Old Dominion University Old Dominion University

ODU Digital Commons ODU Digital Commons

Computer Science Faculty Publications Computer Science

2019

Electroencephalogram (EEG) For Delineating Objective Measure Electroencephalogram (EEG) For Delineating Objective Measure

of Autism Spectrum Disorder of Autism Spectrum Disorder

Sampath Jayarathna Old Dominion University

Yasith Jayawardana Old Dominion University

Mark Jaime

Sashi Thapaliya

Follow this and additional works at: https://digitalcommons.odu.edu/computerscience_fac_pubs

Part of the Artificial Intelligence and Robotics Commons, and the Communication Sciences and

Disorders Commons

Original Publication Citation Original Publication Citation Jayarathna, S., Jayawardana, Y., Jaime, M., & Thapaliya, S. (2019). Electroencephalogram (EEG) for delineating objective measure of autism spectrum disorder. In C.-H. Chen & S.-C. S. Cheung (Eds.), Computational Models for Biomedical Reasoning and Problem Solving (pp. 34-65). Hershey, PA: IGI Global.

This Book Chapter is brought to you for free and open access by the Computer Science at ODU Digital Commons. It has been accepted for inclusion in Computer Science Faculty Publications by an authorized administrator of ODU Digital Commons. For more information, please contact digitalcommons@odu.edu.

Chapter 2

Electroencephalogram (EEG) for Delineating Objective Measure of

Autism Spectrum Disorder Sa m1>ath Jayanuhna

Old Dominion Univeniry. USA

Yasith J ayawardana Old Dmninion Universiry, USi1

ABSTRACT

Mark.Jaime Indiana Univenity-Purdue Vniveniry

Columbus, USA

Sashi Thapaliya Califvmia Swle Poly1ech11ic Univer.'iity

- Pomnua, USA

Au1ism ,pee/rum disonler (ASD) i.- a developme11tal disorder Ihm ofien impairs a child'.,· 11ormal del'e/oprne111 of the brain. According 10 CDC. it is estimmed 1h01 I in 6 children in the US suffer f rom developme/11 disoniers, and I in 68 chilclre11 in the I.JS .wjferfrom ASD. This condition lws a negative impacl on a person ',\ ability to hear, soci.alize, and communicate. Subjective met1.rnres oj!en f{l_ke more time, resources, and haw1JG/se positives or}O/se 11egati1•es. There is a need for efficienr objective meamres tha1 ca,i help in diagnosing this disease early as pos,·ible with less effort. EEG measures the elec1rh; .,·ignals of the brain via elec1rodes placed on variuus places on the scalp. These signals can be used /0 study comp/e:, neuropsychiatric iswe,. S111dies hm,e shown that EEG has 1he polenrial 10 be us~d as tt biomarker for various ne11rological conditions including /\SD. 11,is chap1er ,viii nu1/ine

!he usage uf EEG m easurement for /he clm·sif,rntion of /\SD usi,ig machine learning algurilhms.

DOI I0.401Sl978-1•5215•74()i-5.ch002

Electroencephalogram (EEG) fOr Delineating Objective Measure of Autism Spectrum Disorder

INTRODUCTION

Autism Spectrum Di.sorder(ASD) is characterized by significant impairmems in social imd commnnicative functioning as well as lhe presence ofrepetjtive behaviors and/or restricted interests. According io CDC estimaLes, the prevalence of ASD ( 14.6 per 1,000 children) has nearly doubled over the lase decade and has a costly impact on the lives of families affected by the disorder. It is estimated that I in 6children in the US suffer from developmental disorders. And I in 68 children fal l under Autism Specu·um Disorder. ASD is a neurological and developmental disorder thaL ha5' negative impact In a person's learning, social interaction and communication. h is a debilitating condition that affects brain developme nt from early childhood creating a lifelong chtillenge in normal funct.ioning. Autism is measured in spel'.rrum because of the wide range of symptoms and severity. The rota I lifetime cost of care for an individual with ASD can be as high as $2.4 millio11 (Buescher ct al. 2014). In the U.S., the long-term societal costs are projected to reach $461 billion by 2025 (Leigh and Du 2015).

One of the main contributing factors for ASD is kno\vn to be genetics. And so far, no suitable cure has been found. However, early intervention has been shown to reverse or correct most of its symptoms (Dawson 2008). And this can only be possible by early diagnosis. Therefore, early diagnosis is crucial for successful treatment of ASD. Although progress has been made to accurately diagnose ASD, it is far from ideal. Tl often requires various tests such as behavioral assessments. observations from care1akers over a period lo correctly determine the existence of Autism. Even with tl1is tedious testing often individuals are mis.diagnosed. However, there remains promise in the development of accurate detection using various modalities of Biomedical Images, EEG, and Eye movement..:;.

Efforts to identify feasible, low-cost, and etiologically meaningful biobehavioral markers of ASD are thus critical for mitigating these costs through improvement in the objective detection of ASD. However, the phenotypic and gcnotypic heterogeneity of ASD presents a unique challenge for identifyingprecursorsaligned with currently recognized social processing dimensions of ASD. One approach w unraveling the heterogeneity of ASD is to develop neurocognitive measures with shared coherence that map onto valid diagnostic tasks, like 1he Autism Diagnostic Observation Schedule Second Edition (ADOS-2) (Gotham et al. 2007), that are the gold standard

35

£1.clrocncephalogr.im (EEG) lor Delineating Objective Measu~ of Autism Spectrum Disorder

iE A.SD identificariun. These 111easure.s ca1~ lhcn be ustd h) stralify children into homogeneous subgroups 1 each repr~scntiug v.-trying degrees of impaired srk:i<.tl neurocognitive functioning. Dc:,:;pitt: the need forobjective, r,hy~iologic,ll measures o~· social functioning. :nachine k arning hJs nm yet t~n w idely appl.ied lo biohch,,vio:-al lllt Lrics fordiag11ostic purpose., in cl1i!<lre11 wilh ASO.

This chapter focuses 0n a :::.ocial ;woctssing dom,rin which, according, ro the Kl\1H lksearch Dom~in Criteria (RDoC), is a cemral defici! of AS D a11d lend::; ir~elf to qL12:1ti fiable ncurncognltive. paucrns: .sot:ial interaction:, choring ADOS 2. The ability to soc'ally t:oordinate viso.oai anection. share a point of v!ew witi1 another pcr~on. and process self- and oth~r-relatccl i:oformation (13:u·resi ancl Moore 1996; Buucrworth and.larn:11 1991; Mnndy cl al. 2009) is a fm,n<latinnal so<:ial cngnitive cupaci\y (Mllrody 2016). Ir, e.mt'r<Jencc in infancy p:-cd:~.ls ir:di vid11;1l differences in language deveJopmi;nl in both chil.dr~n with ASD ar~tl in typically developing children (Mundy ~l

ol. 1990; \111mly and Newcll 2007). Moreover, a1:.cnrion is r.,.;ognized in the cfo-'lgnostic cr!te ria of the DS·M .V as one uf ;he centl'al impairments of cnrly, nonverbal social commt?1·1icai:ion in ASD. While 1he empirical evidence on the physiological nature of auention deficiL~ in ASD i~ emi.:rging thal. can index a1Ler:1ion: -~0dal brain fu nctional co1mcctivity (FC:) {h!ring reaJ. Ji fo soc it'll ir~terac.:r io11.

At. tht": same t rme, i t is \\•·eH•cstablished iu the lircramre lhal rhe ne:ural syste.ms thctt subserve social '-=Ognil'ion ,,re f~mclionally c.ompn>1t •isec..l in childre,o wi1h ASD (Uaron-C:ol,~n et al. 1985; Lomi:>arco et al. 20 1 l; Hill and Frith 2003; Kuna er :,!. 2009; /Vfosi>r: el' al. 2008). 'Pie research ~cogge;ts there is a fum:tional (fro111al-1er.1poral-pmietal) ovcrl~p ill neural S)Slem ac1ivity dm·jng .A. DOS·2 and ~ucial cogn ir~ve proce:-;sing (Mundy 20 16; Kennedy a11d Adolpl•s 2012; Reckay et :ii. 2Ul2; Schurz et al. 2014; Lombardo el a l. 2010; Caruana et al. 2015). Taken TOg$th~r, l'here l~ ;i.rnple evid~nce 1·0 st1pport r.h~l

aberrant fnJ!1tai temJXlrai -parieral FC j ~ a potential nexus for lalent social cogni1ive disrurbance i11 ear!y ASD.

M any studies reveal eilher under- or over-co11ne,:ll:d <:l re::l./'i in 1he arni st.ic brain, depend ing on ' "·hefher 1hc suhject is at rest nr er.gag et.I in cogniti VI!

pruces,ing (Cobell et ,,; . 2008; Just el al. 2004; .l11st cl u l. 200(i; Kun,1 ~, ~I. 20 !4; Koshi no el al. 2005; Koshi no ct al. 2007; Laza:.:> el al. 20 IS; Lynch ct al. 2013; Uddin e t al. 20 :3; Shih e: al. 2010; Nooo,a" e l al. 20()'.); Jones el al. 20 JO; Damarlr: el al. 20!0: Muhum:nad-Re, ~zudeh c l al. 20! 6). Reduced FC wiLhin fron1.al1 superior teu:ro rai, and temporal- pariernl r;;!gion~-regions tha1 comprise 1be social brain system- liav~ been consi•aently reporleC i!l mosl fMRl sl1:dies l!Xami~ing f-C duri!lg ~ocial info:-r11alioc proces.,;ing

Efectrocricepha/ogram (EEGj for Delineating Obje_ctlve Mo.1surc of Au!/sm Spcclrum Disordar

(Ko,hi110 ct al. 2007; Castcili c, nl. 20(:2; Kleinham ct al. 2008; Rudie c1 al. 2011; Wekhe-. et al. 2005). The presence of.1l rered social brain system FC in .;arly neur0dcvclnpm~nf. c,1:1 potentially rev~a.l the or:set of ::mcia1 djffjcolties

(Keehn ct al. 2013). as ~Ile.red FC dlsruprs efficien! information Ho-.v between para] lei i~nd distri~uted neur:-:tl systems i:1volved in tht! proc~ssi1ig ol".';.;ocial ~rnd cormm:nicative infornrnt.ion (rVlundy et al. 2009). Thus, children wilh ASD may develop with fmited nc11nx:ognilive rrsnurces to efficieraly deal ,vilb :.he processing dtm:l.misof Jy:1amic social exchange~. This s.oci.a.l deficjt may emerge as irlio~yncr~tic p;itlerus ufEEG during boots of joint sodal mtention

LITERATURE SURVEY

Social Interaction Tasks

ro date, [he few ~tudie~ that bave examinc.d FC du,·ing art;:.nrion h,n,e dor.c so using non•clinica] paradigms th,1.r involve the observa~ion olt:ttemioE-eliCJti 11g videos~ however. data from such paradigms mc1y Pot. reflec t. I he rrtH"~ person· to-per.;;.on internc~ivf! nature. lvlore. :wportanLly, video paradigms m~1y 1mly tap ir~ro one of two fac~l~ of attcmi.;rn: respo:Hling to jol1H. a11.entio'!1 (R.lA),

which serves 1-;11 imperative. function. \Vhat is not rcpn:scntcd :1~ .IA-clicif.in~ \'i<lco para<.bgms '.s initiating joint attention (It-\), ,vhlch serve:; a dedarnli ve function and taps into social rcwat(~ ~y:-:.tcms that ,1rc integral to the social sharing o' experiences iCaruana et al. 20 l 5: Schilbac-h et ai.2010; Gordo,; c1 al.2013). '.\1oreover, RJA and LIA show a developmental dissociation during the first and second years of life (Yoder el al. 2009; lbaiiez et al.2013: !v!undy et al. 2007). Although RJA and l.JA 'iotl1 h~ve predictive value ic infancy, !JA is a more stable marker of ASD thai, RJA in iarer childhood (Mundy e, al. 1986). Some:· ncurofrnagi.ng rese.archers have de~lt \.Vith the above issues by usmg a live face ro-avatar paradigm '.O simu,atc IJ/\ bids (Rcdc~y ct al. 2.012; Gordon cl al. 2013). However, the movernentconslraints inside the MRI ::.ca.·1ner create :.esting conditions th.:.il can b~ difficuh for younger children, wi1h and wltf10ut ASD.

Eye movemc:nl behavior is a n:~sull or complex ncurolugicaJ processes: thcref~ve, eye gaze met.ri.:.·s c.?Jn revei1I objecti·,,e <.rnd q~1a11tifiabie J11 t·uru1atiun at.>oul· lhl~ predictability aml cu,1sistency uf cuvcrl :mciai cogniLive processes. incli.l<ltngsociril ;:Ueation (Cl1itn·Tegrna1·k 2016; GuiHon etal. 20.14),emOliDli rcrngr.iti1)n (Bal Cl ,II. 2010; I.Jlack ct al. 2017; Sawyer ct al. 2012; Sassnn

37

EJectroencephitlot;r! tr. (EEG) for De-lit,eatlng Objective Meas 1Jre of Autism Sper:tr11m Dfscm1er

et al. 201 6; Tsang 2!) ; 8; W;igncr et al. 2016; Wieckow,ki an(I While 20 17), perspe~li\'e taking, (Symeonidou et al. 2016) an<I joint aHcntion (Ilcdford c, al.201 2; 13illeci Cl al. 2016; Falck-Y11er cl al. 2012; Falck-Yncr el al. 2015; Swansou el al. 20 l3; Thorup e, al. 20 i6; Thorup CL al. 20 I 8; Vivan ti et aL 20 l 7) for children will; and wi thout ASD. Eye gaze m~asureme.nt includes scvl!.raI metrics rdevim tn ocuJommor control (Komogorrsev el al 2013) such as sac.cadic lraje.ctoric~. fixations. :rnd olher relevant measures SllCh a, vclc~ily, dura1.ion, ampli1.11de, and pupil dila1iun (Krejtz ct al. 201 Hli). We believe th>it combined analysi, of t:xat.ions and ,:1cc2des during nau1ral and dyna1~1ic joint aue11tio!"I tasks, 1..:11rrei:rly used as 2 r~!iablc m.easure or ASI) dicgnosiic criteria, will represent valid !">iomarkcrs for objectifying ~nd delineating thl~ clime-nsion:tlity of ASD di~1gnrn;;s in the fot11,.-e . Pre-vi~)1: . ..; work i11 thi!i area have successfully de.monstrated deveJo?mem of l<.. the­coefficient of .c111bicntf foc"I aetcm:011 (Krejtz ct al. 2016) and prcvim:s work h ..:.s. ~upported the re latio11s!1ip between eye trackir,g mc:1rics an c.} sr:vcrity

of ASD diagnosi.1 (Frazier et al. 20 18; De! Valle Rubido cl. al. 2() 18; and communicative co111m:.1ence (Korbury et al. 2009). Jr v:sual attention in fluences

stabiiity of fixations d~peuclent u;,on 1he demands of <lynamic joint alien ti on las k,. a natural :it.xi ,tep is w look into how relevance may be rct1ec1ed in similal" m:urophy~ioiogic features !or atypical social br;:i i11 sy~tems, such as 111 the co111ex: of ASD (Hoticr er al. 20 l 7),

EEG Based Machine Learning for ASD

Stm.bes have shown that EEG has the potential to be u,;,;:d a, biom:u·ker tor n rious ne,1rological conditions ind udi11g ASD (Wang e1 al. 20 13;, Ecn measures the clecrrica'. .signals of th~ brain via t:lt:~\Jodes t !1at Ere placed 0 11

various places on the ,calp. The.se ~kctrical sig11als are postsynapt.ic activity i11 the neoconex and can be used ro smdy complex ncum psych,atric ,ssurs. EEG has various frequency bands and its a11c,lysis are performed on these varying bandwidths. Waves bet\vc~n 0.5 u:1d 4 HZ a:·e deltat beL,1.:een 4 and 8 HZ are theta. between 8 and 13 HZ an, alpha, 13 to 35 HZ are beta and over 35 are gamma. Saccadic eye movcme11l plays c. big role in Lhe attcnt ion

anci he.havior of an i:1dividual which dirt!ctly affects both ~ang11~ge and ~oci;-il skills (Flerc!1er-Wa1son et ill. 2009). Autistic children seen, to have c' ifforcnt eye movement hehaviors than non-aub ltc children. They tend IO avoid ~ye contact and lookin~ al human foce while focusing more n11 gco111etric ~h,tpes

38

Eiec.troenccph.:i(ogr.:tm (EEG) for D~fineatiug Objective Measure of Autism Spectrum Disorder

(Kli" ct al. 2009). While a typical chilt'. doesJ1't find any imerest in geome,ric sJ1apc-.: and rer:d ta make more eye co:1tdct, .and hm~rnn face pcrceplion.

In Grossi et cl. (2017), authors u~e a complex Hl·'.G proce~.,ing aigoriihm c.:1llecl MSR0~·1/I -FA.STaJong v.;ith multiple machine l~arn.i:ig algorithms t;) cla,siiy Auti,1ic pstie,Hs In t;1i, s:udy 15 ASD i~divirl11a:., and lfl non ASD \Vee .~elected. ASD grnup comprised of 13 males and 2 females be~ween 7 and 14 years of age. Coi~~rol gn.)up comprised of .~1 males and 6 female-~ betv.,ecn 7 ar.(: 12 years of age. Resting State. EEG of both clo~ed and open eyes \Vere recorded using 19 electrodes. ?alien!.~ sat i1~ a quiet room wtt.hout speiking or performing any mentally demamling ac1·ivi1y \vhile ihe ELG 'NHS hclng rccnnklL The proposed fFAST algorith1n t..:onsisr...; nf exactly tlHcc different 1>hases or par!>. In the Ctrst srage also called Squashing phase, rl1c raw EEG signals 1:1rc converted into fci.itu:·e vectors. Anthors presem a workflow of the system from raw rl:c:rn to dassificativn to make comparis.on between <l,fferent algorithms such a, Muhi Scale Entropy ('v1S Ej a 11'1 the. M11lti Scale Ranked 0;-ganizing /l·laps (\IS-ROM). MSROM is a novel algorithm based on Singk Org1.~nizing J\.1ap Neural Netv,·ork. In lh!S study, ttc dataset is rn<ldomly divided inlu 17 lrainiag c<msisting of 11 ASD, 6 co:;trols aHi eight test records conslsting cf 4 ASD, 4 cont;ul. The nois~ elimination is perl'Onned only on the training set:. Abo, it (.:ornp'.etdy depends on 1Le algorithm ,deeted tor extra~tion of fcalu:·c vectors, For MS-RO~'v! fe~rnres the-, utilize a,, algDrithm calkd TWIST. In the final classification st.::.ge, :.hey us~ mulLiple mal'.t1iric learn~!lg algorith!ns a!ong wit.h mulliple validaticm pi::Jtoco1s. The. validmion protncol'-i are tn=iining-rest.ing a11d leave rn1e out cross v.alidm;or;, For classification purpo3;c~ they make use of .Sine Net Ne11ral N~hvork, Logi1:lic Regre~sion, Sequential l\1linimal Optimization, kN>i1 K-Cnmrncr:ve i\·1ap, Nah,c llaye.s, anc R~tndom forest. \.Virh ~1SE feature exLraction the best results were given l;y Logistic and t\nivc Bayc.s with exactly 2 errors. Whereas, MS-ROM with training test protocol hatl 0 erto:·.'\ ( l 00% ac(:uracy) with 1111 the c.la~sificatirn1 models.

Bosl ct aL (201 f). crin(llld a s111dy using mlvISE as feature vectors along wich multiclass Support Vcct·nr \1r1chinc. t.o difl'~r~ntiat.e devclopi1tg and high-risk infant groups. In this ~tudy they u~c 79 &ffercnt infanrs of whkh 49 were considered high risk a:id 33 typica]y developing infants. The 49 infants \.vere high risk based on one oft.heir older sibJtngs having a confirmed ASD diagnosis. The ot.het 39 infants wc[c nm. high fi-:.k .c:ince no orie iii Lheir ramily ~ver was diagnosed ,i..·hh ASD. Data Vl'as collected from each i11frJr~t. during multiple s:e::t'>iori,; with :-..ome intervaJ. Duta excracted from an info.r.t in five different .,~:ssions in various mmJths bdw~cn 6 to 24-month period

39

EIKUooncepha/ogram (EEG) for Deli11~9ting Objecli11t1 Measure of Autism S~c:trum Oiso,der

were cons idered unique. Resting stale f-:EG with 64 clecc,-odes wa, extracted by placing th~ infanl. in a d.irnly lir room in rheir mofaer's hip ,,,_,h~re the research a~sic;iant blew bJhb]es ioca1ch thefr i.me111ion. T he raw ,ignais were preproces.,ed os i11g M odified Multiscale Entropy. L()\>, higb, and mean for each c•.:rvc from intvlSE were caJcu]aLcd to ere.at~ a li:allll'~ set of 192 values. T he hest fit for the dasslfic;Hion for High r~sk and normal infa111s Y\•·as al age

9 inontlu with over 90W, accuracy. Abdulhay et " ;· (20 17). use EEG imrinsic fo1wtion pLOlsntion 10 identify

p11ttern:,; in Aul ism. They maihemalica!lyoomput~ EEG fealeres andco11:pare ASD wi1h typically developing. Jn this study they selected 10 children wit.h AS D and 10 nC\n•aucistic child;.cn within the age group of 4 to 11. T hey colkcr~d resting state EEG u~i ng 64 cl~ctnxles with a 500 H Z sa111pli1:g frequency. Ini tially the ~ignal~ w.:re bund pas.Ii fi]fcred and all the art ifacts including eye movements were removed by usi11g l ndepcnden\ Compone.111 Analysis. En:pirical J\.1o<le dceompnsirion was applic:d Lo extract lntrinsk vlodc f uncrion from each of the channels of the part icipants. 11,en pti inr by poiul pt1ls.1tious of analytit: intrin'ik modesan::computcd which is then µio1tcci Lt1 rnakc cornparison with the c.ounierp:i.rt intri11sic niode 111 annl'l ler chc1nncl. Any exisr.in_g stability ;oops are Hnc1lyzcil 1"0 1· abnormal neun.1l connccri,·i1y. 111 addition, they pcrfor:n 3D mappi11g LO visuali1e and :)pot unusual brain activiri~~. In the t"ii·st IMF of channel 3 versiis the first IMF in channel 2 for typically developing ar,ci ~utistic chi id, it was found that :he qahil ity onocal pulsati(>n pati1\vays maintained ,:consisLcn,y whiie i~ \vas rand(1m i~1 :;y µlt;-:1 l!y developing. Similar patterns were seen in channels I and 2 and 16 and 37 of non•aulistic and u1.i l istic chikfre n. Overall l l1is computa tional metl1od "-Vas able to diffcre111ia1e the ab11onnal EEG activities bcrwe.en ASD and typically developing childrc.n.

Alie et al. (201 t) use :'l-1arkuv M(Kleis with eye tracking to cl,issify Autism Spe<:t rum D isorde r. Unlike most od1er studies that collected data from children wh,1 were 3 year, or older. in this study they collect data from 6- 111onthvo!d infants. Tnere \V~re in total 32 subjec:s <W t of whid: 6 wer,;; la;cr

a t 3 years of age d~agnosl!d w itil /\SD and the rest were 1:oL. During. lh c data col!ection the ~uhje.c1·s were p1aced in fro11l ofth::ir nw thers llnd fourdiffrrcnt came.ms from different angles recorded the video for about 3 minutes. The ~ye tracking \Vas simply based on eitlu.::· the ~ubject inoked r!: the~ :r:olher's face or not. Through this they get a hi nary sequence of scbjccts' eye patrec:, which i s then ccrnvened into alphabet sequence of :1 .-,pecific lc-ugl h. Then

1he se.qucnce ""''.ls fi ltered using a low pass filter ami down sa111pled by fa~tor of 18. T his i~ done ro enhance Markov Models :o produce c1Tectivt :·cs>1lrs.

40

€/ectroencephalogram (EEG) fo( Delineating Objective Measure of Autism Spectrum Disorder

Using rh1s data, they compare Jlidden M,nkov Ivloc/els and Vrn·iable-order Markov M,xiels for lhe cl;,ssification of !\SD. Hidden Markov Modeh was able IO correctly identify 92.03% of the typically developing subject while icl::ntifying only J3.JJ% of Arni~tic ~ul~jcct. \Vhereas the Vl\·1M cotrec:Jy ide11t.ifi~d I on'/!, of the Amistic and 92 .03% of ,yp,cally devd.opi11,,; subjects. It was cicar from thi,~ re .. ~ult that Variable· Otlkr lvlarkov models are superior ir: fi :1rling Autisa!c eye pattern whjle both i\1arkov Models ;;we: the same iri f indi;1g typieal~y d~vcloping. The aut.l~ors point out rhi'.) diilere.r.ce bec.ause of ,·ari(~us spcctnm1s (>f Aut.isP1 \1.-'ith difterent eye p,u.rcrn~. Neven.hele-s.s, rh;:

VM \,f algorithm used in this st;_1ely looks effective in ~cten:.ifying Aur.isn1 i;1 an early age. Sim1larly. Liu :::I ;-11. (2015) propose ,:t m,~d1i11c icarning framcv,mrk for the diagnosis or Autism u~in.g eye movemem. T11ey utilize lwn different duta~et.s from pre.vious studie."- One of 1hc datasets had 20 A.SD children, 21 typically developing. and 20 typical developing IQ-matched children. Tlie other dataset con,pri,ed of I') ASD, 22 lntcllccrnally disabled, dild 28 typical young ,1dults and adolescents. They comp11te. R,1g of V>'ords tor Eye Cc'l<r dinatcs and Eye moven1erH) N-Grams and AOf from tile datasets. And they lrain five differentSuppor·, Vector machine 11Hxkl wilh RB F kernel. I ;acl1

nf the model used diff~.rent· form of fc~1ture.s Hke BO\V of eye coorC.inaLCs, 8O\V or eye muvcme.rn, c.ombJnation, N-GrJm.--., a11d AOl. 111e :·esJll \.vas good for both groups with C.ombi11c1tion or ruston d,1ta. Ho1weve.r, the children data~el W!th fusion \\'as the best with around 8711,) acc\lracy.

Jiung and Zhao (2017) u:-.G eye movement with deep n~ural neLworks 1n idt:nlify individual~ v,•irh Autism Spectrnm Disorder. Tbey used dataset from a previous study with 2G .-\SD arn: 19 health controls. Here the sllbjects ob,erved arou,icl 700 images from ihe OSJE datab~se. OSIE d:,tabase is a 9opular eye rraL'.king daus.ct tsed for in1ag~ ,1,alie1:cy !icnchmarking. I :irsl.,

they use C1u.~ter Fix a!gori lhm on the raw data to compute fixations and sm:c:-!des. l\ext, they work on find ing the- d:scrim!native iruages as the OSlE datas.:t !s noi s;iccjfically buil·~ fora1nism suidtes. So, hot.h groups might have tbc sam~ visual patte-rn for some of the images. For this purpo:;e, they use Flshe-r s.core mctl1od by wl~ich t.hey sc:ure each of ,hl: images and select oP.ly ,·he one w~th the higher sco:·es to be processed further. Aftt:r Lhi;-; pnx:.t:-s.-; of im~1ge ~election, they compute fix al inn maps tocliffi.::n.::nt!ai-e rix:lttoro.s t>e:tween two gtoups. ,;1 xati0!l maps i.ill~ :,;jmply a probability distribut'ion of :tll the. eye fixaLions.Inaddition, they L;.Se a Gaussian Kernel for .m1oothi11gand :1orn1;jl iie by their SLun. Norrnalizalion i-. 11su:1'ly done r1,.1 hen we are comp.:1.ring t\VO difTe:·em fix:1tion n~aps as is the case here. T;1en they compme difference or fix mi on map between tbe Autistic and 110L-Autl:-:t.ic gmup. Thi.~ is tl:eoriginal

41

l:itctrocncephelogr.im /EEG} for Dellneati11g Objeciive Measure of Autism Spectr:,m Di,;ordor

targ~t which they llsec to trnin n SAUCO'l network top-edict thc~e vabe~. SAUCON network is one of the st111.e-of-thc-arr irr:age saliency predict inn atgorhhrns. hnage saliency µr~dicrion is nbuu! predicting the\' isual pucren1 of user., given an image. SALICON network use., two VGG with 16 laycrs. One ofthc VGG uses the original imilge todeteCl lhc small salienLrcgions when~as 1he other VOG uses the down ~amplc:l in::agc: Lo detect the cenh::r of large salient n.>,gions. Al. I.he end bo:h the ou1puts arecombinecl ro get a better resulL This :)nly predict~ the image salic,icy. So, to predict the difference of fixation m:1p they add anuther convol"tion layer with Crnss l"'n tropy Loss foccrion usi ng the c>riginal Difference. of fix:itiou map. /\ext, they stud the predicted difference of fixmion maps 10 the final prediction layer. Jn this part they firs, apply t.anh function 1.u the feature, 1he11 concatenate the foature w ccors of al' rlx~tion to cousi<ler dynan:ic change of anemion. After wh:ch they reduce t~c dimension by to.sing local average pooling. At. la, I they trnin an SV1vl 10 make the fi 11al classification between /\.'ii) end control. They make use of 1he popnlar lel!.ve--one .. oul cross v~Jid~aion 10 measure the performanc1:: of thei1 nio<lcl. The ac.cr,rncy of this :11odcl showed real promise in eye tracking lor ASD wir.h about 92% accuracy.

METHODOLOGY

Curre:it techniques in prac.:t ice for idemi fying AS D a:·~ most.~y subjecti vi:! <ll~d prone to er ror and llSllltlly takes a lot of time for ii cal diagnc;sis. \ 1os! ofrhe children Wllh ASD are <!illgno.,ed af1er 3 years of age. Eidy diagnosi., is the key for reversing ur ttcating ASD r.hrongh early inh::n·cntion. As time is of an essence we need a method of di'1gnosis Lfrnt is fa,t. and c~ficii~nt unlike 1he currcm prac:.ke that CO'.Jld take months ro yea:~. Medical lr.1aging a!ld blood Lesting (Sparks cl al. 2002: Spence ct al. 2004) arc promising anc a lot of work :s being do1,c with these modalitie, to diagnose ASD. However, EEG and Eye movement 3re cosL effective HIid hence can X: accessible in l'onsumcr level. The ai,H of this rcsea.-ch is to study !Ile idemification of Atitison Spectrun: Disorde-r using EEG dming ADOS-2. Compa,·ison of :he cbssificatior: perforn:ance betw~ea EEO f•~awres c:rn pute11:ially re.suit in Lnciing the bctt~r fea1.ureset. We hypothesize as the- top performing.signal most likely ha.s 111ore oftheuniqut: da1a point_< and pallcrn of l\SD and similarly, r.hc leas1 ;1erlilnni11g .signals l1avc less of the data jK>int, and p:-,nans rela,ing 10

A2

Elar:tr::,~r,cephalogrnm (EEG} for Deiineatjng Db}JK:tive Measure of Aurjsm Sp~crmm Dls.ordc-,·

Fig1-1ni I. Lf:,-G l'roccssing Pipditot: j(w Swdy /, EEG Omrr pnipmcesst.'d using Mak om P(oeh'ne fiAlu•,1:s rhi.,· /li/H::.·liiu~ UJ trmh SVM, Lo.~istic, DNN tilltl G'aussic111 1°'/r1fve iJt1_w,.~- Models

U_

ASD. The secondr:.ry goal is toco:npare variou;-; maci1 inc le;3rni11g n lgorirlinb for rhc cla5sificatiou purp1)ses. Condilion~ like ADHD, and other learn;ng djsabilities can also share similar comparative patten~:-: for di ffcrcnl. ft,;a\t1re.s.

Machine Learning With EEG Measures During Joint Attention

We have recem,y ernp!oyecl preliminary fca,urc analysis 011 acquired raw EEG data from the work o(Jairne e, al. (201.6), wherein 1hc EEG was recorded frorn adolescents with ASD (N;::;:2.:!) and typically de\1elopi11g adok.:scents (~::::::28) \vbilc they waJched ti seri.es of30-secom.i join~ atte.nlion clidtlng video ~lips.

43

Elcclro~ncephalogrttm {F.EG) for Oeline1,ting Ot,jcctlve /tfeas1m:t of Autism SpectlWn Oisordt:r

Figure 2. EE(i Fedi!.!N Proce.\·sins Piµelirwfor Veep .l\/t11.1m! /'hirwork. t::m:h iay!!r of 1he tfet:.p nturai n,!t,,•ork. is shown in the fig1'!'~, with its Jimr.tiniMlity

First, we ilpplied the pre-processing pipeline (described i11 n.3.,) on the raw UEU time ~eri~s lo remove noisy channds ar.d d&la ~gmcnl.s coniaining movemem and ocu jar artifocls from 1he EF..G data. The prc .. pnx:essed dar.a was 1hen classif'crl using EE<~ Analyric.,· Pipeline (implemcnrcri i11 Python) (T hapsliya et ai. 20 ll!).

fr,im attention is tbc abilily to ;St)cia!ly coordinate visual mr.enLion, ~liarc a iXlinl of vie\\• with arn>rher pcroon. and process self ;!nci orhcr•rcl~.ted informatlot?.11en~e. the data rc~rie1,.•al was r)l;!rfc>r1~1ed wh ile m~1ki11g thc~ub_jects watch video clip~ that ,vm1ki help in examining jo)nt at1en1io ri. There was a roLal of 12 vi,~::::o~ each of which '-"as 30 seconds Abo111 o ne serond gap was provided hi:twee:1 ~ach video. RoLh the EEG rmd Eye moven1e:11 Vl··e.rc

44

£/€,cfroeucephalogr,;,m (EEG) for DeJine;;t,-ng Objective Meas11re of A11tism Spectrum Disorder

cr:-Ee.cteri while the parlicipanrs warc:1c<.1 tlie vide-o. A rotal of34 par-tjcipants EEG ~fata was used in this paper :-ifrer lhe preprocessing step.

Tl1..:::rc are m;iny \\•ays toexLracrfoat11re frumEEG data_ Entropies, waveiets1

FFT .!rid variou~ other statist:-:al rnc~bnds nre common}y cumputed 1"ertrures {,r\1-Fahoum and Al-fralhat 2014). lJJ :his \vork we use: Sta1lsrical and Enn-o:)Y , 1;-1.h1es. Stadsr:cal featEres con:qxise or f\·1ean. Stand;-1rd Deviatior;, a1d

cmnbin;;d mean and standard dr:vi~tion of Lht: flllered. dal.iJ. r.-ur 1.h~ frELure

analysis, we used sL.ati~1icHl and entropy VJlur:s includiug meau, sta:;11.fard dcvlation, and combined me.an and ~tHndard devta1io11 en :lie pre-proce~se:i

data. L::nlropy is compl!lcd by using Shannon crnrnpy foncrion (Lin 1991 ), whi:.;.h is J,e average rate at v.-fiich i.nformation is produced by ~-l stochastic ~uurcc of dala given by, H~. = - )~ p

1 log.

1 P,. \1ean t\rnction takes in a 20

m:1trlx con~isting of the EF.G sig11al oi" a pe.r:';on ;Jnd returns a feature vector

wi ih mean Vli~ucs for each char: net <.wer wiu(lows of :-;ignal. for the me-au, eaeh of the 128 channel'.-1 wen_• coir,putcd. For each subject a foawre vector

cousisting of mean of single channel was crealed. So, r.he mean f111H:dor I akes in a 2D ma!rix consisting of the EEG signal uf a person an<l rel urns a fec1ture vcc1or with mean value& tOr each ch;;.nncl. For' the ~ta11dard de.vintio:i, each of the 128 c)u!nneJ:.; were computed. For each stil~jccta feature vector consi..;Ling of mean or single channel was cr·eated. So, the deviation fnncrion takes in a 2D rn:-i.u-i.)( consisting of the ELlG signal of a person ;.rnd reu.1rns .a feature v.;.~tt.Or with :..;ranch!rd rlr:viar.ion vtJues for ca:..:h channel. This L~ sho,v11 in ihe F!gun: 1.

For classification SVM, Logistic, Oei::r Net1ral Nd\.Hirk (Dl\N ;i, and

Gaussian Naive Bayc~ is t:sed. For tbedee.p ni;ural 11d wo;·k (see Figure 2·1 with five hidden layers with sigmoid a.;.;tivaficm function i . .; used. Fo.t op:imiz;-1tion

c.megorical cross entropy for loss and Adamax optimiz.er(Frci valds 2.11d Liepins 2017) is u.,t.:d. \Ve cl~ptured th;-ee difterent fe-rnur~ ser; entropy i:eaturi.:s, 1'1 T

:fohle l. Clus_,·tficution Ac1.'UttJ(.'_•.,:o_fE/.:,U du ting .loinr Ai/em ion Swtly. 'f Jw F.r.:ropy, FFJ; Afrtm and Sumdr.m.f Oe·,,i:uion 1;t1h,e.1: rm:· p,i,,.t~;: . .fi)r etid1. dr.1s.f{_{ier us,!dfor !hi.r Irnciy

MCml Shi

O,U-, i O . .Sj 0.5.>

050

(J55

45

l:J&e.trocnc~phalog,am (EEG) for Oe/Jneating Objective ,ite~uure of Aullsm SpMtrom Diso<dcr

and stat ist ical leatures. We also calculate mean, and surnlard deviatioi;. In total there. ai·e 4 difforent fomun:s from EEG aml 4 dii'fr.,rent models for each type of c las,ifier ancl, overall 1hcre arc 16 diffe.re.nr mcx\cl variations base.cl on the fomurcs (4 te.iture-set x 4 classil iers). For each fear;,re there are !:tree models for each algori1.hm, two rr,ocleh usi:ig 1-'canrre Sele.:rion aml the 11:irci one without using any foau,re se.lect:c)n For r-eacmc selection PC/\ t~id sequt!nlial feature selc-:.:tjon is used.

Classification of EEG During Joint Attention: Results

Tl1c Table l. p;·c,ems an analysis anrl c:omparison of EE(j darn. Note that two models were crc:ited for each :n:">dei with on:y EEG aad combined <laia hy using PCA ,111d wirhoet usi11g PCA. Like SVM with l'CA and without PCA. For some models wirh PCA did bener while for some without PCA did better. Fur e.,ample, D~'.'J almost aiways without using PCA did worse hec2use Dt the cu,·se D[ di mensionality. Th~ highest performini;; SVM witl1 about 56% accm,,cy was using 1--l"T with 21! the features witholll PCA. T!1e highest perti.)iming Logi.stic regrcssio:1 '.vith 78% accuracy wa.-. using FFT without PCA. SVM, Logistic Regre;-;•don~ und Gaussi;m Naive llayes do het.:cr without PCA which means that wirh PCA it loses data points thar these models find c,e.ful. T 11is i , intcres,ing because PC\ is supposed to tiud the mos.t disc1 i1111nrtnt features and remove rtdundm~t or n<•fsy fea1ures. A ,~d this issllpposecl:ohelpr.1aehi11e learr.ing model, prod11cc~!l::r resuils. For SVM most models wilh PCA d ie: better e.,cqt 1he bigl1es: performing model. Tilis mi_ghl. mean that the Enlropy data is more linear than the or her dl-lt.asets. Fnr DN!\ the curse of tlimer.sionalily is nh\'ious. Wnerea,; for Gaussian Naive Bayes all 1hc high pcr:"ormi ng mo,iels did not use PCA except the one witl1 C:EG mean. l his isaa exception a11d must be clue to :he naw recf !he I cEG mean data. Bui in general case-l\ai vc Bayes <.locs be1ter wi1hou1 PCA. This might be since probabilistic mode's arc able ro maks sense of rigkr dimensional dataset much ~asier :han ocher modeis like DNN. Thci, with using Sequential Feature. Selection algor ithm al.mosl all the modeb perfonned h~tter lhan either PCJ\ or 110 Feature Sell!Ction.

In th's study we have used PCA, am.I Se<JUeutial Fe1Lure S~lee1,011 algorithms. The:·c are oth~r l'eature Selection algori,bms like Ge;,eric algorithm, Particle Swcrm Optimization, and TWJSTwhich c:an be compared to find features w optimize the. performanc~ of the. :11odd~. A~so, this will tell us which feature selection algoritlom wil! work beuer for tl1e cor.1bi11cd

45

Elcctroe11cephalogr~m [EEG) for Def!neati119 Objective Mf'as,:,re of Autism Spectrum Wsorder

d.:Ha sc1s. Gaw.;sian Naiv~ Bayes with .~ome. of 1.he- tea tu res had perfect score. But we need to reproduce this rcsulr wirh huge n=.1:nber of pa.rticipanF; to I).;

able to 11"-C lhis \n :~ t:linical set~ing. Ct:rrcnt nL1Jnber of 34 panicipa~11.:s is loo lt)W 10 cor:firm our results. Hm\.·cver, this is ii fi,:sr. step towards d.-:vcloplng an optima: Auj-;m Ding1io:-;is system.

EEG Coherence During Live Social Interaction

The notion that soci:-tl brain .~y:srer;.1 FC n~ay be ~;,. usefrl i[•dex or !)ocial impairrnem ,s suggested by both the liter~lt!re (Mundy 2016; Jaime ct al.2016) and by our prdimjnary fi11dings obtained from our pilot ~ample Clli11posed or individunb. between the ages of 5 and 17 y eill'S who completed an ADOS-2 assessmc-.nt while we sirnullaneClusfy recorded their E£(j_ Despit.c a small sa1npie size (ASD = 8; TD ~ 9), our prd!mint~ry result'~ indicate a tre~1di:1g negative association bclwc;._;ll right hemisphere dcll"a and thcla band Ef-:(J coh~rY~nc~ and level of socia! symplom scv~riry (according to the ADOS-2 a:gorichm scoring) in children with ,\SD (,ee Table 2 below), bu not in"'" pilot sa,nple of typically developing (TD) chi.!dren. Our µreliminary resnl:s palm H co11ceptual plCturc th.er:: is i!• line wirli our prior \vork evaluating EEG coherence durillg joint social anent ion perception in ASD (Jaime ct al. 2016), 1.hat there are diagnostic group difference~ in the ?.sso{:iar.-:on betwe-:n

right he.misphere- front~d- rcmporal--paieu:J re and standardized measures nf soci~':.I functioning. Su{:h diagnoslic grm.:p di ffcr~ncc.s in FC association patLerns reflect a tendency fer children ,,vitb imrain~d ,.;ocial capucity to hav ... ~ idio!-iyncrntic parterr1:-i or social hraira sy:.;tem functional organizaiion rchnivc to Lypical nern·odevelop~ent. Tlms, EEG measures of social brai!l system FC ncquired during live :-.ocial inte:rc1c\"ion shO\\.'S promise as a candidate 1t01'.-i11va.-;ivi:: hiomarker of early e1:lerging i:lberra:1t social ncutocog,~itivc dysfunction in ASD.

EEG Acquisition and Pre-Processing

Our prel irn inary FC 1:easures \.Vere analyzed from t:ach pilnL ~ubjl:e!'s EEG record lng, acquired tl1rougl10L1l th(; critin; (iur:-1tion ot·rhe ADOS-2. \Ve used a 32-dmnncl LiveAmp \.1..-irek;-;s EEG syst~m with ac~ive electrodes and a digilal samp'ing rate of250 Hz (Brain Products GrnbH) tc,r EEG lime series acquls1t1on. Use of a wirek-ss EEG syslem :tllowcd for head movements

47

Electroencephalogram (EEGJ f0t· DeliruJ.atlng Obj(!ct/vc Me,7sure of A.urism Sf)(tctrum Disord•r

flgure J. E£(j Fca,ure Proc-essing and Cl:!.v:ifin,,ion. T::e row signcfr naJuilr.d jrr,m cm £F,(i nre stored (md i:,; _\·t ,J~fft'lt:t.l w preproc.·e.~.··•iflg 10 obwir. elem, :im~-serie.\ data, O,uxt !It;.\• C.'.m·-;p{eies, .1l1e dean datil fa p,1.r.w:d 1/.n.:ugh band J}las.fifters wul

,fomu.re e.tln:iction fa 1urfnrmed. 1'hen the exnw.:tt:dji:a111r~.\· rs re fed iri!o a clt;.nifu!r. which u3e:; cm,s w;lidauon 1oew1luatr i ;,; pe1J01mance'd.:J1e,1ding on hn·v ;, prt!dfr rs /\M) rmd Tn cinss tahels.

..... c . ...... i ""''"':'_j

: . ' ~{}~~~-.. ~'

~;;.~=· l _,_ ... s··::::-J •.. ,..-c,..·.

. ....... j~~ti ~ .. __ ,,_, ..... ~~~ ....

®Im~ '

0~ "- . ,,,

e

anJ lhe active electro<!~~ increased speed n' appEcation thereby increasing probability of st1ccessfu l EEG d,,u, :,cqui, ition with special populations.

Ail 32 c:han~c-ts wert: continuousty 1\::corded D~ing th.; FC1. c;Jectrocle as r~ferencc. To maximiU! t·hc c.ous?stency of lhc recordi!lg quc:.lity dcro-;,...:;

conditio ns, a singk epoch was ,ccorded per experimental eondilion. In be tween cp<)Ch recordings an impedance check will he rx~rfo:mcd. This \vas rCSlJlred in 6 d ifferent epochs per subject. Prior to 1hc :·ccnrding of e ~ch expcrin:cutal epoch, ~ 90 second epoch of eyes d osed while resting will lle :1::con.le.::: . This served as a nccc.ssary hasc!in(- mcu·ic fo1 r!1e EEG an~lysis. After accp1is i1io11 , the raw EE(; datu output ,v::.is i .-npon cd into the op~n-soun.:e ,'vl ATLA!l toolbox: EEGLAB (De lorme and M akeig 2004). Nex1, following prcpmcessin)l pipeline ,s applied:

l . Rcmovt: low freque.n~y bHscline drift with a J H,. l,igh-pass filter. 2. Remove 50-60 H:,; AC l ine noise by a;,plying the CleanLine plugin. 3. Clean continuous raw data using rhe clc~m_r.iwda1a plugin (M t1llen ct

al. 201 5). The.clean_r::wdala plugin fi rst performs Oudcham1d rejection based un lwo critc1 ia: (1) channeJs rhat have rial signals longer 1hm 5

seconds and (2) cl1ar.nels poorly co1Tdated with udja1,,;e111 clrnnnds. It

48

Electroeaccph<1/09ra,n (EEG) for Delino.1ting Objective Me-as-11f't! of Aufism Spectwm Dtsorde.r

then applies ~u·tjfact subspac.e rec.011structjo1: (ASR) -- an algoritlun l.har removes 11cmsta1.ionary, higb vartance signals from tl1e EEG U1en

u:-.es calibrarior: data ( 1 ni in sections of clec1n l·~ECi) to recons1;-11ct £be mls~ing dat~t u~ing a spatial mixiHg matrix.

4. lnterpolate tc.mov(;d channels.

j. Re-ret'erei:ce channel~ to average referetx-e. 6. Separarc ncu1-brai11 m·1\fac1~ frcm tlie E!~Ci recording via EEGLAB's

Independent Componcllt Ana!y,i, {IC/\)1• Briefly, !CA irsvolves the

linc.-1r d~compo~\tion of the aggregate channel activity im,) a :-;ctics of

indcpcndcIH cornpor.ents t.hm an~ spatiaily rih:~red from :he recorded EEG time ~erie-s. Components .a:pr~semi:ig ~ye. cardiac, ~mJ muscle artifocl arc rc.m«wt!d and cornponcnL:-; repres,;::11ti:1g genuine. brn~n activity are rcrained.

Tahle 2. ADOS-2 Score olthe .t\SJ) vs TD so'.r/JjC!ct·>. /!ere. 8 s!1bjeas 11-'r!n~ d.iagno.•wd M:h AST> {aho"'' horiz.on.ral lilie ) and the orlwrs wett: ,ypically de ve/op~·.,1;; (TDi

i\1 12

I" ;;{I

:M

M

\I

" ; 10

f--------- tr

!i It . !f,

17

19

ll

:6

!n

AS[;

ASLl

A!'-il)

,\SD

!

Tl)

m ~ ------+------11)

! 'J'I)

49

E/ectroe,,r;ephalogf'tfm (El!GJ for Delineating Ob_iectfve M£ilsure of Autism Sµ~crrum Dlsotder

EEG Measures of Functional Connectivity

We firsL cxtracl.ed !80-sccond epochs bcgini;ing fron' the miud le one-third ;:,onion of ead1 subjec1's prc-r,nJCe,sed EEG time series 10 calcuh,te a flmcrjo11 ;J conncctivi,y ( l'C) measure of tht eng;;ged social brain sysiern. Wi1h each subjt~t'se;,ochcd F.EG time series trca1cd as ad,screL<!-time ~ignill

u = ·i;, (1.) for EEG cha,u,el i , we used EF.G cohe.re11<·~ ,1, a ,,ari~blc of FC.

E EGcuherencc,c: normalized magniruJe-sqm,red cohercncc(~1SC), c;. ( u..·), is a sta1.is!ical ~Ii mate of the amrnmt ofp!latic synchro11y h~1ween two EEG

time serie s> u and ri: c:i ', .• ,,)· = lrP (~·)[2 t r'rf; 1

1~:)o fw ).•1 where the sq_lrnre-~i " ' \ . ,., ' . '1. ..... ~ \ _.

ntagnitude of tl1e cn>~s spectrum <lcn:.;ity r,-', .. ~ ( u.J )r'' (a measure of co~variance)

hc1n:e,~n th~ two signals !) .and ·1; ttl a given frequency ~' , is normalized. by !he Power Specrrnl Dens,lies (PS Os) (variance} of ea<:h channel P.,. and '""·

so that OS c,:. (t,c.;; .S 1. I-I ighe!r val11es represent greater syncl1ronou::. act.ivily

between disrin<.;r channe'.s whereas lower vali.les represent reduced or non• synchn1nous activity (Nunez and Srinivasan 2006). Cohcre1"~e is a fu nction or ~·re-qL1cncy~ lo compute a ~ingle sirni larity metric between a pai r af signals, we in1·egzate over frequer.cy to obtain l<>Utl power (or variani.:~ i1: a slalistica)

T

.sense) P = 1- ('c' k-\ w hee T is :he exttnl of frequenC)' cornpllnem,

I) :7' ~) b,•, I

sampled. T he MSC oi • signal which it~!r µrorn1Cr.< r.o variance (in the statislil';.; I sense; and hence .~

1 =-=~. , gives tt convenicr. t, norrnalized metric

of sim' lancy. Acco,diagly. im ra-hemi,pheric MSC between e lectrode posi1ions thar.

are !)path"llly coll.(}Cil{ed over ar~a~ comprt~ing tJ1e social brc.i11 s.y~1e111 (Saxe 20()/,; Adolphs 2009) wer~ examined. Ekcrrode pairs were se'ected based on Homane~ a l.'s [ 19l\7; elecrrode placernentccrrela,csofcortical locatio11. Using the int.ernational I0/20 placcmenr sys1.em (Klem et al. 1999). the follmv:11g electrodes w ere selected: F7. F8, T7. T8. TP9. TPIO, P7. P8, C1, aed C4.

Classification of EEG During ADOS-2: Results

\Ve genc11tiec1 fi'vc fealure c.:et:i catcg.orizc<l according to Lhe frequency bands: 1) del~a. 2) thei·a, 3) alpha, 4) beta and 5) gamina 1,,vith each sel" reprcsemlng the a'llplimdc and :x,wer of the signal :i om euc,1 e lectrode. The.,c fcawr::

50

Electroencephalogram (EEG) for Delineating Objective Mo~sc,re of Autism Spectrum Disorde,

sets were entered to 43 different Cliissifiers yielding precision rates, recall rates, F l scores, and percent accuracy. We identified six the top performing classifiers: Random forest, Logistic, Bagging, JR IP, LMT and Ada Boost Ml .

The six top performing classifiers for the 5-band feamre set are listed in Table 3. TheJR!Pclassificr yielded the highest percent accuracy with 98.06% indicating that a 5-band feature set collected during an ADOS-2 test classifies ,!diagnosis of ASD with grcaterthan 90% accuracy. From these six classifiers, the AdaBoostM l classifier yielded the lowest percent accuracy at 92.14%.

The evaluation results in Table 3 were calculated based on features from all electrodes. We also conducted an evaluation by selecting only F7, F8, T7, T8, TP\J, TPlO, P7, P8, C3 and C4 electrodes based on Homan et al.'s j 1987 j electrode placement correlates of cortical location. The results of this evaluation are listed in Table 4. When comparing the results, it was observed that the Random Forest classifier yielded the highest percent accuracy with 97.04%. The AdaBoostM l classifier yielded the lowest percent accuracy at 79.75%.

Table 3. Precision, ReCtll!, Fl and Accuracy of six classifiers 11sedj(Jrclas.r1fica1ion of ££G dur;ng ADOS-2

f'n-cision l{t'cull

09' 0.98 0.9S:

Logisti: 096 0.96 0.% 96.6.1>'l

0.'}5 {J.95 0.95 95.66~

HUI' 0,9S 098 I I.MT 0.95 095 0.9~ 95.79<;

0.92 0., ll.91 92. !4';{-

Ta/1/e 4 . Precision, Recall, Fl and Accumt.:y ofsix classifiers used.for cl.t1.uif,ccairm of EEC during ADOS·2 us ing only a selected se, ojfeatlfff!S

Cl.t$.1>ifier l'rcclilon Fl ,h'CtlrllCY

R~11dom forr,st 0.97 0.97

0.'4 0.84 I H72%

0.95 0.95 0.95 95.50~-

JR!I' 0 .94 0.94

LMT O.:B ! 0.82

AJ:iAws1MI 1).1,:0

51

Electl'oencepl)ulogram (EEG) for De/Jneating Objective MeasurtJ of Autism Specr,·um Discm:ier

DISCUSSION AND FUTURE OUTLOOK

Duc: 10 it3 iow co.;t "nct feasihility, electroeurnphalogrnphy \EJ.ov) shows pt,ccnrial as a!l enec:_ivt: n~urophysio logical inslrument rn the classific;.ilion or 1\SD (Lenan owic1. ,rnd I ,oo 2014: Snyder e: al. 2015; Gloss er a l. 20i6), anrl there is ~-mergi:ig evidcucc that--cn:11bined w ith machi ne learning apprnachc:s--quantilative measuresofE F,G c:in pr~dic1 ASD with hi~h hovels of,.,rc, itivity and specificity (lfosl et al. W 18; Grossi ~l al . 2017; Djemal ctal. 201.7). An advantage of EEG is its ubility to be applied to ecologica l!y v.!licl contexts (i.e., per-..on-to-pc1.~on social imcraction) via wircles., sobtions thus allowing for the ~imul1 anc-.ous acquisition of data from 1nullipl~ part.;.cipants ic rerd-wurld scrlir:gs.

To establish p:-uot of concept-th:::r our classifiers show uli;i ty to predic:. fearnres in l i11e wi1h diagnostic n iteria of ASD-wc colkct b iobehavioral metrics wi1·hin the context uf standardized lasks 11sed in a gold sfandmd asses!-menl of ASD syrnptomHtology: The /\uti sm D iagnos[ic Ob:,;ervatio11 Schedule Second Edition (i\DOS-2) (Uotham ei al. 2007), The ADOS-2 has bctn careful)y di.-!velopeJ ro create sn'1pshuts of naturnl isrir sol'ial ..;ctnarios that can reveal observable features central to ASD (i.c .• _il1i11l HtlC 11t ion, soc.in! o, crtures). thereby :illowing us lo mca~ure brain acri,ity that nre temporally concurrent with the,eo bserv:,bk ASD features within relatively bric f pcrio<ls. It is also in,portant n, nme that we dicl 1101 use lhc·s..: ADOS-2 tasks a , a cl inic~] tool to diagno:,:;c iJartlCJ j1<1nts; rather, we.capilali i.ed on i.hescmi~:itrncturcd a!u1 st.mdardlzed riarur~ nf rhe:,;;c social la~ks in ll1c /\DOS-2 tu cr.:atc:: a context that engages the social brn:n sy~~~m and elicits joint visual aa .;;nt.ion behavior fm arquis iL,on of h10behav•oral m.:tri" ' · Thus, part icipa111s recruitc<I for this :,;turly IHt\•'c: ,1 lrcudy t"eceived a diagnosis of ASD hy a chnic:al profcs~ioEal prior cu enrolling in 1hjs slutly.

D~e wits high tcn:pnrnl r~solmion and kasibil iry, clecttlkiiceph,i;ogrnphy (EEG) shows po1enrial as a~1 d-Yective ncun,phys iological insln iment in the classification nf ASO (Lenar:,,wicz and Loo 20 i 4 : Snyder e r. al. 2[) 15; (.l]oss et a l. 2016). An a,h·amage of EE() is its ability 10 oc applied w ecologically va1id contcxls via wirel!!sS solrnions that allow fort he .sinn1J1;i11eous acqllisition of data from mul:iplc par1icip211ls. Tl:is nrnkes EEG an apprnpri:rlc choice l·or cx11mlni:1g rdevam neurophysiolog:cal feature~of !\SD i11 re<ll-\vndd selt 1ngs

(Lee :m<1 Ta11 2006). Dc."pite lhcs~ advantages. mosl I ~F.G research occurs in :1ighly control!cd c:xperimenlnl e'.1vinmmems, regairing datu coll t cwd over

52

Elactroenc<Jf,)f::;ilogTam (EEGJ for Dclinl!ating Objac!ive fllosare of Autism Spectrum Disorder

many trials v.-it11 m!nimal head rn.oveme.nt. V•lc \Viii ilddrcss this Ce.fic.ie.ncy hy comhining EEG and eye lracker usage in ihc fut1rn:: studies

Ear~y diagnosis :s crucial for s.uc.::~~s.ful Lrc~tlmcnl of ASD . .Although pmgress l1as been tnade to acc:urdely di~1gnn;;c ASD, il i-; far from ideal ,J)11wsrnt 2008). lr oft.en requi:-es various .sub_jcc1jv~ measures, bciJ;:v:ora'. assessments, obs~-rvations fron1 carctakcfs over a period 1·oc:orrectly di,1gno-;e ASD. fvL~n •xith this 1edlous resting often individuals are :nisdi<1gnosed. Hmvever, there remains. proin;se in thi.: lkvelcprnent of accuralt: fkwc1io1: -.rsing subjective modalities of EEG, an:i Eye rnnvemenrs. In the fulu:·c we will ohtain lwo sets of hiohehaviora] me,::.smes re.presenting j oim aucr:tion: fu ncional intcgrarion or ncurocognjrjve nel work:-; i1..-..socia1.e<l with the social hrnin (i.e. 1 EEG metrics) <1nd visucil behavior (i.e. cyl: rrac.king rr.ctrics). Regar<li:tg visual behavior, we wi II coJ\ecr, analyze, and produce a b.atwry ,Jf traditional positional eye movement met.f:cs thought to be pot.cnti::ti indicators ofjoim attenhon, including number offi~mtons CJucob an:1 Karn 2003), fixation durations (Fitts e; al. !9.50; Ju,1 arnl Carpc.nler 1976), ,u1J rn:niber of n-.:grt.:ssitms (Aluma et aL 2014), during naturalis1ic1 ciync.mic communication tasks.

REFERENCES

Abd:iihay, F .. Alafeef, M., Hadoush, H., & Alomari, N. (2017). Frequency 3d m:ipping and imcr-channi2;l srr:bilit)• of EEG intriJJsic fu11ctio11 pt1ls,:nion Indic:mor.-; toward.~ nuLism spectrum diagnosis. In Electrical a,uf Elei:-tronics

!:·nxineering Corrf"eH:nu: (.JI f"l'.FC,J , 2017 lOrhJordm,.1i(m. lmenwtional. lLEE l O. l l09iJIEEEC.20l 7.8051416

Ado"lµh~. R. (2009). Tht social hr.1in: Neural hasis of so'..:ial k1i0wledge.

Annual RevieY.: of Psyclwlog)\ 60( 1 }, 693-'716. cloi:] 0. ~ I 46/annure-v. psyc1.6(). l ! 0707.1635 l 4 PM If): I 8771 ?>88

AI-Fahoum, A. S., & Al-Fn1ihat.A. A. (2014). Me.thodsofEEU signal katures extracrion nsing linca'.· analy.sis i11 frequency and time-frequency domains

JSRN Neums,:,irnce, 20/4, l-7. Joi:,(). l 155i20l4/730218

Alic, 0., Mahoor, :,,1. H., Mattson, W. L AndcrsO!l, D.R., & Messinger, D.

S. (2011 ). At;alys~s of eye gaze patcern of i nfanrs at flsk of autism spectrnm disord:::rusing M,irkov model.-:. In Apl1/.iaaion.srfC:ompul(:r Vision (\VACV),

20/ f IEEE Workshop on. IEEE. l(J.l 109/WACV.20! 1.57; 1515

53

Electroencephalogram (EEG} for Delineating Objective Measure of Autism Specfrum msorcter

Azuma, M., Minamoto, T., Yaoi, K .. Osaka, M., & Osaka, N. (2014). Effect of memory load in eye movement comrol: A study using the reading span test. Jot1rnal ofl:."ye Movement Resea1d1, 7(5), 1- 9.

Bal, E., Barden, E .. Lamb, D., Van I leckc, A. V. , Denver, J. W., And Porges. S. W. 20 IO. Emotion Recognition in Children with Aurism Spectrum Disorders: Relations to Eye Gaze and Autonomic Stare . .lounwl of A111ism and Developmental Disorders, 40(3), 358-370.

Baron-Cohen, S., Leslie, A. M., & And Frith, U. (1 985). Does the autistic child have a "theory of mind"'! Cog11iti011, 21( I), 37---46. doi: 10.1016/00I0-0277(85)90022-8 PMID: 2934210

Barresi, J., & And Moore, C. (1996). Intentional relations and social understanding. Behavioml and Bmin Sciences, 19(] ), I 07-122. doi: 10. 1017/ S0 140525X0004l 790

Bedford , R , Elsabbagh, M ., Gliga, T., Pickles. A., Sen ju. A., Charman, T .. & Johnson, l'v1. H. (2012). Precursors to Social and Comnrnnication Difficulties in Infants At-Risk for Autism: Gaze Following and Auentional Engagement. Journul o/Autism and Developmentcrl Disorders, 42( I 0), 2208- 2218.

Billeci, L., Narzisi, A., Campa1clli, G., Crilaci, G., Calderoni, S., Gagl iano, A ..... Murarori, F. (2016). Disentangling 1he ini1iation from the response in joint arrcnrion: An eye-tracking study in toddlers wirh autism spectrum disorders. Trans/Mional Psychiatry, 6(5), e808. doi: I 0. I 038/tp.20 l6. 75 PM ID:27 l87230

Black . .\1. H., Chen, N. T. M., Iyer, K. K., Lipp, 0. V., Dolle, S., Falkmer, :\1., ... Girdler, S. (2017). Mechanisms of facial emotion recognition in autism spectrum disorders: ln;ights ti-om eye rracking and elecrrocncephalography. Neuroscience and Biobelwvioml Reviews, 80, 488- 515. doi: JO. l0l 6/j. neubiorev.2017.06.016 PMID:28698082

Bosl. W., Tierney,/\., Tager-Flusberg, H., &Nelson, C.(20 11 ). EEG complexity as a biomarker for aurism spectrum disorder risk.. I.JMC Medicine, 9, I, 18.

Bos!, W. J., Tagcr-Flusbcrg, H., & '.'lelson, C. A. (20!8). EEG analytics for early detection of auLism spec1rurn disorder: a datu-driven approach. Scie111ific Repons, 8(1), 6828.

54

E/ecirccnc.ephalogram (E.EG} for Delinerttlng Obj8'ctive Measure o( Autism S.oectmm Disorder

Baescher, A. V., Cidav, Z .. Knap~ • .'.l., & Mandell, D. S (2014). Cosrs of aarjsm speclrurn di.sorde~s in the Lni:cd Kingdom and the united states. },\MA J'ed1atrif:s, 168(8), 721-728. doi: l I). I 001ijamapc:diatrics.201 4.2 l () PMID:24911948

Bullerv,:onh, G., & Jarrett, N. (199 ! ). \Vrlat m?nd.s have in common is space: Spatial mccl1a:1!sms ~crvJ:1gjoinL vis uni attemio~ i:1 infancy. British Journ.r.il ,:fDeveiopmentnl Psyclw!ogy, 9(1), 55-72.. Joi: ,0.1 l; J/j.2014-835X.l99 l. th00862.x

Caruana. t\i., H:ock~.1.,&Vloo1gar,A.C2015).A front tcmpo:.--oparicLalnetwork LOJr.1mon ~o ini Liming :md responding tojoim attention hds. _,V,mrofou.:w~- 108, 14-,16. doi:JO.JOJ6/1.nenroimag~.20!4.l2.04l PMID:25514111

Cas,elli, E, r-r;th,C., Happe,F.,& Frith, LJ. (2002). AuLism, i\spergasy11dro1Ee

and bra~n mechani:-.;ms for die aLtri but ion of me11·'.al states to anim:..1ted shapes. Bram, 125(8), 1839-]8,19. doi::0.109:ihrain/awfl89 PMID: l 2 l 35974

Cltira-Tegmark, M. (20 I 6). Social ancntio:i in ASD: A review and met;;­m:alysis of eye-tracking sludie:-;. !?esear,:h in Dt}Pelopmen.t,:i/ Disabilities, 48. 79 93. cloi: IO. 1016(1.ridd.2015.10.01, PMI0:26547134

Cobcn, R., Clarke, A. R., Hudspeth, W., & 13arry, R. J (2008). EEG power

.and :.:nhL:.rcncc i!1 at.J.ist.ic spectn1m disorder. Cl.inicallileurophysic:,fogy, J J 9(5), 1002-1009. doi: I 0. IO I 6/j.cl;r,pil.2008.01.()J:l P\HD: I 833 I 812

Damarla, S. R., Keller, T. A., K•tna, R. K., Chers:assky, V. L., Williams, D. L. \fi1"11ew, 'sf. J., & .lust, \1. A. (2010). Cortic,J 11nderconncctivi1y coupled with preserved vi~uospatial cognition in autism: Evlctcn~e from anF:VIRI study of an embedded figures task. A1<rism Research, 3(5), 273-r/9, doi: 10.1002/ aur. l 53 P.'.l!D:20740492

Dmv~on, G. (2008). Eatly bebwioral intervention, brJ.in p1asticity: ;md the

prevention of autism :spectrum <lisord('r. Developmem and P.\):cho_,J(i'iholo){ii 20(3), 775-803. doi: IO.JOJ7/S0954579108000370 PMTD: 18606031

Del Valle Rubido, M., llfocracken, J. T., Hollander, E., Snic, F., Noeldeke, .1., Boak, L., .. , Umbricht, D. (20l8). In Sca,·cl1 of Biomarkcrs for Autism Spectrum Disorder. Au!ism Resemd·;, 11(11), 1567 1579. doi:J0.!002/ aur.2026 l'MID:3!):l246'i6

55

Ei•ctroencepht,Jogrnm (E£G) for Delintating Objective Me,ii.Stffe of Autism Spe.ctn,m O,sorck,·

Delorme, A., & Makcig, S. (2~)- EEGLah: An open socir<Ce toolbox for analysis of sjngJ(!-trial EEG dynamics including indepemk nt co111ponenr analysis . .lounral o}Neum scicnr.« Merhod,1·, U4(1;, I, 9--2i. doi: J().J0"/6/j. jnc.imeth.20(H. I0.009 !'MID: I~ 102499

Djt:n:al , R., Alsharnhi. K .. [brahim, S., &Alsuwaile:u, A. ('2017). EEG-base<l computer aided diagnn~is of autism spenn1111 disonkrusing wHvc]er,.e;itropy, c.nd ann. llioMed Research /i,temalioi,al. PMID:28484720

Duchowski, A. T., Krcjt,., K .,KrcjtL, L, Ride, C .. Nie<1z;e1ska, A .. Kiefer, P., ... Giannopnuros, I. (20 IS). T11c f :,dex of l'upill"ry Acti,ity: Measuring Cognitive Loatl Vjs-il-vi~ Tusk l)ifficulty with Pu pil OsCJ;lation. fn Proceeding.\ (d° !he 2018 Cf-JI Cun}ere11ce m1 /iumon n,c1ors in Co111pu1ing S:,.,1ems. ACM. I.0. 1 !45/3 173574.3 173856

Falck-Y!tcr, T., Fernell, E., Hedvall , L., Von Hofsten. C., & Gillberg, C. (2012). Gaze Pcrfonna11ce in Cllildren with Autism Spcclru m Dis0rde.r when ()b~erving Communicative A cl'ions. Journal oj'Autfa·m and DP.l1elopmen1al Disordets, 4Z( 10), 2216 -2245.

F.1lck-Yttcr, T .. T hornp, E., & Bolre, S. (2015). llric1· Report: Lack of Procc.,~ing Bias for th~ Objects Otha People /\\tend 10 in ]-Year-Old., with 1\cori,m . .Jo,mw! uJAutwi!llnd Dew!(;pmenral Di.,orders, 45(6), 1897-1904.

Fitts, P. M., Jones, R. E., & Mi!ton.J. I .. {1950). !'ye Movements or Aircraft Pih>ts During Insrrnme.nt~Landi11g Appro11ches. A1(ronamicaJ 1-:11gineering l?ei>iP.11. 9(2), 24-29.

Fletch~r-Watsun, S., I .eekam, S. R., Benson, V., Frank. M., & Finulay, l. (2009). L::ye-mov,.:;.menls reveal aacntion to sodal infonuation in aLJ Lism spcc1rum disorder. ,Veumpsy:hnlogia, 47(1), 2-18- 257. doi:10.1016/_;. neuropsychologi:1.2008.07.016 PM IO: 18706434

Frazier, T. W .. K lingc111io, E. W .. Parikh, S., Speer. l .. , Slraass, M. S., fag, C., ... Youngstrom, E . A. (2018). Dcvelopmcn, and Validation of Ohjocrive and Q11antitative EyeTrack.ing-Base<l ~-1easmes of Autism Risk a11d Symptom Levels. Journal oj'1h,~ . .l\mericcm.Acadtimyof Child mul Adof<, rceru P.".H .. :hiatry, 57(11), 858-866. doi: 10.101 6/j.jasc.20! 8.06.023 PMID:30392(,2.7

Freivalds, K .• &. Liepins. R. (2017}. !mpnwir.g r}re. neural GPU ord·,.ftc<:u,re f(n· al3oriihm learniH_g. arXiv preprint m-Xiv: 1702.0&727

56

Eic-crroencaph:,/ogram (EEG) fer Dellnc.itiny Objective Me~$1Jfe of Aulism SPfiCl!'11,n DisordRr

Glos,, D., Varn:a, J. K., Pringshsim, T., & NHwc1, M. R. (2016). Prnctice advisory: The utility of EEG theta/h~~ITt po·wer rnt]o in ADHD diagn(J.~is n:por1 of rh~ guidclim:::. d;~velopmern, disstminatirm, and implementation subcomnjttec of th.;; American ~caderny of neurology . . Neurology. 1 O~ 1212. P\1lD:27760867

G-1rdor, I., Eilbott. J. A., Feldman, R., Pclph0 ey, K . . A., & Vander Wyk. B. C. (2013). Social., rewmd, anC attention brain ndworks are involved when cnline bids for joint a::re!ll"iOlJ are mel ,,..·ith congrnc:.nr vc1·sc-. incongruent n.:spunses. So,::ial Neuroscience, 8(6}, 544 554. doi: ~ 0.1080/ 17470919.20] 1832374 PMID:24()44427

Gorham, K .. Risi, S., Pickles, A.,&. Lord, C. (2007). The mtt.isrn Ji,,gno,1'c observation scl1edu]e: revis.ed algorithms for improved Oic.gno.,1.ic validity . .Journal of.4.urism. and Developmeni.af Oi.vo,·rh:rs, 37(4-), 613.

GroS:-ij, E., Olivieri, C., & Bu seem&, J\1. {2017). Dia.~nosis of ~:L:tism through EEG pnx:ess~d by adva11ceJ cumpl11r11ionaI algorifruns: A pilor srndy. Cornpuicr.h1e1lwd.r and Prognims in Himm~,iicin.(1, 142, 73 ·;;g_ doi: l 0. 10 l 6/j. c•mpb.20 17.02.002 PMID 2832S448

Guillon. Q., Hadjikhani. N., Baducl, S, & Roge, B. (2014). Visual social atlt'lltion in ;u1tisrn spectrun1 cli::.onk:: Insjghts from eye nacking s:':..1die,')

,-,,,,!t1uro.}dence and Biobehaviurol l<eviews, 42, 279-297. doi: 1.0.] 016/j nei,hiorcv.2014.o:l.O 11 PM Jf):24694721

H1D, E. L, & Fr;th, U. (2003). Undemanding autism: insights from mind and hrain. Phi/o.rnphit:al h-ons(lctilm.) of rh.e Royal So,:ic!t)' B: Biolo,~icof S,.frnces, 358(1t,3D), 28 l.

Homan, KW., Herman,.!.,& Pt,1,ly, P. ( 1987). Ce•·ebr,il location ofinternmioPal 1.0-20 system ck:clrode plac:::mcnt. Electroencephalography and Chnicr.d Neurophy,·iolugy, 66(4), 376-382. dui: 10.1016/0013-4694(87)90205 9

P\'llD:2435517

Ho:ier, 5 ., Leroy, F., Boisgon1ier, J., Laidi, C., .\fo11gin, J.-F., Delorrne, R. , ... Houenou, J. (2017). Social cognition in amisrn ls as.sociated with the

ncu.-odcv,:.;lopmcnt of the post.criorsuperiortempor:11 ~u:cu:-.. A<:rn Psyr.hiazri<::7

Scandinavica, /36(5), 517- 525. doi: IO l l l l/aq,s.12814 PMID 28940401

c7

Electroet1cophalogram (EEG) for Dcllneating Objective Mc;:,sure of Autism Spec1rum Df.:,;ordcr

lbaiiez, L. Y., Grar.Lz, C. J., & Mes,inger, D. S. (2013). The development of referenLial commuruca1lon :md autism symp1onia1ology :n high-risk infan ts. Jnfoncy, 18(5), 687-707. doi:i0.l I l l/j . l 532-7078.2012.00 l42.x PM'l D:244038M

Jacoh, R. J. K., & Karn, K. S. (20m). Eye Tr:K·kin[: in Hunian-Co:nputer lnteraciion and Usability Research: Ready to Deliver the Promises. ln Tile Mind's Eye: Cognitive and Appbce Aspects o! Fye M overnl':11L ll esearch. Els~vier Science.

Jairne, M., lvkmalion, C. :-.-r., Davidson, B. C. , Newell, L. C., .'v1undy, P. C., & And Henderson, II. A. (2016). Brief report: Reduced 1e11:pmal-central EEG alpha coherence during joim auer:tion percep1io11 in aclole.sce.nts wi11, _quti !) tn spectrmn disorder . .loumal ofAuti.wr: and Dev,lopmenta/ Di.wmlers, 41'i(4), 1477-1489. dci: I0 . , 00710803-015-2667-3 PM ID:266~9R 13

Jiang, M., & Zhao, Q.(20 17). Learning visual attention to identify people with autism spectrum disorder. P,vceedin.g~ ~fthe IEEE International Conference rm Computer Vision, 3267-:1276. 10.1109/LCCV.2017.354

Jones, T. B., Banrlettini, P. A., Kenworthy, L., Case, L. K., ~fillcville, S. C., '.\.lanin, A., & And Birn. R. M. (201 0). Soarccs of group dtl'ferences in functional cor.11ectivity: An invc:.:aig;H.ion applied to autism spectrum disorder. Neurolmag1:,49( I), 401-414. d1Ji: I 0.101 611.neuroimnge.2009.m.0.5 l PMJJ): 19646533

Just., \.1. A.,& Carpe11tcr, P. A. ( 1976). l'.yefixatio:1s arnlCogni,ive Proccsscs. Coi:rutive Psydinlogv, 8(4), '14 l 480 .

.lnsr:, M.A., Cherkassk.y, V. L., Keller, T. A., Kana, R. K. & Minshew, N. J. (20()6). Functional ilnd :m;itomical cortic:~I undercon:.cctivil)' in ,niiism: Evidence. l'rom an FMRI siu<ly of an e,cc11t.ivc fu nction task ,l~d corpus rnllosum morphorneLry. Cerebnd Cnr1ex, 17(4), 951 .. ,96 1. cloi: I 0.i 091i ccrcor/bJ-,]006 PMID:1677231 3

.lust, M . /\., Chtrkllssky, Y. L., Keller, T. A., & Minshew, N. J. (2004). CmticHl aclivatiou ;md synchroni7arion d~1rlng senlence comprehension in high-fun~tioning autism: ~vidence of tmrlerc:onnec;ivi1.y. B1·l4·in, 127(8), l8 l 1- 1821 . cloi: J0. I093/brair./~whl 99 PMlD: I 521 5213

58

[5/ectroenr.;.epha,ogram (EEG) fer Celi11eating Objective Me;1s11ro of Autism SpeclrtJm Disorder

Kana. R K., Keller, T. A., Chcrknssky, V. L., Minshew, N. J., & Just, M.A. (2009). Atypical l'rn111al-puster•or synchronization of •heory of mind regions in autism during menral state anributior:. Soda/Neuroscience, 4(2), 135- J 52. cl(>i: 10 JOS0/171709J080L1985JO PMID:J8633B29

Kana, R. K .• Uddin, L Q., Ktilct, T., Chugani, D., & 'I-Hiller, R.-A (2014). Bra,n c.:.:.nrnectivlty in .:tl!L1sm. Vrontiers tl'i Humm, .Neuroscienctt, 8, 349. dni: 10.3:l89/fnhum.2014.00349 PMID:2,;917800

Keehn, 13.,Miiller. R.--1\., & Townse~d,J. (20 l 3). Atypical attentional networks a11-:J ~hcc.mcrgence of amism. Neurosciencttlind Hiohr!haviorriiRe.vim•}s,37(2)> 164-181. doi.: 10. JOI 6ij.ncuhimev.2012.l l.014 PMlD:23206665

Kern1edy. D. f' .• & Adolph,. R. (2012). The social brain in psychiatric ancl 11eurologica1 disorder<;. Trends i.•1 Co,.Qn.i.'iw: S'cienct!S, 16(1 n, 559-572 doi: 10. !())6/j.r:cs.2012()9 006 PMTD:23047070

Klcinh;;ns, N. M ., Richards, T.. Sterling, L., Stcgbauer, K. C., Mahurin, R .• Johnson, L. C., .. Aylward, E. (2008}. Abno•·mal tunc:.:ona' co1111ecLiv1ty in iJUlism spectrum disordersduri11g fat:~prnces:.;lng. Brain, 131(4), 1000-1 O l 2. doi: lO. ! 09:,ibrnin/;awm334 PMID: 18234695

Klem, G. 11., Uidcrs, 11. 0., Jasper, H., & Elger, C. ( l 999). The !C1Hwemy electrode system of the. i nten~ational focleratioll. Ele,:rro~ncephc1lographywul Clinical Neurophysiology, 52(3), 3-6. PMID: I 0'.590970

Klin, A., Lin, D. I., Gorrindo, P., Ransay, G., & Jones, W. (2009). Two-ycar­olds v-.iith amjsm oriem lo non-social cont.ing~ncH:::.s nHher t.lian biological mot:on. N<1111rc, 459(7244), 25i.

Komogomcv, 0., Holland, C., Jayarathna, S ., & Karpov, A.(2013). 2d Lnear 1;x·ulomot.orplan1 mathemat;cal model: Verifica:ion and biom~r:ic applica1io11-.. ACM Tran.sac/ion, on ;\pplied Perceprion (JAP), 10(4), 27.

Koshii,o, H.,Carpenre,·, P. A. ,Minshew. N.J.,Cherbssky. V. L., Keller, T. /\., &Just, M.A. (2005). Functior,al connectivity in an f"vlRJ working memory 1a,k in high-functioning autisn,. Ne um Image, 24(3).810-321 . cloi: lO. JO 16/j. ,,eJroinrage.2004.09.028 PM!D; 156523 [6

Koshino, 1-1., Kana, R. K., Kclkr, T. A., C!1crkassky, V. L., Minshew, N. J., & Just, NL A. (2007). F~1RJ ~i:vestigaLion of \VOrk.iug memory fur fHces i11 at:::sm: Visual cod:ng and underconnecn.viry with frontal areas. Cerebral Corra, 18(2), 289--300. doi:10.l093/ccrcor/bhm054 PMID 175,.7680

5B

Elec:roencepnJIO!]ram (EEG.I for DelineaCing Objective Measure of Au/ism Spectrum Oi$Ordcr

Krejtz, K., Duchowski. A., Szmidt, T., KrcJ lz, I., Penlii, F. G., Pires, A., Vilaro, A .. & Villalobos, N. (2015). Gaze transition entropy. T•·,msaclio11.1· on AppNea' Perception, .13( 1), 4:1-4~20.

K:·ejtz, K., Duchowski , A. T., Krejtz, I.. Szarkowskc .. A ,, & Kopacz. A.{2016). Discerning J\mbi~.nt/Focal Atte111'io1~ with Cocff1ci~nt K. ·n-ansac:r.ions rm Applied Pen:ep1ion, J 3, 3.

Kre_jtz, K,, Dud;ow,ki, A. T .. Nicdzicl,ka, A., Biele, C., & Krejtz, I. (2018a). Ey~ tracking cognitive. load using pupil diameter a11cl 1Hicru sJi..'.Clde~ v.,itli tixeJ gaze. PloS One, I 3(9).

Krcjll, K., Dnchow.~ki. A. T., Nicdzielska, A., Biele. C .. & Krcjl1,, I. (2018b). Eye tracking cognitivt; lnad using pupil di:rn1e.1er and r1ic;o sac<·~1des with fixed gaze. 1-'lnS One, I 3(9), l-23.

Lazarcv, V. V., Pontes, A., Mitrofanov, A. A , & deAzevedo, L. C. (20i5). Reduced i11terhemisphcric connectivity in childloocd aurism ue'.ecied by elecrroenecpha.lograpti ic photic drlvi11g coherence. Journal t4' /lu1ism and /Je,,elvpmemaf rnsorders, 45(2), 53:-547. doi: 10. 1007!0803-01 3-1959-8 !'~111):24()97142

1.ee. J. C, & Tan, D. S. ,200G). Usillg a low cost electroe11ccphalDgraph for t·ask classification iH 1 fCJ research. h; Procl.~edin.gs of 1hf! l9rh animal

ACtW sympo.1·ium rm User inrerface s0Jh11ar<' and r.echno!ogy. AC.l'vI. IO l l45/l l66253. I l 66268

l .seigh, J.P., & Du, J. (2015). Brief re~ort: Forecasting tloe econon°ic buoden of uutCsm in 2015 and 2025 ir: lhc u~ited states. Jounw l ri Au;,:sm and De,w!opmenrai Disordns,45( 12), 4135-4, 39. doi: l0.J0071080:'l-Ol ~-2521-7 PMJD:26183723

Lc11artowicz, A. , & Loo, S. K. (2014). lJse or E FG 10 diagnus~ ADHD. C,arem P.,ychiatry RepmlS, 16( 11). 498.

Lin . .1. (1991). Divergence measures hased on the Shannon entropy. IEEE "honsm.:livns rm b1forn;a.rfon. Theory, 37(1), 145-15 l . den: 10.1109/ l 8.6I1 15

L u. W., Yu. X., Ra.J, R. Yi, L., Zm,. X., & ;\nd Li, M. (20 15). FJficielll :-ullisrn :.;pecrrum disorder preciic.tioil with eye move,nent: A nu:c..:hi.ne leurning framework. In ,\.fj't-:cri','e Computing and fo:elli1:cni fmeractim1 (AC!/), 2015 lmemwio,;a/ Co11jerence 011 .. I EFE. 10.J 109/ACll.201 5.7344638

60

EJec1rocmcepha/ogram (EEG} for D~finu8'ting Obje~tive Mcclsute of Autism Spectrum Disotdcr

Lombardo, \-1. V., Ch:ikrah,1rli, R , Bnllmorc, E. T., Baron-Cohen, S .. & Consorti•.1m, M.A. (201 l ). SpecializatioH of right remporo-parietal j1mctior

for merna)i1.ing ,ind irs relation to social impairments in autism. N"eurolmage, 560\ 1832-1838. doi: I 0., Ol 6/j.:1eurnirnage.2011.02.0(17 PMID:213563 16

L;)mhardo, .\·1. V., Chakrabani, B., Bullmore, E. T, \'l'heelwrigl11., S. J., Sade.k, S A .. Suckling, J., ... llaron-Cohen, S. (2010). Shired neural circuit,;

for mcntalizing ti bout lbe sdf and orhers. Joi-inwl. tfCog1iitiw: Nei4ro.w:·i,mu~, 22(7), i /i2J-l 63.5. doi: 1.0. l 1 (i2/jOCll. 2009.21287 PMID: l 9580380

Lynch. C. l, Cddin, L. Q., Supckar, I<., Khou mm, A., Phillip,, J ., & Menon,

V. (2013 ). Default mode network in ct, iidhood a utism: Po.,teromcdial t:<Jrte,

he:·erogi:11eity and relalion~hip wilh social deficits. BioloJ{tcal P_;;ychialry, N(3), 212 2 I 9. doi; J0.1016/j.biopsych.201'2.12.013 PMlD:23375976

Mason , R. A., Wi\lirn1s, D. L. Kane, R. K., Minsl1ew, N., & lust, M. A.

(2008). Theory of mind disruption and recruitrr.cnl or lhe right hemisphere during r:J:rrative :..:omprd1en~ion in aut.ism.NeunJpsychologin: 46( I), 269- 280. doi; 10. 1{)!6/j.neuropsycholog:a.2007.07.018 PMTD: l78ri9314

Mo,rnmmaci-Rezazadeh, I., Frohlich, J., Loo, S. K., & Jcstc, S.S. (2016). Rrai11 cor"!ncctivity in autism s~clrurn disor:ler. Currem Opinirm in _Nf:urology, 29(2). 137.

Mdlen, T. R., Kothe, C. A., Ca,, Y. M,, Ojed~. A., Kerth, T., IW,keib, S., ,

Cauv..,.enbergb,, 1 G. (2015). Real-t.imcncumimaging and cognitive mon.lf_t)ring 11sing \\·\~;,Hable dry EEG. IEJ~E Tmn.1·ae.tton.1· on 8ionu-:dir:m' r.:n.f.!,ineering, r,2('. I), 2553-2567. doi:10.llil9ffBME.2015.248l482 PMID 26415149

Mendy, P., Block, J., De'gado, C., Pomares, Y., Varc Heck~, A. V.. & !'arladc, M. V. (2()()/). Individual d ifkrcnccs and the dcvc io,,rncn!. ofjoint al'.~11\ion in i11iirn:y. Child Development. 78(3), 'l38-954. doi: IO. l .111/j . 1467-·

8624 .2007.01042.x PMID:J 75 i 7014

Mr,ndy, P., & Newell, L. (2007). AU~ntion, .ioint attention, an,I social t:Og11itlo;1. Curren/ Direc!ion..-.. in Psychological .\'dence , ./6(5), 269-274. cloi:10.J J 1 Uj.l467-1!72l.2007.00518.x PMID:19343102

Mundy, P., Sigrn,n;, M., & Kasari, C. (1990j. A lor:gitudinal s•.udy ofjoir1l

arrention an<l language developmenr in autistic children_ Journal of Autism. and Developmental Disurder.,, 20(1 ). 115- 128. doi:10.1007/BF0220686! PMiD:2324051

61

F.lec:!ro.:.ncephillogram (EEG) for Delintatlng Objective Measure of A11lisrn Spectr11m Oisorder

Mundy, P.,Sigman, M., Ungcrer,J., & Shermar ,T. ( 1981:). Defii:ingthe sori,d defidts of all! ism: The co11trihntion of non•verbc:.I communication measure5. Jounuii r>j' Ch.ild Psyr..:hol.tJgy and Psychicirry, nm/ Alli<id Disr:ipiinrs, 27(5), 6~7-669. <loi: l O.ll I l/j.1469-76 10 . 1986.rb00,90.x PMID:377 1682

Mundy, P., Solh-an, L .• & Mastergeorge. A. M. (2009). A parallel and disrrilmkd~processing 1nodi::I ufjoint atu:ntion. socia.1 cognition and autism. /\11ti ,m Re~earch, 2(!), I, 2-21. doi: 10.)002/,rnr.61 Ptvl1 D:i93.'i8304

1'.-fundy. P. C.. (2016). Aurism (J.Juljoint at1en1ion: Deveiopmcn1, neuroscit'!ICe, (111.d c/fri icalfundanum,ials. Guilford Publications.

l\oonan, S. K., Hai~t. F.. & Moller, R.-A. (2009). J\berranc functional connectivity in auti~m; Evii.:ence from kw.: fre,quency bold signal fluctuations. llr'1 in. Re.1ecm :·h, 1262, 48- 61. doi: 10 . 1016/j .brainrcs.2008. 12 076 PMtD:1940 1185

Norbury,C. r'., Brock,J., <:ragg, L.. Einav, S., Gr,ffitl,s, Fl., & Nation, K. (2009). Eyc-rnove.ment patterns a~c assoi;jat~d wHh comml111icMivc competence in autistic s~ltUrum disorders. Journui af<.hikl Psychol.ogyt1nd P.,ylhiatry, and A!!ied Disciplir.es. 50(7), 834-842. rloi: 10.11 l 113.1469-76 I 0.2009.02073.x PM fl): 19298477

Nunez, P. I .. , & Sriniv3san, R. (2006). A theor~ti~al hasis for standi:1g and iraveling. llmir,: ,,..·aves measured 'A'ilh hnn?au .EEG wilh implica1ions fur an inLcRrated consciousness. Clini,:at Neurophysiolo8Y, 117( 11 ), 2424-2435. rloi: JO. I 0l6~j.clinp:1.2006.06.754 PMlD: (6996301

Redcay, E., K!cincr,M.,& Sa,e, R.{2012). Look at rh,s: TheneuralcMrclatcs of injtiaf.i11g ancl responding to bids for jojm :-me11;.]or1. Frontiers in Human. Neuroscience, 6, J69. doi: I 0.3389/fnhur.i.20l2.00 : 69 PMID:22737112

Rudie, J. D., She~zad, Z .. lle,nandc,, l... M ., Colich, N. I .. , Bookhe,mc:. S. Y., lacoboni, !vl., & Oaprecto, M. (20 11 ). Reduced func1 ional ii:te.g~Ht 1011 and segregalion nf ciist:iln:ted ;icur1-1.I 6ysu~n1s ur.dcdyiug social and cmotionnl information pr<:x:.es."ing ia autism "-(leCtrnm disorder~. Ccn::_bral Corf ex, 22(5), l025-1037. doi: I 0.1093/ccrc11r/bhrl 71 PMTD:21784971

Sas,on, N. J., Pinkham, ,\. E., Weittenhi ller, L. P. , Faso, D . .I., & Simpson, C. (20 l6). Context Effect, on Facial A fleet Re.:ognition i ,i Schizophrenia ;.met Autism: Bchavi(lnt! and Eye-Tracklng Evidence. Schi'l.ophrenia Bull<~rin. 42(3), 675--681. doi:10 l093chbulbv17(o PMID:26645375

62

EJectroencephalogr:rm (EEG) ror Defineating Ob}euive Me.asuu; of A1Jtrsm Spectrum Di,sorrJcr

Sawye:· . .-\. C. P., Willbmsoc, P., & Youi:g, R.L(2012). Can G~7.e Avolflance Explain \Vhy Individuals \Vith ,\sperger'sSyndrorneCan't Recognize Emolions from Facial Lx.pressions? Journal afAuri.\H,! and Developmental Disorders, ·12(4). 606-6,8.

Saxe, R. (2006). Uniquely human social cognition. Current Opinfrm in Neurobiology: /6(2)1 235-239. ctoi: lO. fOl 6/j.~onb.2006.03.001 PMID: 16546372

Sd1ilbach, L., Wilms, M .. E~i<hoff, S. B., Romanzetti. S., Tepest, R., Bente, G., ... Vogeiey, K. (201 Ot Miuds made for sharing: Initiatingjoim attemtion recruits reward-relateJ neuroc.ircu~try . .lounwi ,i CugniJ.iF<~ N<!1.,mscfon.ce, 22( 12), 2702-2715 doi:10. l }62/iocn.2009.2140l PMJD: l992976I

Sclu:r~. M., Rs<ll!~, J., A•<:ii,torn, M., Richlar •. F., & Perncr, J. (2014 ). f:·Hctionating theory of rni:1d: A mcta-analysjs of functjonal br.ain l111aging studie~. Neuroscience ond /Jiob(!ht1Pioro} Reviews, 42, 9-34. doi: IO. l016/_j ncubiorev.2014.01.009 PMID:24486722

Shih, P., Shen, M .. Otrl, R. , Keehn, 13., Garfrey, M. S., & Mlilier, R.-J\. (.2010). Alypical network cunnecti\'hy for imitation in autism ,pec:rum disorder. Neuropsychm'ogiu, 48(10). 29:l i-29:19. r:oi: 10.1016/j. nc11rnp.1ychologia.201 o.osms PM ID:20558187

Snydcr,S. lv1.. Rugino~ T. A. 1 Hornig1 M .. &Stc1n, M.A. (20 i 5). Tnte:grnt.!onof aIJ EEG hinmarkc;- with a cli nlcian's ADJ fl) evallutiun. /Jn:iin and Beha·✓-ior,

5(4), 3-30. doi :JO 1002/brbJ.330 PMJD:25798338

Sparks, B., Friedman. S .. Shaw, D., Aylwai-d, E., Echclard, D .. Artrn. A., .. Dage r. S. R. (2002). Brni;~ .;trucnirnl alwormc1li:ies in young children with ,wt.ism spectnw1 disorder. Ne1.1mlogy, 59(2), 184-192. dui:LO.l212/ WNL.59.2. I 84 PMID: 12 I 16055

Spence, S. J., Sharifi, P., & Wimitzer, M. (2004). Auti,•n spectrum di,mller: scre-en~ng. diagnosis, and medical evakation. In Seminars in Pediatric Ner.rology (Vol. l l , pp. 186--195). Elsevier. del i: 10.1 Ol 6ii.,pcn.2004.07.002

Swa1hOi1, M. R. , Serl 111, G. C., & Siller, M.(2013). Bruarl Alltism P!ieuotype iu Typica lly Devduping Children Predi..'.tS Pt'.:-form a nce on an Eye-Trac.:klng Meas;1: eof joint At.te111·io11.Joanw/.of.Aui'.ism wu/ Developmental Disorders,

43(3), 707- 718.

63

E/ec(ro11ncephalogrom (EEG) for Deline.;,ting Dbjec JiYe Measure of Au!i$m Spectrum o;sorder

Symconidon, l ., D',1111,ir.the:I, I., Chow, W. Y., & Br~hcny, R. (201 6). Development of cmlinc ttse of tllem) 1 of 111ind darin~ adole-.;cenci..!: An eyr.:. · tracking study. Journal of Erperime!oW! Chi id P.-.ychoiogy. 149, 8 i- 97. doi: 10. 10 I6/j.jecp.1015.l J.007 PMID:2672347 l

Thapaliya. S., .layara thna, .S., &. Jaime, M. (2018). e valuu1i11g ih,: F.[G and eye moi;c:mell! \"f<>r c.mt.ism spr.<.·trwn disorder. At:adc1uic Press.

Tho rup, E., 'lys1.riim, P., G redthiick, G., B,~h~, S., & Fakk-Yttcr, T . (201 6) . .-\ hered gaze following during live interaction in illfamrs at risk for autism: an eye tracking ~wciy, t1rlolecular/w iism, 7(1 ), i 2.

T horup, E .. Nywom, P., Grcdeback. G, Bollc, S., & Fakk.Yuer, T. (2018). Reduced Alternming Gau During Social i,ueranion in Infancy is Associa1ed " 'ith elevated Symµt.oms ur Autism in TodJkrho,xt. Journal ,fA/innrmal Child Psyciw/ogy. 4fi(7), 1547-1561.

Tsang, V. (201 8). Eye-trac~ing sludy 01.1 fncial emorion rec.ogni1jo:1 t:tsb i1.1. Jndividllals wilh high··hua.: tioning autism "ipectnnn disorders. Auri.c.m, 22(2). !61-170. doi:l0.l l 77/J 3623613l 6667810 PMID:29490486

lJddin. L. Q .• Supekar, K., 1.yuch, C.J., Khou,am, A .. Phiilips,J .. f ei11slein,C., .. . Mellon. V. (2013). Salience network- based classification and predic:ion of symp101.a severity i~ children with a,nism.JAMA Psychiall')', 70(8). 869-879. doi: 10. 1 OOJ/Jamapsychiat.ry.2013. !04 PMID:23803(,51

Vivau1i. G , Fanning, P.A. J., Hocking. D.R., Siever,, S .. & Dissil.n&yakc. C. (201 7). Social Atleution, Joint A:reution and Sus1aii,d A11.en1ion in Autism Spectrum Disorde.r ar1J ~!j}!jams .S yrnlrome: Con,·ergtnces and Divergcr.ces . . lournol <~f.4ulism amt Dev<~lopmentol Disonierr, 47{6), J 866-- ~ .~77.

Wagner, J.B., Luyslcr, R. J., Tagcr-Flusberg, H., & Nelson, C. /\ . (2016). Greater Pupil Size in Response 10 E1.1101ional Face~ :1s an Early Marker of Social-Communic,,tive Difflcuhies in fn fant;,; at High Risk for Autisrn. h~/fmc~,. 2/(5), 560-.581. doi: ,0 . ll l l l i~til.121 28 PMID:2761 6~38

Wang, J., Bars1ei11, .I ,, Etbridg~, I .. E ., Mosconi, M . W., Tabrne, Y.. & Sweeney, J. A.(201 3). R-:.:sting ~late. EEG ,1bnonr:alifies ln :.:utism speclrum disun.lers. Journal nfNeurodci•dopmr.ntal Disorders. 5( 1). 24.

64

E.'ectroencephalogram (EEG) for Delineating Obj~ctive Me.1sure of A1Jtism Specirum DisorrJer

Welchew, D. E .. Ashwin. C., Bcrkm1k. K., Salvador, R., Suckling, J., naron­Coi'ien, S., & Bullmore, E. (2005). Ft1r.chrnal Dlsconnectivi:y of the mediHI te-ni~iral !oh.:: ia :'\sperg:er's synclrome.Bioiogic!d r~iychiatry,57(9), 99 !-998. doi: l 0. l 0 I 6-'.j.biopsych.2005 01.028 PMlD: 15860'39

V.-'i~ckowski, .'\. T., & Whi'.e, S. W. (2017). Eye--Gaze Ai,alysis of Pacial En1n:,on Rccognilinn and Expression in Adolescents \Vilh ASD. Jounw! (y'° Clh•ical Child and Adolesce111 f'c.1dw!ogy, 46(; ). 110-124. doi: 10.1080/15 374416.20 I 6. 12(),-1924 l'M1D:276SL'.i'l0

Yoder, l'., Stone, W. L, W,1,dcn, 'J'., & Maba, E. (2009). Predicting social impairment and ASD Oic.gnosis in younger siblings of children with audsm spectn.1rn di~nnkr. .lounwl ofA.utism and /Je1.,efopmenwl Diso;Y.i.~,:~, ]9l 10), I 381-1391. doi: 10.1007 ;0803-009-0753-0 PylfD: l 9449096

ENDNOTE

D~tails regard;.11g performing ICA in EEGLABcaE b~ fot;nd her~: Swart7, Ccn1er J·or Compurarion,1} Neuroscience (2018, September 19). Chapte1 09: Decompming Data U,ing JCA. EEGLAB \Viki. t::rps:Jlsl'c1uicsd. cdu/wikiiClrnpter_09:_Dc,,0n:posii:g_Data _Using .JC;\

65