who acted accordig to the decided sceario. 4) Auditory stimuli were made by a hamburger advertisemet (the cotets of the advertisemet make people have the desire to eat a hamburger: beef ~ burger, burger, burger ~ Wow! Beef or shrimp which will igite flame from ma s heart) 5) The coditios of the sceario to collect the brai wave DB are to add of each sese, oe by oe, with miimized 5 sese, ad to make the applied distiguished. There were break ad rest sectios betwee the sectios to distiguish the sectios from each other. Table 1 shows the type of 5 seses i the behavior of eatig food. The brai wave DB 3 x o x x x 4 x x o x x 5 x x x o x 6 x x x x o 7 o o x x x 8 o x o x x 9 o x x o x 10 o x x x o 11 x o o x x 12 x o x o x 13 x o x x o 14 x x o o x 15 x x o x o 16 x x x o o 17 o o o x x 18 o o x o x 19 o o x x o 20 o x o o x 21 o x o x o 22 o x x o o 23 x o o o x 24 x o o x 0 25 x o x o o 26 x x o o o 27 o o o o x 28 o o o x o 29 o o x o o 30 o x o o o 31 x o o o o 32 o o o o o Sigle sese Double sese Triple sese Quarter sese Five seses Figure 4. The sceario of brai wave DB collectio (ex: o1 set) Experimet 1 This experimet was doe to aalyze the overlappig rate betwee sectios, which cosist of chaels with high correlatio coefficiets. For the correlatio feature values of brai waves betwee chaels of 65 sectios (32 sectios ad 33 break sectios) per oe subect, the overlappig rate of the chaels which had high

4 correlatio i each sectio were calculated. The average overlappig rates were obtaied for 32 sectios ad 33 break sectios each. From the results of the experimet, the average overlappig rate of 32 sectios for 11 people was 13.05%, ad the average overlappig rate of 33 beak sectios was 17.67%. The average overlappig rate of the total 65 sectios was recorded as Experimet 2 This experimet was doe to aalyze the overlappig rate betwee experimeters. The aalysis was doe o chaels with high correlatio coefficiets. From the results of the experimet, the average overlappig rate of al 65 sectios was recorded as 7.17%. Betwee sesory stimuli o the same cross-sectio overlappig of subects were lower tha the correlatio betwee EEG chaels Experimet 3 This experimet was doe to aalyze the overlappig rate betwee frames, ad was doe o chaels with high correlatio coefficiets. From the results of the experimet, the average overlappig rate of al 8 kids was recorded as 18.71% Experimet 4 We were recogitio experimets from 1 perso (subect 11). 65 kids of cadidate models of recogitio were used for the high-correlatio chaels from each sectio of the average correlatio coefficiets. This experimet chaged the extractig egative umber ad the positive umber chaels which have high correlatio. 3 to 30 of positive umber ad egative umber chaels which have high correlatio from 1 kid of feature values about 8 kids of correlatio coefficiets were extracted. 48 to 488 from a whole feature value betwee chaels 728 (8 x 91 x1 matrix) were extracted. I the results of the experimet, 26 extractig umbers showed the highest correlatio. The total average rate was recorded as 60.14% Experimet 5 We were recogitio experimets from 11 people. 26 positive umber ad egative umber chaels were extracted, each of which has high correlatio from 1 kid of feature value o 8 kids of correlatio coefficiets. From the results of the experimet, the average rate of recogitio was recorded as Coclusio A study o correlatio feature aalysis of brai waves betwee chaels related to the of five seses was performed i this paper. To aalyze correlatio features of brai waves related to the of five seses, brai waves were collected, ad a correlatio feature aalysis experimet was performed o the brai waves betwee chaels from the collected brai wave DB. I the results of the experimet, whe stimulatig the five seses, the overlappig rate of correlatio features of the brai waves betwee chaels was foud to be low, thus, it seems that the s could be distiguishable from each other. It seems that the method for extractig correlatio features of brai waves betwee chaels preseted i this study could be used for techology based o brai waves for recogitio of the five seses. I the future, the study of figure aalysis of brai waves related to the of the five seses durig movemet behaviors will be eeded. 6 Ackowledgmets This research was supported by the MKE, Korea, uder ITRC IPA-2012-(H ), by PRCP through the RF of Korea, fuded by MEST( ), ad by the Basic Sciece Research Program through the atioal Research Foudatio of Korea (RF), fuded by the Miistry of Educatio, Sciece ad Techology ( ). 7 Refereces [1] M. Tepla, Fudametals of EEG Measuremet", Measuremet Sciece Review, Vol 2, Sectio 2, (2002). [2] Syed M. Saddique ad Laraib Hassa Siddiqu "EEG Based Brai Computer Iterface", JOURAL OF SOFTWARE, VOL. 4, O. 6, pp , (2009). [3] Tadaori FUKAMI, et al., "The Relatioship betwee Agig ad Photic Drivig EEG Respose", IEICE TRASACTIOS o Iformatio ad Systems Vol. E94-D o. 9 pp , (2011). [4] Mabir Sigh, Sugheo Kim, ad Tae-Seog Kim, "Correlatio Betwee BOLD-fMRI ad EEG Sigal Chages i Respose to Visual Stimulus Frequecy i Humas", Magetic Resoace i Medicie, Vol. 49, pp , (2003) [5] YIJU WAG, XIAOROG GAO, BO HOG, CHUA JIA, ad SHAGKAI GAO, "Brai Computer Iterfaces Based o Visual Evoked Potetials", IEEE EGIEERIG I MEDICIE AD BIOLOGY MAGAZIE, pp , (2008) [6] Heidi Yppärilä, et al., "The effect of iterruptio to propofol sedatio o auditory evet-related potetials ad electroecephalogram i itesive care patiets" BioMed Cetral, Vol. 8, o. 6, (2004). [7] E. Maarrez, O. Diez-Martiez, I. Medeza, ad A. Flores, Stochastic resoace i huma electroecephalographic activity elicited by mechaical tactile stimuli, eurosciece Letters 324, (2002). [8] A. A. Cheriski et al., "Modificatios of EEG Related to Directed Perceptio ad Aalysis of Olfactory Iformatio i Humas", europhysiology, Vol. 41, o. 1, (2009). [9] Juliaa Cristia Hashida, et al., "EEG patter discrimiatio betwee salty ad sweet taste usig adaptive Gabor trasform", eurocomputig 68 (2005).

5 [10] M. Tepla, et al., "Phase Sychroizatio i Huma EEG Durig Audio-Visual Stimulatio", Electromagetic Biology ad Medicie, Vol. 28, o. 1, pp , (2009). [11] Dog-Gyu Kim, Yog-Wa Roh, ad Kwag-Seok Hog, "A Portable EEG Sigal Acquisitio System", The Secod Iteratioal Symposium o Mechaical Sciece ad Techology, December 20-21, (2011).