Blind Visualization of Task-Related Networks From Visual Oddball Simultaneous EEG-fMRI Data: Spectral or Spatiospectral Model?

ORIGINAL RESEARCH published: 26 Ap il 2021 doi: 10.3389/ neu .2021.644874 F on ie s in Neu ology | www. on ie sin.o g 1Ap il 2021 | Volume 12 | A icle 644874 Edi ed by: B unno Machado De Campos, S a e Uni e si y o Campinas, B azil Re iewed by: Elias Eb ahimzadeh, Ins i u e o Resea ch in Fundamen al Sciences (IPM), I an Jonas Vibell, Uni e si y o Hawaii, Uni ed S a es *Co espondence: René Labounek [email p o ec ed] Igo Nes ašil [email p o ec ed] Special y sec ion: This a icle was submi ed o Applied Neu oimaging, a sec ion o he jou nal F on ie s in Neu ology Recei ed: 22 Decembe 2020 Accep ed: 22 Ma ch 2021 Published: 26 Ap il 2021 Ci a ion: Labounek R, Wu Z, B idwell DA, B ázdil M, Jan J and Nes ašil I (2021) Blind Visualiza ion o Task-Rela ed Ne wo ks F om Visual Oddball Simul aneous EEG- MRI Da a: Spec al o Spa iospec al Model? F on . Neu ol. 12:644874. doi: 10.3389/ neu .2021.644874 Blind Visualiza ion o Task-Rela ed Ne wo ks F om Visual Oddball Simul aneous EEG- MRI Da a: Spec al o Spa iospec al Model? René Labounek1*, Zhuolin Wu1,2, Da id A. B idwell3, Milan B ázdil4, Jiˇ í Jan5and Igo Nes ašil1,6* 1Di ision o Clinical Beha io al Neu oscience, Depa men o Pedia ics, Uni e si y o Minneso a, Minneapolis, MN, Uni ed S a es, 2Depa men o Biomedical Enginee ing, Uni e si y o Minneso a, Minneapolis, MN, Uni ed S a es, 3Mind Resea ch Ne wo k, Albuque que, NM, Uni ed S a es, 4Cen al Eu opean Ins i u e o Technology, Masa yk Uni e si y, B no, Czechia, 5Depa men o Biomedical Enginee ing, B no Uni e si y o Technology, B no, Czechia, 6Cen e o Magne ic Resonance Resea ch, Depa men o Radiology, Uni e si y o Minneso a, Minneapolis, MN, Uni ed S a es Va ious disease condi ions can al e EEG e en - ela ed esponses and MRI-BOLD signals. We hypo hesized ha e en - ela ed esponses and hei clinical al e a ions a e imp in ed in he EEG spec al domain as e en - ela ed (spa io)spec al pa e ns (ERSPa ). We es ed ou EEG- MRI usion models u ilizing EEG powe spec a luc ua ions (i.e., absolu e spec al model - ASM; ela i e spec al model - RSM; absolu e spa iospec al model - ASSM; and ela i e spa iospec al model - RSSM) o ully au oma ed and blind isualiza ion o ask- ela ed neu al ne wo ks. Two (spa io)spec al pa e ns (high δ4band and low β1band) demons a ed signi ican nega i e linea ela ionship (pFWE <0.05) o he equen s imulus and h ee pa e ns ( wo low δ2and δ3bands, and na ow θ1band) demons a ed signi ican posi i e ela ionship (p<0.05) o he a ge s imulus. These pa e ns we e iden i ied as ERSPa s. EEG- MRI F-map o each δ4model showed s ong engagemen o insula, cuneus, p ecuneus, basal ganglia, senso y-mo o , mo o and do sal pa o on o-pa ie al con ol (FPCN) ne wo ks wi h as HRF peak and no iceable ough. ASM and RSSM emphasized spa ial s a is ics, and he ela i e powe ampli ied he ela ionship o he equen s imulus. Fo he δ4model, we de ec ed a educed HRF peak ampli ude and a magni ied HRF ough ampli ude in he on al pa o he FPCN, de aul mode ne wo k (DMN) and in he on al whi e ma e . The equen - ela ed β1 pa e ns isualized less signi ican and dis inc sup a h eshold spa ial associa ions. Each θ1model showed s ong in ol emen o la e alized le -sided senso y-mo o and mo o ne wo ks wi h simul aneous basal ganglia co-ac i a ions and educed HRF peak and ampli ied HRF ough in he on al pa o he FPCN and DMN. The ASM θ1model p ese ed a ge - ela ed EEG- MRI associa ions in he do sal pa o he FPCN. Fo δ4, β1, and θ1bands, all models p o ided high local F-s a is ics in expec ed egions. The mos obus EEG- MRI associa ions we e obse ed o ASM and RSSM. Keywo ds: simul aneous EEG- MRI, ask- ela ed ne wo k isualiza ion, spec al and spa iospec al models, isual oddball pa adigm, gene al linea model, GLM, independen componen analysis Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks INTRODUCTION I es e al. and Huang-Hellinge e al. op imized ini ial simul aneous EEG- MRI da a acquisi ion (1,2) and Allen e al. and Goldman e al. implemen ed i s algo i hms supp essing g adien MR a i ac s induced in he simul aneous EEG eco dings (3,4). The de elopmen o a ious mul imodal da a usion s a egies has aken o d i en by he mo i a ion o gain he mos in o ma ion om EEG high empo al esolu ion and MRI high spa ial esolu ion. The i s published da a usion app oach c oss-co ela ed EEG αband powe luc ua ions wi h he MRI- BOLD signals o he es ing-s a e pa adigm (5,6), ollowed by he gene al linea model (GLM) implemen a ion (7,8). The GLM became a p ominen me hod equen ly applied in he ield and no only o he EEG spec a in eg a ion wi h es ing-s a e (9– 14) o ask induced (15–20) da ase s. GLMs inducing e en - ela ed po en ial (ERP) ampli udes o imings (21,22), and spike- in o med GLMs (23–25) ha e been p oposed and op imized. The oxelwise GLM esul s sel -o ganize in o la ge scale b ain ne wo k (LSBN) s uc u es (19). Concu en usion s a egy o en o a es MRI da a in o space o linea ly mix u ed spa ially independen componen s, i.e., he LSBNs, wi h hei ep esen a i e clus e wise induced BOLD luc ua ions, which a e compa ed o a ious EEG dynamics (26–31). In pa allel, eg ession o co ela ion app oaches in e ing EEG and MRI dynamics, join independen componen analysis (32), g aph build app oach (33), dynamic unc ional connec i i y (34), o mul imodal dynamic causal modeling (35) ha e been p oposed o use EEG- MRI da a wi h a ious esul in e p e a ions. Many eg ession o decon olu ion app oaches epo ed ha EEG- MRI hemodynamic esponse unc ion (HRF) demons a es a ying imings and shapes (10,19,20,36–42). In physiologic si ua ions, he BOLD signal is delayed o EEG e en s bu an ex eme example is he epilep ic spike EEG- MRI whe e BOLD signal peaks can p ecede he EEG spikes (41,42). Thus, i is a p e e able app oach o model a iable HRF han o use ixed canonical HRF, which has s ill been domina ing in he common p ac ice (5–9,11–18,21–26,28–33,43). O e a ious exis ing EEG- MRI da a usion echniques, he abili y o blindly and au oma ically isualize and quan i y obus ask- ela ed unc ional ne wo ks and hei EEG- MRI associa ions (e.g., ia a iable HRF) is lacking. We ha e ocused on he simple GLM usion app oach wi h a iable HRF agg ega ing au oma ically induced EEG spec a (19, 20) and es ed whe he we can iden i y usion se ings ha blindly isualizes ask- ela ed ne wo ks. This au oma ed me hod may o e high ep oducibili y wi h emendous po en ial in he clinical esea ch o e en clinical p ac ice applica ions o quan i a i ely measu e cogni i e dys unc ion. Cogni i e dys unc ion may occu in a ious neu ologic and psychia ic condi ions including epilepsy and can be es ima ed om EEG, e.g., by measu ing cogni i e e en - ela ed esponses such as P300 po en ial. The P300 esponse is ime-locked o an e en and is elici ed by a ask/e en when a es ed indi idual is eques ed o espond o a single s imulus o a se o s imuli as in he oddball pa adigm. The P300 has been inc easingly in es iga ed as a ma ke o cogni i e p ocessing. Speci ically, he P300 esponse ep esen s a neu al signa u e o he p ocessing o s imulus con ex depending on he a en ion and s a e o a ousal leading o an app op ia e esponse (44). Al hough he P300 has been almos exclusi ely assessed in he empo al domain ia ERPs (45,46), he cha ac e iza ion in he equency/spec al domain, since ime and equency a e ully complemen a y domains, may p o ide addi ional insigh in o he da a. Spec al (Equa ion 1) (16,17,43,47) and spa iospec al (Equa ion 2) (19,43) models ha e al eady been p oposed o he blind isualiza ion o ask- ela ed ne wo ks om simul aneous EEG- MRI da a. The 1s model (Equa ion 1) assumes ha local MRI BOLD signal luc ua ions (b) a e p opo ional o luc ua ions o he equency- speci ic (ω) weigh ed EEG absolu e/ ela i e powe spec a (p) wi h modeled be ween-signal delay ia HRF con olu ion ke nel (h). The weigh ing unc ion g(ω) can be conside ed a equency speci ic il e modula ing he powe spec a and inal powe luc ua ions a e o en es ima ed as an a e age o e channels (16, 17,43,47). The 2nd model (Equa ion 2) is simila bu conside s he il e ing p ope y o be channel (c) speci ic. The iden i ica ion o a obus ask- ela ed weigh ing unc ion g(ω)o g(c,ω), i.e., e en - ela ed (spa io)spec al pa e ns (ERSPa ) in EEG spec a, emains a no ully op imized challenge in he usion p ocess. b∝Zg(ω)p(ω)dω∗h(1) b∝Z Z g(c,ω)p(c,ω)dc dω∗h(2) The EEG absolu e/ ela i e powe spec a consis o a linea mix u e o s able spa iospec al pa e ns [i.e., di e en s able g(c,ω) unc ions in Equa ion (2)] wi h empo al luc ua ions ha we e mo e ask- ela ed o he ela i e powe a he han he absolu e powe (48,49). Absolu e EEG powe iden i ied 14 s able pa e ns wi h highly signi ican EEG- MRI associa ions a isual oddball da ase (19,48). Rela i e EEG powe iden i ied 10 s able pa e ns simila o he absolu e powe pa e ns and wo o he ela i e powe speci ic s able pa e ns, o which he EEG- MRI ela ionships ha e no been in es iga ed ye (49). Se e al s udies u ilized he spec al model o he isualiza ion o ask- ela ed neu onal ne wo ks om EEG- MRI da a (15– 18) o he spa iospec al model (19,43) wi h ew mu ual disc epancies: (i) he esponse unc ion was ixed o a iable; (ii) di e en asks we e used. The e o e, a di ec and ai compa ison be ween models s ill needs o be in es iga ed. Wi hin he cu en s udy, we a e p esen ing he ull compa ison be ween absolu e/ ela i e powe based spec al (Equa ion 1) and spa iospec al (Equa ion 2) models o ully au oma ic EEG- MRI usion. The goal is o op imize he au oma ic isualiza ion and quan i ica ion o ask- ela ed neu onal ne wo ks. The obus ness o e models was objec i ely assessed. F on ie s in Neu ology | www. on ie sin.o g 2Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks MATERIALS AND METHODS Expe imen al Design The iden ical simul aneous EEG- MRI da ase o isual oddball pa adigm was used as p e iously desc ibed (18,19,48,49). The e en - ela ed designed isual oddball ask was pe o med by 21 heal hy subjec s (13 igh -handed men, one le -handed man, se en igh -handed women; age 23 ±2 yea s). Th ee s imulus ypes we e p esen ed andomly o each subjec . Each s imulus consis ed o a single yellow uppe case le e shown o 500 ms on a black backg ound. In e -s imulus in e als we e ei he 4, 5, o 6 s (d awn uni o mly and andomly). A o al o 336 s imuli (di ided in o ou consequen ial sessions) we e p esen ed, consis ing o a ge s (le e X, 15%), equen s (le e O, 70%), and dis ac o s (le e s o he han X and O, 15%). Subjec s we e ins uc ed o p ess a bu on on he box held in hei igh hand whene e he a ge s imulus appea ed and no o espond o dis ac o o equen s imuli. This s udy was app o ed by and ca ied ou in acco dance wi h he ecommenda ions o he Masa yk Uni e si y E hics commi ee guidelines and all subjec s signed he app o ed w i en in o med consen in acco dance wi h he Decla a ion o Helsinki. Simul aneous EEG- MRI Da a Acquisi ion The scalp EEG da a, wi h e e ence be ween Cz and Fz elec odes, we e acqui ed wi h an MR compa ible 32-channel 10/20 EEG sys em (B ainP oduc s, Ge many) and a sampling equency o 5 kHz. Two channels we e used o ECG and EOG. Via he B ainVision Reco de sys em (B ainP oduc s, Ge many), he EEG da a we e synch onized and acqui ed simul aneously wi h MRI da a du ing g adien echo imaging sequences [1.5 T Siemens Symphony scanne equipped wi h Numa is 4 Sys em (M ease)]. G adien echo, echo-plana unc ional imaging sequence was acqui ed wi h ollowing pa ame e se ing: TR =1,660 ms; TE =45 ms; FOV =250 ×250 mm; FA =80◦; ma ix size =64 ×64 (3.9 ×3.9 mm); slice hickness =6 mm; 15 ans e se slices co e ing he whole b ain excep he in e io pa o he ce ebellum. The ask was di ided in o ou equal uns o 256 scans and 84 s imuli. Following simul aneous EEG- MRI acquisi ion, high- esolu ion ana omical T1-weigh ed images we e acqui ed using an MPRAGE sequence wi h 160 sagi al slices, ma ix size 256 × 256 esampled o 512 ×512; TR =1,700 ms; TE =3.96 ms; FOV =246 mm; FA =15◦; and slice hickness =1.17 mm. EEG Da a P ep ocessing The EEG da a we e p ep ocessed as desc ibed in (19,48) using B ainVision Analyze 2.02 (B ainP oduc s, Ge many) wi h he implemen ed manu ac u e ’s pipeline. G adien a i ac s we e emo ed using a e age a i ac sub ac ion (used sliding window wi h window leng h=21∗TR) a he acquisi ion sampling a e 5 kHz (3) and il e ed wi h a Bu e wo h ze o phase 1–40 Hz band-pass il e . Then, EEG signals we e esampled o 250 Hz (an ialiasing il e included). Ballis oca diog am (BCG) a i ac s we e emo ed by a e age a i ac sub ac ion (used sliding window wi h window leng h =21∗BCG epochs) wa e o m om each channel (50) and signals we e e- e e enced o he a e age. Eye-blinking a i ac s we e emo ed by conduc ing a empo al ICA decomposi ion and emo ing eye-blink a i ac s om he back- econs uc ed ime cou se. EEG Spa iospec al Decomposi ion The decomposi ion was he same as implemen ed and p e iously desc ibed (48,49). The p ep ocessed EEG signal om each lead and session was no malized o 0 mean and a iance 1, and di ided in o 1.66 s ( epe i ion ime o MRI scanning TR) epochs wi hou o e lap. Each epoch was ans o med o a spec al domain wi h he as Fou ie ans o m (FFT), gene a ing a ec o (leng h =67) o complex alued spec al coe icien s be ween 0 and 40 Hz. Complex alues we e con e ed o absolu e powe by aking he absolu e alue and squa ing, o con e ed o ela i e powe alue by di iding he squa ed alue by he powe o he whole epoch. The ou pu ec o o 67 eal absolu e/ ela i e powe alues comp ised a 3D ma ix Ewi h dimensions n ,nc, and nω. The dimension n ep esen s he numbe o spa iospec al epochs (n =256), he dimension ncis he o al numbe o leads (nc=30), he dimension nω is he o al numbe o spec al coe icien s (nω=67). The EEG spa iospec al decomposi ion (Equa ion 3) decomposes he ma ix o spa iospec al maps Ein o a sou ce ma ix S o dimensions S(m,n∗ cnω)con aining independen spa iospec al pa e ns and a mixing ma ix Wo dimensions W(n ,m) con aining he pa e ns’ dynamics. Dimension mis he numbe o decomposed independen spa iospec al componen s (m=20). E=WS (3) Using he GIFT oolbox (h p://mialab.m n.o g/so wa e/gi /) (51), he ma ix Ewas dimensionally educed using PCA (single- subjec educ ion o 50 p inciple componen s and g oup-based educ ion o 20 componen s), ollowed by INFOMAX g oup- ICA (52) wi h 10 ICASSO uns (53). Only spa iospec al pa e ns, which we e epo ed o be s able and obse ed in bo h absolu e/ ela i e powe spec a (48,49), i.e., 10 pa e ns, ha e been selec ed om ou pu sou ce ma ices So sepa a e g oup-ICA uns o absolu e/ ela i e powe s. Indi idual subjec ’s ime cou ses we e gene a ed by PCA based back- econs uc ion (i.e., he indi idual pa i ion o he PCA educing ma ix is ma ix mul iplied by he indi idual pa i ion o he agg ega e educing ma ix) (51) o he g oup spa iospec al pa e ns on he indi idual subjec spa iospec al maps and ime-cou ses. Selec ion o S able EEG Spa iospec al Pa e ns Wi h Rela ionship o he Task Fo each subjec and session, we ha e one ma ix Wwi h dimensions W(n , 20) con aining he back- econs uc ed ime cou se o each spa iospec al componen . Le a ma ix U o dimensions U(4∗n ,10) o be a single-subjec ma ix o luc ua ions o s able spa iospec al pa e ns o e all ou sessions. Rela ionships be ween hese dynamics and s imulus ec o imings (in ma ix X) we e assessed wi h a single-subjec gene al linea model (Equa ion 4, GLM) (54) sol ed wi h he leas mean F on ie s in Neu ology | www. on ie sin.o g 3Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks squa e algo i hm (Equa ion 5) and a con inuous g oup one- sample - es o he k- h s imulus ec o (Equa ion 6) (48). Va iable cis he ec o o bina y posi i e con as a he s imulus ec o o in e es , he b acke s <> cha ac e ize he expec a ion o e subjec s, a iable σis he s anda d de ia ion and a iable sis he o al numbe o subjec s. Model ma ix Xcon ained equen , a ge and dis ac o imings in 12 sepa a e bina y ec o s o each s imulus and session and ou ec o s o he DC componen in each session. U=Xβ+ǫ(4) β=(XTX)−1XTU(5) k=√shcT kβki σhcT kβki (6) These spa iospec al pa e ns, whe e any | |- alue was highe han 3.25 (c i ical alue a pFWE <0.05 o 10 mul iple compa isons, i.e., 10 selec ed s able pa e ns, and 16 deg ees o eedom, i.e., 16 a iables in model ma ix X) o any s imulus ype in absolu e o ela i e powe , we e conside ed as a pa e n wi h ask- ela ed EEG powe luc ua ions. Second selec ion c i e ia was o demons a e mean | |- alue a e aged o e all spa iospec al/spec al models wi h p<0.05 (∼| | >2.0) wi h ela i ely small s anda d de ia ion in he a e aged | |- alue o e models, i.e., STD| |<0.5. Task-Rela ed EEG Spec al Pa e ns All 10 s able spa iospec al pa e ns obse ed in bo h powe ypes we e a e aged o e leads and p o ided 10 spec al il e s g(ω) (Equa ion 1). The il e s we e used o il e ing o powe spec a o ma ix E eshaped a dimensions E(n ,nc,nω). Sepa a e o each il e , ime poin and channel, he absolu e/ ela i e powe alue p( ,c) was il e ed as Equa ion 7. Time-cou se p( ) o each channel cwas no malized o mean 0 and a iance 1, and inal absolu e/ ela i e powe luc ua ion ¯ p( ) was es ima ed as an a e age o e channels o each speci ic spec al pa e n. p( ,c)= nω X ω=1 g(ω)E( ,c,ω) (7) Task- ela ed spec al pa e ns we e e alua ed and iden i ied wi h he same me hodology as desc ibed in sub-chap e Selec ion o S able EEG Spa iospec al Pa e ns Wi h Rela ionship o he Task, bu ma ix U(Equa ions 4, 5) consis ed o 10 pa e n-speci ic a e aged ¯ p( ) luc ua ions. MRI Da a P ep ocessing The MRI da a we e p ep ocessed wi h SPM8 (Wellcome T us Cen e o Neu oimaging, London, UK) so wa e lib a y. Mo ion a i ac s we e minimized by alignmen o all unc ional scans, ollowed by co- egis a ion wi h he subjec ’s ana omical image and no maliza ion in o s anda dized MNI space (Mon eal Neu ological Ins i u e empla e) (55). Func ional scans we e spa ially smoo hed wi h an iso opic 3D Gaussian il e (FWHM =8 mm) o inc ease he signal o noise a io (SNR) and o make he andom e o s mo e no mally dis ibu ed. Pe iods longe han 128 s we e linea ly de ended o emo e slow d i s and physiological noise. EEG- MRI Gene al Linea Modeling Wi h Va iable HRFs Rela ionships be ween MRI oxel ime-cou ses (Y) and EEG ask- ela ed spec al/spa iospec al pa e n ime-cou ses we e examined using he indi idual GLMs (Equa ion 8) (54) wi h he EEG ime cou se con ol ed wi h he canonical HRF (x1), con ol ed wi h he 1s empo al de i a i e o he HRF (x2) o con ol ed wi h he 2nd empo al de i a i e o he HRF (x3) (19,20). In addi ion o he h ee EEG eg esso s x1-x3, he model ma ix Xcon ained a DC componen . Reg ession ma ices βwe e es ima ed o e all GLMs wi h he ReML algo i hm (Res ic ed Maximum Likelihood) implemen ed in SPM12 so wa e (Wellcome T us Cen e o Neu oimaging, London, UK) in he MATLAB p og aming en i onmen (Ma hWo ks, Na ick, USA). Y=Xβ+ǫ(8) G oup-a e aged EEG- MRI esul s we e es ima ed wi h a one- way ANOVA es (implemen ed in SPM12) o h ee EEG- de i ed single-subjec spa ial β-maps o each o h ee EEG eg esso s. The βweigh s se ed as dependen a iables in sepa a e ANOVA es s conduc ed o each spec al/spa iospec al pa e n, gene a ing g oup-a e aged spa ial EEG- MRI F-maps. The inal F-maps we e h esholded o objec i e e alua ions a p<0.001 unco ec ed o mul iple s a is ical es s (i.e., wi h a c i ical absolu e F- alue o 5.7), and he c i e ia ha clus e s con ain 100 oxels o mo e. Fo isualiza ions, he F-maps we e h esholded a p<0.0001 (i.e., F >8.1) due o high esul obus ness. G oup-a e aged EEG- MRI HRFs (hi) in each oxel i we e es ima ed om g oup-a e aged eg ession coe icien s β (es ima ed wi hin Equa ion 8) wi h Equa ion 9 whe e is he canonical HRF and numbe s 1–3 a e indexes o eg esso s x1-x3 (19,56). hi=βi,1 +βi,2 d d +βi,3 d2 d 2(9) Assessmen o EEG- MRI Resul s G oup-a e aged EEG- MRI F-maps and HRFs we e isually inspec ed o a subjec i e simila i y/dissimila i y e alua ion o e he simila spec al/spa iospec al pa e ns. Fo he objec i e assessmen , olume, mean F- alue, median F- alue, and maximal F- alue we e au oma ically ex ac ed om he sup a h eshold oxels (p<0.001) o he inal g oup-a e aged EEG- MRI F-maps o e e y s able ask- ela ed spec al/spa iospec al pa e n. The spec al/spa iospec al model wi h he highes objec i e me ics was conside ed as he mos success ul one in blind isualiza ion o a ask- ela ed neu al ne wo k. RESULTS Task-Rela ed EEG Spec al/Spa iospec al Pa e ns Signi ican nega i e linea ela ionship be ween EEG luc ua ions and equen s imulus was obse ed in wo F on ie s in Neu ology | www. on ie sin.o g 4Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks TABLE 1 | G oup - alues o linea ela ionship be ween s able EEG pa e n luc ua ions and s imuli ec o s. Spec al Spa iospec al Mean STD Absolu e Rela i e Absolu e Rela i e F equen δ1−0.09 0.46 −1.13 −0.58 −0.34 0.59 δ2−5.24 −2.42 −0.81 −0.92 −2.35 1.79 δ3−3.71 −1.71 −1.45 −1.45 −2.08 0.95 δ4−3.44 −4.70 −2.09 −4.32 −3.64 1.00 θ1−1.76 −0.87 −0.34 −2.03 −1.25 0.68 θ4−2.54 −1.95 0.39 −1.01 −1.28 1.10 α11.65 3.54 1.72 2.59 2.38 0.77 α30.62 2.03 2.48 0.12 1.31 0.97 β1−3.64 −3.25 −3.23 −3.28 −3.35 0.17 β2−1.89 −0.64 −0.17 −3.74 −1.61 1.38 Ta ge δ12.80 3.35 −0.40 −1.86 0.97 2.18 δ22.53 2.93 1.77 2.63 2.46 0.43 δ32.26 2.20 2.43 2.46 2.34 0.11 δ41.08 0.35 0.24 1.04 0.68 0.39 θ12.18 1.49 2.37 2.17 2.05 0.33 θ41.93 0.65 0.39 1.45 1.11 0.62 α12.05 0.51 2.62 1.73 1.73 0.77 α30.93 0.17 0.58 0.21 0.47 0.31 β1−0.11 −2.30 −0.52 −1.44 −1.09 0.85 β2−0.82 −2.06 −0.06 −1.40 −1.08 0.74 Dis ac o δ1−0.19 0.30 0.56 0.21 0.22 0.27 δ2−1.30 −0.40 0.31 1.30 −0.02 0.95 δ3−1.05 −0.56 0.21 −0.35 −0.44 0.45 δ4−3.11 −2.76 −2.24 −1.46 −2.39 0.62 θ1−0.58 1.45 −0.25 0.01 0.16 0.77 θ4−2.06 −2.52 −1.00 −0.43 −1.50 0.83 α1−0.23 1.17 0.56 1.65 0.79 0.70 α3−0.39 0.60 −0.34 0.40 0.07 0.44 β1−0.79 0.19 −0.61 −0.97 −0.55 0.44 β2−1.06 −1.00 0.12 −1.31 −0.81 0.55 Bold highligh ed - alues demons a ed signi ican ela ionship wi h pFWE <0.05. G een-colo highligh ed ows indica e he pa e ns wi h signi ican ask- ela ed ela ionships. These pa e ns demons a ed pFWE <0.05 o a mean o e models o p <0.05 (i.e., mean | | >2.0) wi h - alue s anda d de ia ion o STD| |<0.5. spec al/spa iospec al pa e ns (i.e., δ4wi h in e -model - alue −3.64 ±1.00, β1wi h - alue −3.35 ±0.17, Table 1, Figu e 1A,pFWE <0.05). The nega i e linea ela ionship can be in e p e ed as he EEG pa e n powe dec ease du ing he equen s imulus onse . O he h ee pa e ns (i.e., δ2 wi h - alue 2.46 ±0.43, δ3wi h - alue 2.34 ±0.11, θ1wi h - alue 2.05 ±0.33) demons a ed a signi ican posi i e linea ela ionship be ween he EEG powe luc ua ions and a ge s imulus (Table 1,Figu e 1B,p<0.05 and STD| |<0.5). The posi i e linea ela ionship can be in e p e ed as he EEG pa e n powe inc ease du ing he a ge s imulus onse . The δ4pa e n luc ua ions wi h - alue −2.39 ±0.62 may also be sensi i e o dis ac o s imulus wi h simila nega i e linea ela ionship (Table 1,Figu e 1A,p<0.05) as obse ed o he equen s imulus. Lowe obus ness o ela ionships be ween EEG pa e ns and a ge o dis ac o s imuli migh be caused by lowe s imulus amoun s. Fi e o 10 analyzed s able EEG pa e ns demons a ed po en ially signi ican ela ionship o s imuli ec o s o he isual oddball ask (Table 1,Figu e 1). F equen -Rela ed EEG- MRI Ne wo ks As he δ4pa e n demons a ed he highes le el o ela ionship be ween pa e n luc ua ions and equen s imulus, he EEG- MRI F-maps also isualized he highes F- alues and la ges amoun s o sup a- h eshold oxels o e all ou in es iga ed EEG pa e ns (Table 2,Figu e 2). The absolu e spec al model (ASM) p o ided he la ges and he mos signi ican EEG- MRI associa ions in compa ison o o he models (Table 2). S ill, s a is ical measu es a e e y high and obus in all in es iga ed models o he δ4pa e n (Table 2). Posi i e δ4pa e n demons a ed signi ican (pFWE <0.05) bila e al EEG- MRI associa ions in cuneus, p ecuneus, insula, basal ganglia, and in senso y-mo o ne wo k, soma osenso y ne wo k and do sal pa s o he on o-pa ie al con ol ne wo k F on ie s in Neu ology | www. on ie sin.o g 5Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks FIGURE 1 | Task- ela ed EEG spec al o spa iospec al pa e ns. (A) F equen ela ed; (B) a ge ela ed. Pa e n sho cu s and indexes (e.g., δ4) a e he same as used in (19) o consis ency pu poses. All spec al pa e ns a e a e ages o e elec odes o bo h absolu e o ela i e spa iospec al pa e ns. Pa e ns’ ampli udes (i.e., y-axis o spec al pa e ns and colo -coding o spa iospec al pa e ns) a e in a bi a y uni s. Fo spa iospec al pa e ns, he da k blue colo is an app oxima e minimum alue in he y-axis o he co esponding spec al pa e n. The da k ed colo is an app oxima e maximum alue in he y-axis o he co esponding spec al pa e n. (FPCN) (57–60) (Figu e 2). Pu amen, pallidum, halamus, and b ains em we e in ol ed wi hin subco ical g ay ma e s uc u es (Figu e 2). The do sal pa s o he FPCN o e lap wi h he do sal a en ion ne wo k (DAN) (57) and may be no disce nable one om he o he in a lowe spa ial esolu ion (Figu e 2). Spa ial dis ibu ion in Figu e 2B ep esen s he same esul , which was p esen ed in (19) wi h absolu e spa iospec al model (ASSM) and a iable HRF modeling. He e, we p oposed ha he ASM wi h a iable HRF modeling o ela i e spa iospec al model (RSSM) wi h a iable HRF modeling inc eased he s a is ical powe and he obus ness (Table 2,Figu e 2). All models demons a ed a posi i e HRF peak as e han a peak iming o widely used canonical HRF (Figu e 2) and no iceable HRF ough (Figu e 2) in he insula, senso y-mo o ne wo k, soma osenso y ne wo k, do sal pa o he FPCN and basal ganglia. Excep his expec ed esponse (i.e., ed HRFs in Figu e 2), e e y model de ec ed b ain a eas wi h educed HRF peak ollowed by la ge HRF ough ampli ude (i.e., he blue HRFs in Figu e 2). T ough peaks in bo h de ec ed HRFs we e again as e han an expec ed ough iming o he canonical HRF (Figu e 2). The HRF wi h a educed peak and an ampli ied ough was obse ed in a eas o supe io on al co ex and pa ie al co ex (Figu e 2A), which migh belong o he on al pa s o he FPCN (57–60) o de aul mode ne wo k (DMN) (57). In e io ly, we no iced signi ican (pFWE <0.05) bila e al clus e spo s in on al whi e ma e a eas (Figu e 2A) whe e o ceps mino , an e io halamic adia ion and in e io on o-occipi al asciculus migh pass [e alua ed by a isual inspec ion o EEG- MRI F-maps o e laid wi h he JHU whi e-ma e a las (61–63) in he MNI space]. As he β1pa e n demons a ed simila nega i e linea ela ionship be ween i s powe luc ua ion and he equen s imulus (Table 1), he EEG- MRI F-maps demons a ed simila loca ions o associa ion spo s (Figu e 3) whe e maximal |F| alues we e obse ed in he δ4EEG- MRI F-maps (Figu e 2), and simila HRF p ope ies (Figu e 3). The β1EEG- MRI F-maps demons a ed a lowe s a is ical obus ness o all models (Figu e 3,Table 2) when compa ed o he δ4F- map obus ness (Figu e 2,Table 2). Again, he ASM p o ided he mos obus s a is ics a he in e -model compa ison le el (Table 2). The ela i e spec al model (RSM) did no demons a e any signi ican EEG- MRI associa ions, which is analogical o he p e ious obse a ion o no ela i e βassocia ions wi h he ixed canonical HRF a he same da ase (18). Again, he RSSM was mo e obus han he ASSM (Table 2). O e all, he lowe obus ness, he β1EEG- MRI associa ions migh appea mo e spa ially speci ic. As in e p e ed om he ASM EEG- MRI β1F-map (Figu e 3). F om he basal ganglia, he bila e al pu amen demons a ed EEG- MRI associa ions wi h no mal HRF peak. F om he senso y-mo o ne wo k, bila e al EEG- MRI associa ions wi h no mal HRF peak we e obse ed in p ima y senso ic, p emo o , soma osenso y co ices, sup ama ginal gy us, le la e alized B odmann a ea 7 (BA7), and p emo o BA6. The educed HRF peak p ope ies we e obse ed o la e alized BA9, BA10, BA46, and igh hand p ima y mo o co ex. The RSSM emphasized soma osenso y BA6 associa ions compa ed o he ASM esul s (Figu e 3). In ega ds o he men ioned spa ial speci ici y, i is impo an o no e ha one can ge simila spa ial dis ibu ion o he δ4EEG- MRI F-map i s ic e h eshold han p<0.0001 is se in he Figu e 2 isualiza ion whe e local spa ial s a is ics exceeded he β1pa e n esul s (Table 2). F on ie s in Neu ology | www. on ie sin.o g 6Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks TABLE 2 | Spa ial s a is ics o ask- ela ed EEG- MRI F-maps. Spec al Spa iospec al Absolu e Rela i e Absolu e Rela i e F equen ela ed δ4Volume [mm3]937,845 431,865 571,266 656,910 Mean F- alue 14.08 9.96 10.65 13.05 Median F-Value 12.48 9.04 9.57 11.4 Max F- alue 53.83 27.35 32.19 45.27 β1Volume [mm3]282,555 - 40,797 106,704 Mean F- alue 8.37 - 7.4 7.61 Median F-Value 7.76 - 7.19 7.23 Max F- alue 22.39 - 12.17 15.31 Ta ge ela ed δ2Volume [mm3]524,205 26,271 58,185 26,865 Mean F- alue 10.14 7.34 7.68 7.35 Median F-Value 9.14 7.00 7.26 6.96 Max F- alue 28.86 13.76 16.12 14.97 δ3Volume [mm3]658,557 15,525 2,646 17,118 Mean F- alue 10.64 7.54 7.01 7.45 Median F-Value 9.45 7.05 6.81 6.98 Max F- alue 36.09 16.5 9.57 14.03 θ1Volume [mm3]852,930 140,994 236,844 334,746 Mean F- alue 12.44 7.94 8.92 9.02 Median F-Value 11.08 7.41 8.25 8.19 Max F- alue 47.32 17.1 21.24 29.78 The alues we e es ima ed om all sup a- h esholded oxels wi h p <0.001 (i.e., |F| >5.7). Bold highligh ed numbe s a e he highes ob ained alues o e in es iga ed models pe me ic. Values |F| >8.1 me he condi ion p <0.0001, alues |F| >12.1 me he condi ion pFWE <0.05. Ta ge -Rela ed EEG- MRI Newo ks Al hough δ2,δ3, and θ1pa e ns demons a ed all he posi i e ela ionship (p<0.05 and STD| |<0.5) be ween EEG pa e n luc ua ions and a ge s imulus, he EEG- MRI da a usion isualized he la ges and mos obus F-maps o he θ1pa e n o e all in es iga ed models (Table 2,Figu e 4). The mos obus s a is ics was yielded by ASM, ollowed by RSSM, ASSM, and RSM (Table 2). All models emphasized le la e alized EEG- MRI associa ions in he senso y-mo o ne wo k (co esponding o he push on he igh hand held bu on box and he a ge push bu on esponse) wi h smalle amoun s o he basal ganglia associa ions, which we e somewha p ese ed o he ASM and pa ly o he RSSM (Figu e 4). The ASM p ese ed signi ican EEG- MRI associa ions in he do sal pa s o he FPCN o e lapping wi h DAN (Figu e 4). These EEG- MRI associa ions p esen ed a non- educed HRF peak and a no iceable HRF ough again wi h iming as e han classic canonical HRF (Figu e 4). The ASM s ill isualized a signi ican (pFWE <0.05) educed HRF peak and an ampli ied HRF ough in a eas o he on al pa s o he FPCN, DMN and supe io on al whi e ma e bu in smalle amoun s han obse ed o he equen - ela ed δ4pa e n (Figu es 2,4). The RSM, ASSM, and RSSM e ealed a compa ably smalle amoun o signi ican clus e s in loca ions as ASM (Figu e 4). The RSM, ASSM and RSSM did no demons a e almos any signi ican EEG- MRI associa ions o δ2and δ3pa e ns (Figu es 5,6,Table 2). The ASM o δ2and δ3pa e ns showed simila spa ial and HRF obse a ion as o he θ1 ASM (Figu es 4A,5A,6A). The simila i y in EEG- MRI esul s be ween δ2and δ3ASMs was expec ed as bo h use almos he same spec al il e ing p ope ies o e all leads (Figu e 1). Dis ac o -Rela ed EEG- MRI Ne wo ks The e idence o a nega i e ela ionship be ween EEG powe luc ua ions and dis ac o s imulus was only no iced o he δ4 pa e n. The δ4EEG- MRI F-maps migh hen ep esen bo h equen - ela ed o dis ac o - ela ed associa ions (Figu e 2). DISCUSSION No el y and Neu oimaging Impac Ou esul s on isual oddball ask da a ep esen he sys ema ic objec i e compa ison o spec al (i.e., ASM, RSM) and spa iospec al (i.e., ASSM, RSSM) EEG- MRI da a usion me hods wi h he a iable HRF pe mi ing he a iable delay be ween he immedia e EEG ollowing he BOLD signal changes. The au oma ically quan i ied e ec o he a iable HRF in he EEG- MRI da a usion was ema kable. Bo h ASM and RSM esul s gained om he same da ase wi h ixed canonical HRF we e a om eaching pFWE <0.05 in EEG- MRI s a is ical pa ame ic maps (18,43). In con as , la ge numbe s o oxels in δ4and θ1EEG- MRI F-maps o all models me he s a is ical signi icance condi ion pFWE <0.05. While accoun ing o he ob ained EEG- MRI map obus ness and due o he in ol emen o insula, senso y-mo o co ex, soma osenso y F on ie s in Neu ology | www. on ie sin.o g 7Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks FIGURE 2 | EEG- MRI F-maps and es ima ed hemodynamic esponse unc ions o all in es iga ed models o he δ4pa e n. The EEG- MRI F-map colo ba is he same o e all models. HRF, hemodynamic esponse unc ion; IRF, impulse esponse unc ion. I HRF demons a ed a educed peak and an ampli ied ough (i.e., blue colo -coded HRFs) he F- alues we e assigned wi h a nega i e sign. The F-maps we e h eholded wi h p<0.0001, i.e., |F| >8.1. The oxels wi h |F| >12.1 me he condi ion pFWE <0.05. The ed colo coded HRFs we e de i ed om oxels wi h posi i e signed sup a h eshold F- alues. All F-maps a e shown ollowing he neu ological con en ion, i.e., le hemisphe e on he le side o he axial slice. (A) Absolu e spec al. (B) Rela i e spec al. (C) Absolu e spa iospec al. (D) Rela i e spa iospec al. co ex, cuneus, p ecuneus, basal ganglia, and FPCN (also known as cen al execu i e ne wo k) in he oddball/P300 asks (64– 73), we ecommend using a iable HRF and absolu e spec al (ASM) o ela i e spa iospec al model (RSSM) o blind and ully au oma ic isualiza ion o ask- ela ed ne wo ks om simul aneous EEG- MRI da a. The cu en app oach is ully au oma ed, blind, and obus . I o e comes EEG- MRI usion ia e en ela ed po en ial (ERP, F on ie s in Neu ology | www. on ie sin.o g 8Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks FIGURE 3 | EEG- MRI F-maps and es ima ed hemodynamic esponse unc ions o all in es iga ed models o he β1pa e n. The EEG- MRI F-map colo ba is he same o e all models. HRF, hemodynamic esponse unc ion; IRF, impulse esponse unc ion. I HRF demons a ed a educed peak and an ampli ied ough (i.e., blue colo -coded HRFs) he F- alues we e assigned wi h a nega i e sign. The F-maps we e h eholded wi h p<0.0001, i.e., |F| >8.1. The oxels wi h |F| >12.1 me he condi ion pFWE <0.05. The ed colo coded HRFs we e de i ed om oxels wi h posi i e signed sup a h eshold F- alues. All F-maps a e shown ollowing he neu ological con en ion, i.e., le hemisphe e on he le side o he axial slice. (A) Absolu e spec al. (B) Rela i e spec al. (C) Absolu e spa iospec al. (D) Rela i e spa iospec al. e.g., he P300) analysis o ampli udes o la encies (22,74–76) and is wi hou a need o supplying an inpu in o ma ion abou s imuli imings. Manpowe expe ise and e o s a e equi ed o he isual iden i ica ion in he ERP analysis. In he p oposed analysis app oach, quan i a i e EEG- MRI ask- ela ed ne wo ks a e gene a ed au oma ically. The po en ial neu oimaging impac o his me hodology is in a quan i a i e measu emen o local da a- d i en de e minan s o cogni i e de ici in pa ien s su e ing wi h epilepsy o o he condi ions wi h a cogni i e impai men . The speci ic ou come de e minan s may be subjec /g oup-speci ic F- alue magni udes o a iable HRF ampli udes and la encies. High a iance in local HRF la encies has been ecen ly epo ed in pa ien s wi h e ac o y ocal epilepsy (40). This obse a ion unde lines he impo ance o a iable HRF models o u u e clinical EEG- MRI applica ions. We demons a ed local EEG- MRI esponse unc ions wi h he educed HRF peak and he ampli ied HRF ough wi h he mos obus loca ion in he cen e o whi e ma e bundles. These bundles may con ey in o ma ion ela ed o execu ion and goal- di ec ed asks (77–79). Al hough he whi e ma e BOLD signal was conside ed as a nuisance signal ha was usually eg essed ou om he da ase du ing he p ep ocessing (80), se e al s udies ha e epo ed he whi e ma e MRI-BOLD signal and dis ega ded i s p e ious ca ego iza ion as a blind spo in he unc ional imaging (81–85). Ou obus whi e ma e EEG- MRI associa ions wi h he educed HRF peak and he ampli ied HRF ough co obo a e his ecen blind spo hypo hesis. Task- ela ed δ4and θ1EEG- MRI associa ions migh be conside ed con o e sial bu we a e con inced ha hey ep esen he bands o majo e en - ela ed inge p in s in he F on ie s in Neu ology | www. on ie sin.o g 9Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks 29. Hil unen T, Kan ola J, Abou Elseoud A, Lepola P, Suominen K, S a ck T, e al. In a-slow EEG luc ua ions a e co ela ed wi h es ing-s a e ne wo k dynamics in MRI. J Neu osci. (2014) 34:356–62. doi: 10.1523/JNEUROSCI.0276-13.2014 30. Ma eˇ cek R, Lamoš M, Labounek R, Ba on M, Sla íˇ cek T, Mikl M, e al. Mul iway a ay decomposi ion o EEG spec um: implica ions o i s s abili y o he explo a ion o la ge-scale b ain ne wo ks. Neu al Compu . (2017) 29:968–89. doi: 10.1162/NECO_a_00933 31. Keinänen T, Ry ky S, Ko honen V, Huo a i N, Nikkinen J, Te onen O, e al. Fluc ua ions o he EEG- MRI co ela ion e lec in insic s eng h o unc ional connec i i y in de aul mode ne wo k. J Neu osci Res. (2018) 96:1689–98. doi: 10.1002/jn .24257 32. Moosmann M, Eichele T, No dby H, Hugdahl K, Calhoun VD. Join independen componen analysis o simul aneous EEG– MRI: p inciple and simula ion. In J Psychophysiol. (2008) 67:212–21. doi: 10.1016/j.ijpsycho.2007.05.016 33. Yu Q, Wu L, B idwell DA, E ha d EB, Du Y, He H, e al. Building an EEG- MRI mul i-modal b ain g aph: a concu en EEG- MRI s udy. F on Hum Neu osci. (2016) 10:476. doi: 10.3389/ nhum.2016.00476 34. Lamoš M, Ma eˇ cek R, Sla íˇ cek T, Mikl M, Rek o I, Jan J. Spa ial- empo al- spec al EEG pa e ns o BOLD unc ional ne wo k connec i i y dynamics. J Neu al Eng. (2018) 15:036025. doi: 10.1088/1741-2552/aab66b 35. Wei H, Ja a ian A, Zeidman P, Li ak V, Razi A, Hu D, e al. Bayesian usion and mul imodal DCM o EEG and MRI. Neu oimage. (2020) 211:116595. doi: 10.1016/j.neu oimage.2020.116595 36. de Munck JC, Gonçal es SI, Huijboom L, Kuije JPA, Pouwels PJW, Hee haa RM, e al. The hemodynamic esponse o he alpha hy hm: an EEG/ MRI s udy. Neu oimage. (2007) 35:1142–51. doi: 10.1016/j.neu oimage.2007.01.022 37. Wu L, Eichele T, Calhoun VD. Reac i i y o hemodynamic esponses and unc ional connec i i y o di e en s a es o alpha synch ony: a concu en EEG- MRI s udy. Neu oimage. (2010) 52:1252–60. doi: 10.1016/j.neu oimage.2010.05.053 38. B idwell DA, Wu L, Eichele T, Calhoun VD. The spa iospec al cha ac e iza ion o b ain ne wo ks: using concu en EEG spec a and MRI maps. Neu oimage. (2013) 69:101– 11. doi: 10.1016/j.neu oimage.2012.12.024 39. P okopiou PC, Xi a-Po xas A, Kassinopoulos M. Modeling he hemodynamic esponse unc ion using mo o ask and eyes-open es ing-s a e EEG- MRI. bioRxi . (2020). doi: 10.1101/2020.06.29.178483 1 40. Van Eyndho en S, Dupon P, Tousseyn S, Ve lie N, Van Paesschen W, Van Hu el S, e al. Augmen ing in e ic al mapping wi h neu o ascula coupling bioma ke s by s uc u ed ac o iza ion o epilep ic EEG and MRI da a. Neu oImage. (2021) 228:117652. doi: 10.1016/j.neu oimage.2020.117652 41. Jacobs J, Le an P, Moelle F, Boo R, S ephani U, Go man J, e al. Hemodynamic changes p eceding he in e ic al EEG spike in pa ien s wi h ocal epilepsy in es iga ed using simul aneous EEG- MRI. Neu oimage. (2009) 45:1220–31. doi: 10.1016/j.neu oimage.2009. 01.014 42. LeVan P, Ty ae L, Moelle F, Go man J. Independen componen analysis e eals dynamic ic al BOLD esponses in EEG- MRI da a om ocal epilepsy pa ien s. Neu oimage. (2010) 49:366–78. doi: 10.1016/j.neu oimage.2009.07.064 43. Labounek R, Janeˇ cek D, Ma eˇ cek R, Lamoš M, Sla iˇ cek T, Mikl M, e al. Gene alized EEG-FMRI spec al and spa iospec al heu is ic models. In: 2016 IEEE 13 h In e na ional Symposium on Biomedical Imaging (ISBI). P ague (2016). p. 767–70. doi: 10.1109/ISBI.2016.74 93379 44. Polich J. Theo e ical o e iew o P3a and P3b. In: De ec ion o Change: E en - Rela ed Po en ial and MRI Findings, Polich J, edi o . Bos on, MA: Sp inge US (2003). p. 83–98. doi: 10.1007/978-1-4615-0294-4_5 45. Tue ing P, Su on S, Zubin J. Quan i a i e e oked po en ial co ela es o he p obabili y o e en s. Psychophysiology. (1970) 7:385–94. doi: 10.1111/j.1469-8986.1970. b01763.x 46. Kok A. E en - ela ed-po en ial (ERP) e lec ions o men al esou ces: a e iew and syn hesis. Biol Psychol. (1997) 45:19– 56. doi: 10.1016/S0301-0511(96)05221-0 47. Kilne JM, Ma ou J, Henson R, F is on KJ. Hemodynamic co ela es o EEG: a heu is ic. Neu oimage. (2005) 28:280– 6. doi: 10.1016/j.neu oimage.2005.06.008 48. Labounek R, B idwell DA, Ma eˇ cek R, Lamoš M, Mikl M, Sla íˇ cek T, e al. S able scalp EEG spa iospec al pa e ns ac oss pa adigms es ima ed by g oup ICA. B ain Topog . (2018) 31:76–89. doi: 10.1007/s10548-017-0585-8 49. Labounek R, B idwell DA, Ma eˇ cek R, Lamoš M, Mikl M, B ázdil M, e al. S able EEG spa iospec al sou ces using ela i e powe as g oup-ICA inpu . In: Wo ld Cong ess on Medical Physics and Biomedical Enginee ing 2018. Singapo e: Sp inge Singapo e (2019). p. 125– 8. doi: 10.1007/978-981-10-9038-7_22 50. Allen PJ, Polizzi G, K akow K, Fish DR, Lemieux L. Iden i ica ion o EEG e en s in he MR scanne : he p oblem o pulse a i ac and a me hod o i s sub ac ion. Neu oimage. (1998) 8:229–39. doi: 10.1006/nimg.1998.0361 51. Calhoun VD, Adali T, Pea lson GD, Peka JJ. A me hod o making g oup in e ences om unc ional MRI da a using independen componen analysis. Hum B ain Mapp. (2001) 14:140–51. doi: 10.1002/hbm.1048 52. Bell AJ, Sejnowski TJ. An in o ma ion-maximiza ion app oach o blind sepa a ion and blind decon olu ion. Neu al Compu . (1995) 7:1129– 59. doi: 10.1162/neco.1995.7.6.1129 53. Himbe g J, Hy ä inen A, Esposi o F. Valida ing he independen componen s o neu oimaging ime se ies ia clus e ing and isualiza ion. Neu oimage. (2004) 22:1214–22. doi: 10.1016/j.neu oimage.2004.03.027 54. F is on KJ, Holmes AP, Wo sley KJ, Poline J-P, F i h CD, F ackowiak RSJ. S a is ical pa ame ic maps in unc ional imaging: a gene al linea app oach. Hum B ain Mapp. (1994) 2:189–210. doi: 10.1002/hbm.460020402 55. F is on KJ, Ashbu ne J, F i h CD, Poline J-B, Hea he JD, F ackowiak RSJ. Spa ial egis a ion and no maliza ion o images. Hum B ain Mapp. (1995) 3:165–89. doi: 10.1002/hbm.460030303 56. F is on KJ, Fle che P, Josephs O, Holmes A, Rugg MD, Tu ne R. E en - ela ed MRI: cha ac e izing di e en ial esponses. Neu oimage. (1998) 7:30– 40. doi: 10.1006/nimg.1997.0306 57. Uddin LQ, Yeo BTT, Sp eng RN. Towa ds a uni e sal axonomy o mac o- scale unc ional human b ain ne wo ks. B ain Topog . (2019) 32:926– 42. doi: 10.1007/s10548-019-00744-6 58. Menon V, Uddin LQ. Saliency, swi ching, a en ion and con ol: a ne wo k model o insula unc ion. B ain S uc Func . (2010) 214:655– 67. doi: 10.1007/s00429-010-0262-0 59. Uddin LQ. Salience p ocessing and insula co ical unc ion and dys unc ion. Na Re Neu osci. (2015) 16:55–61. doi: 10.1038/n n3857 60. S idha an D, Le i in DJ, Menon V. A c i ical ole o he igh on o-insula co ex in swi ching be ween cen al-execu i e and de aul -mode ne wo ks. P oc Na l Acad Sci USA. (2008) 105:12569–74. doi: 10.1073/pnas.0800005105 61. Mo i S,Wakana S, an Zijl PCM, Nagae-Poe sche LM. MRI A las o Human Whi e Ma e . Ams e dam: Else ie (2005). A ailable online a : h ps://www. else ie .com/books/m i-a las-o -human-whi e-ma e /mo i/978-0-444- 51741-8 62. Wakana S, Cap ihan A, Panzenboeck MM, Fallon JH, Pe y M, Gollub RL, e al. Rep oducibili y o quan i a i e ac og aphy me hods applied o ce eb al whi e ma e . Neu oimage. (2007) 36:630–44. doi: 10.1016/j.neu oimage.2007.02.049 63. Hua K, Zhang J, Wakana S, Jiang H, Li X, Reich DS, e al. T ac p obabili y maps in s e eo axic spaces: analyses o whi e ma e ana omy and ac -speci ic quan i ica ion. Neu oimage. (2008) 39:336–47. doi: 10.1016/j.neu oimage.2007.07.053 64. Linden DE, P ulo ic D, Fo misano E, Völlinge M, Zanella FE, Goebel R, e al. The unc ional neu oana omy o a ge de ec ion: an MRI s udy o isual and audi o y oddball asks. Ce eb Co ex. (1999) 9:815– 23. doi: 10.1093/ce co /9.8.815 65. S e ens AA, Skudla ski P, Ga enby JC, Go e JC. E en - ela ed MRI o audi o y and isual oddball asks. Magn Reson Imaging. (2000) 18:495– 502. doi: 10.1016/S0730-725X(00)00128-4 66. Kiehl KA, Lau ens KR, Du y TL, Fo s e BB, Liddle PF. An e en - ela ed MRI s udy o isual and audi o y oddball asks. J. Psychophysiol. (2001) 15:221–40. doi: 10.1027//0269-8803.15.4.221 67. B ázdil M, Dobsík M, Mikl M, Hlus ík P, Daniel P, Pazou ko á M, e al. Combined e en - ela ed MRI and in ace eb al F on ie s in Neu ology | www. on ie sin.o g 16 Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks ERP s udy o an audi o y oddball ask. Neu oimage. (2005) 26:285–93. doi: 10.1016/j.neu oimage.2005.01.051 68. B ázdil M, Mikl M, Ma ecek R, K upa P, Rek o I. E ec i e connec i i y in a ge s imulus p ocessing: a dynamic causal modeling s udy o isual oddball ask. Neu oimage. (2007) 35:827–35. doi: 10.1016/j.neu oimage.2006.12.020 69. Chen C-C, Syue K-S, Li K-C, Yeh S-C. Neu onal co ela es o a i ual- eali y-based passi e senso y P300 ne wo k. PLoS ONE. (2014) 9:e112228. doi: 10.1371/jou nal.pone.0112228 70. Li F, Chen B, Li H, Zhang T, Wang F, Jiang Y, e al. The ime- a ying ne wo ks in P300: a ask-e oked EEG s udy. IEEE T ans Neu al Sys Rehabil Eng. (2016) 24:725–33. doi: 10.1109/TNSRE.2016.2523678 71. Li F, Yi C, Jiang Y, Liao Y, Si Y, Yao D, e al. The cons uc ion o la ge-scale co ical ne wo ks o P300 om scalp EEG. IEEE Access. (2018) 6:68498– 506. doi: 10.1109/ACCESS.2018.2879487 72. Zhang Y, Tang AC, Zhou X. Synch onized ne wo k ac i i y as he o igin o a P300 componen in a acial a ac i eness judgmen ask. Psychophysiology. (2014) 51:285–9. doi: 10.1111/psyp.12153 73. Li F, Yi C, Liao Y, Jiang Y, Si Y, Song L, e al. Recon igu a ion o b ain ne wo k be ween es ing-s a e and P300 ask. IEEE T ans Cogn De Sys . (2020). doi: 10.1109/TCDS.2020.2965135. [Epub ahead o p in ]. 74. Lieben hal E, Ellingson ML, Spanaki MV, P ie o TE, Ropella KM, Binde JR. Simul aneous ERP and MRI o he audi o y co ex in a passi e oddball pa adigm. Neu oimage. (2003) 19:1395–404. doi: 10.1016/S1053-8119(03)00228-3 75. O’Connell RG, Bals e s JH, Kilcullen SM, Campbell W, Bokde AW, Lai R, e al. A simul aneous ERP/ MRI in es iga ion o he P300 aging e ec . Neu obiol Aging. (2012) 33:2448– 61. doi: 10.1016/j.neu obiolaging.2011.12.021 76. Campanella S, Bou guignon M, Peigneux P, Me ens T, Nouali M, Goldman S, e al. BOLD esponse o de ian ace de ec ion in o med by P300 e en - ela ed po en ial pa ame e s: a simul aneous ERP– MRI s udy. Neu oimage. (2013) 71:92–103. doi: 10.1016/j.neu oimage.2012.12.077 77. Al ie i R, Melca ne A, Junemann C, Zeppa P, Zenga F, Ga bossa D, e al. In e io F on o-Occipi al ascicle ana omy in b ain umo su ge ies: om ana omy lab o su gical hea e . J Clin Neu osci. (2019) 68:290– 4. doi: 10.1016/j.jocn.2019.07.039 78. Dal ymple-Al o d JC, Ha land B, Louka enko EA, Pe y B, Me ce S, Collings DA, e al. An e io halamic nuclei lesions and eco e y o unc ion: ele ance o cogni i e halamus. Neu osci Biobeha Re . (2015) 54:145– 60. doi: 10.1016/j.neubio e .2014.12.007 79. Schepe s IM, Beck A-K, B äue S, Schwabe K, Abdalla M, Sandmann P, e al. Human cen omedian-pa a ascicula complex signals senso y cues o goal-o ien ed beha io selec ion. Neu oimage. (2017) 152:390– 9. doi: 10.1016/j.neu oimage.2017.03.019 80. Ba on M, Ma eˇ cek R, K ajˇ co iˇ co á L, Sla íˇ cek T, Kašpá ek T, Zemánko á P, e al. E alua ion o di e en ce eb ospinal luid and whi e ma e MRI il e ing s a egies—quan i ying noise emo al and neu al signal p ese a ion. Hum B ain Mapp. (2019) 40:1114–38. doi: 10.1002/hbm.24433 81. Maze olle EL, Beyea SD, Gaw yluk JR, B ewe KD, Bowen CV, D’A cy RCN. Con i ming whi e ma e MRI ac i a ion in he co pus callosum: co-localiza ion wi h DTI ac og aphy. Neu oimage. (2010) 50:616– 21. doi: 10.1016/j.neu oimage.2009.12.102 82. Gaw yluk JR, Maze olle EL, D’A cy RCN. Does unc ional MRI de ec ac i a ion in whi e ma e ? A e iew o eme ging e idence, issues, and u u e di ec ions. F on Neu osci. (2014) 8:239. doi: 10.3389/ nins.2014.00239 83. Ding Z, Huang Y, Bailey SK, Gao Y, Cu ing LE, Roge s BP, e al. De ec ion o synch onous b ain ac i i y in whi e ma e ac s a es and unde unc ional loading. P oc Na l Acad Sci USA. (2018) 115:595– 600. doi: 10.1073/pnas.1711567115 84. G ajauskas LA, F izzell T, Song X, D’A cy RCN. Whi e ma e MRI ac i a ion canno be ea ed as a nuisance eg esso : o e coming a his o ical blind spo . F on . Neu osci. (2019) 13:1024. doi: 10.3389/ nins.2019. 01024 85. Li M, New on AT, Ande son AW, Ding Z, Go e JC. Cha ac e iza ion o he hemodynamic esponse unc ion in whi e ma e ac s o e en - ela ed MRI. Na Commun. (2019) 10:1140. doi: 10.1038/s41467-019- 09076-2 86. Rek o I, B ázdil M, Nes asil I, Ba es M, Daniel P. Modi ica ions o cogni i e and mo o asks a ec he occu ence o e en - ela ed po en ials in he human co ex: cogni i e and mo o ask modi ica ions o e en - ela ed po en ials. Eu J Neu osci. (2007) 26:1371–80. doi: 10.1111/j.1460-9568.2007.05713.x 87. Ba¸sa E, Düzgün A. The CLAIR model: ex ension o B odmann a eas based on b ain oscilla ions and connec i i y. In J Psychophysiol. (2016) 103:185–98. doi: 10.1016/j.ijpsycho.2015.02.018 88. Ba¸sa -E oglu C, Ba¸sa E, Demi alp T, Schü mann M. P300- esponse: possible psychophysiological co ela es in del a and he a equency channels. A e iew. In J Psychophysiol. (1992) 13:161–79. doi: 10.1016/0167-8760(92)90055-G 89. Be na EM, Malone SM, Williams WJ, Pa ick CJ, Iacono WG. Decomposing del a, he a, and alpha ime- equency ERP ac i i y om a isual oddball ask using PCA. In J Psychophysiol. (2007) 64:62–74. doi: 10.1016/j.ijpsycho.2006.07.015 90. Ha pe J, Malone SM, Be na EM. The a and del a band ac i i y explain N2 and P3 ERP componen ac i i y in a go/no-go ask. Clin Neu ophysiol. (2014) 125:124–32. doi: 10.1016/j.clinph.2013.06.025 91. Peng W, Hu L, Zhang Z, Hu Y. Causali y in he associa ion be ween P300 and alpha e en - ela ed desynch oniza ion. PLoS ONE. (2012) 7:e34163. doi: 10.1371/jou nal.pone.0034163 92. Bende S, Banaschewski T, Roessne V, Klein C, Rie schel M, Feige B, e al. Va iabili y o single ial b ain ac i a ion p edic s luc ua ions in eac ion ime. Biol Psychol. (2015) 106:50–60. doi: 10.1016/j.biopsycho.2015.01.013 93. Ba¸sa -E oglu C, Demi alp T. E en - ela ed he a oscilla ions: an in eg a i e and compa a i e app oach in he human and animal b ain. In J Psychophysiol. (2001) 39:167–95. doi: 10.1016/S0167-8760(00)00140-9 94. Makeig S. Audi o y e en - ela ed dynamics o he EEG spec um and e ec s o exposu e o ones. Elec oencephalog Clin Neu ophysiol. (1993) 86:283– 93. doi: 10.1016/0013-4694(93)90110-H 95. Gün ekin B, Emek-Sa a¸s DD, Ku P, Yene GG, Ba¸sa E. Be a oscilla o y esponses in heal hy subjec s and subjec s wi h mild cogni i e impai men . Neu oimage Clin. (2013) 3:39–46. doi: 10.1016/j.nicl.2013.07.003 96. Lin Y-Q, Cui S-S, Du J-J, Li G, He Y-X, Zhang P-C, e al. N1 and P1 componen s associa e wi h isuospa ial-execu i e and language unc ions in no mosmic Pa kinson’s disease: an e en - ela ed po en ial s udy. F on Aging Neu osci. (2019) 11:18. doi: 10.3389/ nagi.2019.00018 97. Luck SJ, Heinze HJ, Mangun GR, Hillya d SA. Visual e en - ela ed po en ials index ocused a en ion wi hin bila e al s imulus a ays. II. Func ional dissocia ion o P1 and N1 componen s. Elec oencephalog Clin Neu ophysiol. (1990) 75:528–42. doi: 10.1016/0013-4694(90)90139-B 98. No i skiy N, Ramau a JR, Vande pe en K, De Vos M, Mennes M, Mijo ic B, e al. The BOLD co ela es o he isual P1 and N1 in single- ial analysis o simul aneous EEG- MRI eco dings du ing a spa ial de ec ion ask. Neu oimage. (2011) 54:824–35. doi: 10.1016/j.neu oimage.2010.09.041 99. Klimesch W, Schack B, Schabus M, Doppelmay M, G ube W, Sauseng P. Phase-locked alpha and he a oscilla ions gene a e he P1–N1 complex and a e ela ed o memo y pe o mance. Cogn B ain Res. (2004) 19:302– 16. doi: 10.1016/j.cogb ain es.2003.11.016 100. Bledowski C, P ulo ic D, Hoechs e e K, Sche g M, Wib al M, Goebel R, e al. Localizing P300 gene a o s in isual a ge and dis ac o p ocessing: a combined e en - ela ed po en ial and unc ional magne ic esonance imaging s udy. J Neu osci. (2004) 24:9353–60. doi: 10.1523/JNEUROSCI.1897-04.2004 101. An ice ic A, Cole MW, Mu ay JD, Co le PR, Wang X-J, K ys al JH. The ole o de aul ne wo k deac i a ion in cogni ion and disease. T ends Cogn Sci. (2012) 16:584–92. doi: 10.1016/j. ics.2012.10.008 102. Fox MD, Snyde AZ, Vincen JL, Co be a M, Van Essen DC, Raichle ME. The human b ain is in insically o ganized in o dynamic, an ico ela ed unc ional ne wo ks. P oc Na l Acad Sci USA. (2005) 102:9673–8. doi: 10.1073/pnas.0504136102 103. Li Y, Wang L-Q, Hu Y. Localizing P300 gene a o s in high-densi y e en - ela ed po en ial wi h MRI. Med Sci Moni . (2009) 15:MT47–53. A ailable online a : h ps://www.medscimoni .com/download/index/idA /869569 104. Wang RWY, Chang W-L, Chuang S-W, Liu I-N. Pos e io cingula e co ex can be a egula o y modula o o he de aul mode ne wo k in ask-nega i e s a e. Sci Rep. (2019) 9:7565. doi: 10.1038/s41598-019-43885-1 F on ie s in Neu ology | www. on ie sin.o g 17 Ap il 2021 | Volume 12 | A icle 644874 Labounek e al. EEG- MRI: Visualiza ion o Task-Rela ed Ne wo ks 105. Leech R, Sha p DJ. The ole o he pos e io cingula e co ex in cogni ion and disease. B ain. (2014) 137:12–32. doi: 10.1093/b ain/aw 162 106. Winkle I, Hau e S, Tange mann M. Au oma ic classi ica ion o a i ac ual ICA-componen s o a i ac emo al in EEG signals. Beha B ain Func . (2011) 7:30. doi: 10.1186/1744-9081-7-30 107. Eb ahimzadeh E, Sol anian-Zadeh H, A aabi BN, Fesha aki SSH, Habibabadi JM. Componen - ela ed BOLD esponse o localize epilep ic ocus using simul aneous EEG- MRI eco dings a 3T. J Neu osci Me hods. (2019) 322:34–49. doi: 10.1016/j.jneume h.2019.04.010 108. Eb ahimzadeh E, Shams M, Rahimpou Jounghani A, Fayaz F, Mi baghe i M, Hakimi N, e al. Localizing con ined epilep ic oci in pa ien s wi h an unclea ocus o p esumed mul i ocali y using a componen -based EEG- MRI me hod. Cogn Neu odyn. (2020) 15:207–22. doi: 10.1007/s11571-020-09614-5 109. Belouch ani A, Abed-Me aim K, Ca doso J, Moulines E. A blind sou ce sepa a ion echnique using second-o de s a is ics. IEEE T ans Signal P ocess. (1997) 45:434–44. doi: 10.1109/78.554307 110. Tang AC, Su he land MT, McKinney CJ. Valida ion o SOBI componen s om high-densi y EEG. Neu oimage. (2005) 25:539–53. doi: 10.1016/j.neu oimage.2004.11.027 111. Tang A. Applica ions o second o de blind iden i ica ion o high- densi y EEG-based b ain imaging: a e iew. In: Ad ances in Neu al Ne wo ks - ISNN 2010 (Heidelbe g: Sp inge Be lin) (2010). p. 368– 77. doi: 10.1007/978-3-642-13318-3_46 112. Congedo M, Gouy-Paille C, Ju en C. On he blind sou ce sepa a ion o human elec oencephalog am by app oxima e join diagonaliza ion o second o de s a is ics. Clin Neu ophysiol. (2008) 119:2677–86. doi: 10.1016/j.clinph.2008.09.007 113. Congedo M, John RE, De Ridde D, P ichep L. G oup independen componen analysis o es ing s a e EEG in la ge no ma i e samples. In J Psychophysiol. (2010) 78:89–99. doi: 10.1016/j.ijpsycho.2010.06.003 114. Hy ä inen A, Ramkuma P, Pa kkonen L, Ha i R. Independen componen analysis o sho - ime Fou ie ans o ms o spon aneous EEG/MEG analysis. Neu oimage. (2010) 49:257– 71. doi: 10.1016/j.neu oimage.2009.08.028 115. B idwell DA, Rachakonda S, Sil a RF, Pea lson GD, Calhoun VD. Spa iospec al decomposi ion o mul i-subjec EEG: e alua ing blind sou ce sepa a ion algo i hms on eal and ealis ic simula ed da a. B ain Topog . (2018) 31:47–61. doi: 10.1007/s10548-016-0479-1 116. Allen EA, E ha d EB, Dama aju E, G une W, Segall JM, Sil a RF, e al. A baseline o he mul i a ia e compa ison o es ing-s a e ne wo ks. F on Sys Neu osci. (2011) 5:2. doi: 10.3389/ nsys.2011.00002 117. Spadone S, Della Penna S, Ses ie i C, Be i V, Tosoni A, Pe ucci MG, e al. Dynamic eo ganiza ion o human es ing-s a e ne wo ks du ing isuospa ial a en ion. P oc Na l Acad Sci USA. (2015) 112:8112– 7. doi: 10.1073/pnas.1415439112 118. Calhoun VD, Liu J, Adali T. A e iew o g oup ICA o MRI da a and ICA o join in e ence o imaging, gene ic, and ERP da a. Neu oimage. (2009) 45:S163–72. doi: 10.1016/j.neu oimage.2008.10.057 119. Edwa ds BG, Calhoun VD, Kiehl KA. Join ICA o ERP and MRI du ing e o -moni o ing. Neu oimage. (2012) 59:1896– 903. doi: 10.1016/j.neu oimage.2011.08.088 120. Mangala hu-A umana J, Bea dsley SA, Lieben hal E. Wi hin- subjec join independen componen analysis o simul aneous MRI/ERP in an audi o y oddball pa adigm. Neu oimage. (2012) 60:2247–57. doi: 10.1016/j.neu oimage.2012.02.030 121. Ma ínez-Mon es E, Valdés-Sosa PA, Miwakeichi F, Goldman RI, Cohen MS. Concu en EEG/ MRI analysis by mul iway Pa ial Leas Squa es. Neu oimage. (2004) 22:1023–34. doi: 10.1016/j.neu oimage.2004.03.038 122. Van Eyndho en S, Ve lie N, De La hauwe L, Van Hu el S. Iden i ying s able componen s o ma ix/ enso ac o iza ions ia low ank app oxima ion o in e - ac o iza ion simila i y. In: 2019 27 h Eu opean Signal P ocessing Con e ence (EUSIPCO) (A Co uña: IEEE) (2019). p. 1–5. doi: 10.23919/EUSIPCO.2019.8902954 123. Jonmohamadi Y, Mu hukuma aswamy S, Chen J, Robe s J, C aw o d R, Pandey A. Ex ac ion o common ask ea u es in EEG- MRI da a using coupled enso - enso decomposi ion. B ain Topog . (2020) 33:636– 50. doi: 10.1007/s10548-020-00787-0 124. Jonmohamadi Y, Poudel G, Innes C, Jones R. Sou ce-space ICA o EEG sou ce sepa a ion, localiza ion, and ime-cou se econs uc ion. Neu oimage. (2014) 101:720–37. doi: 10.1016/j.neu oimage.2014.07.052 125. Jonmohamadi Y, Fo sy h A, McMillan R, Mu hukuma aswamy SD. Cons ained empo al pa allel decomposi ion o EEG- MRI usion. J Neu al Eng. (2019) 16:16017. doi: 10.1088/1741-2552/aae da 126. Wa b ick T, Reske M, Shah NJ. Do EEG pa adigms wo k in MRI? Va ying ask demands in he isual oddball pa adigm: implica ions o ask design and esul s in e p e a ion. Neu oimage. (2013) 77:177– 85. doi: 10.1016/j.neu oimage.2013.03.026 127. F is on KJ. Func ional and e ec i e connec i i y in neu oimaging: a syn hesis. Hum B ain Mapp. (1994) 2:56–78. doi: 10.1002/hbm.460020107 128. F is on KJ. Func ional and e ec i e connec i i y: a e iew. B ain Connec . (2011) 1:13–36. doi: 10.1089/b ain.2011.0008 129. F is on KJ, Ha ison L, Penny W. Dynamic causal modelling. Neu oimage. (2003) 19:1273–302. doi: 10.1016/S1053-8119(03) 00202-7 130. F is on KJ, P elle KH, Ma hys C, Cagnan H, Heinzle J, Razi A, e al. Dynamic causal modelling e isi ed. Neu oimage. (2019) 199:730– 44. doi: 10.1016/j.neu oimage.2017.02.045 131. Ha licek M, Roeb oeck A, F is on K, Ga dumi A, I ano D, Uludag K. Physiologically in o med dynamic causal modeling o MRI da a. Neu oimage. (2015) 122:355–72. doi: 10.1016/j.neu oimage.2015. 07.078 132. S ephan KE, Ti gemeye M, Knösche TR, Mo an RJ, F is on KJ. T ac og aphy-based p io s o dynamic causal models. Neu oImage. (2009) 47:1628–38. doi: 10.1016/j.neu oimage.2009.05.096 133. Anemülle J, Sejnowski TJ, Makeig S. Complex independen componen analysis o equency-domain elec oencephalog aphic da a. Neu al Ne w. (2003) 16:1311–23. doi: 10.1016/j.neune .2003.08.003 134. Be na EM, Williams WJ, Geh ing WJ. Decomposing ERP ime– equency ene gy using PCA. Clin Neu ophysiol. (2005) 116:1314– 34. doi: 10.1016/j.clinph.2005.01.019 135. Kauppi J-P, Pa kkonen L, Ha i R, Hy ä inen A. Decoding magne oencephalog aphic hy hmic ac i i y using spec ospa ial in o ma ion. Neu oimage. (2013) 83:921– 36. doi: 10.1016/j.neu oimage.2013.07.026 136. Shou G, Ding L, Dasa i D. P obing neu al ac i a ions om con inuous EEG in a eal-wo ld ask: ime- equency independen componen analysis. J Neu osci Me hods. (2012) 209:22–34. doi: 10.1016/j.jneume h.2012. 05.022 137. Ramkuma P, Pa kkonen L, Ha i R, Hy ä inen A. Cha ac e iza ion o neu omagne ic b ain hy hms o e ime scales o minu es using spa ial independen componen analysis. Hum B ain Mapp. (2012) 33:1648– 62. doi: 10.1002/hbm.21303 138. Zhong R, Li M, Chen Q, Li J, Li G, Lin W. The P300 e en - ela ed po en ial componen and cogni i e impai men in epilepsy: a sys ema ic e iew and me a-analysis. F on Neu ol. (2019) 10:943. doi: 10.3389/ neu .2019.00943 Con lic o In e es : The au ho s decla e ha he esea ch was conduc ed in he absence o any comme cial o inancial ela ionships ha could be cons ued as a po en ial con lic o in e es . Copy igh © 2021 Labounek, Wu, B idwell, B ázdil, Jan and Nes ašil. This is an open-access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License (CC BY). The use, dis ibu ion o ep oduc ion in o he o ums is pe mi ed, p o ided he o iginal au ho (s) and he copy igh owne (s) a e c edi ed and ha he o iginal publica ion in his jou nal is ci ed, in acco dance wi h accep ed academic p ac ice. No use, dis ibu ion o ep oduc ion is pe mi ed which does no comply wi h hese e ms. F on ie s in Neu ology | www. on ie sin.o g 18 Ap il 2021 | Volume 12 | A icle 644874