This is a sel -a chi ed e sion o an o iginal a icle. This e sion
may di e om he o iginal in pagina ion and ypog aphic de ails.
Au ho (s):
Ti le:
Yea :
Ve sion:
Copy igh :
Righ s:
Righ s u l:
Please ci e he o iginal e sion:
CC BY-NC-ND 4.0
h ps://c ea i ecommons.o g/licenses/by-nc-nd/4.0/
Func ional connec i i y o majo dep ession diso de using ongoing EEG du ing music
pe cep ion
© 2020 Else ie
Accep ed e sion (Final d a )
Liu, Wenya; Zhang, Chi; Wang, Xiaoyu; Xu, Jing; Chang, Yi; Ris aniemi, Tapani;
Cong, Fengyu
Liu, W., Zhang, C., Wang, X., Xu, J., Chang, Y., Ris aniemi, T., & Cong, F. (2020). Func ional
connec i i y o majo dep ession diso de using ongoing EEG du ing music pe cep ion. Clinical
Neu ophysiology, 131(10), 2413-2422. h ps://doi.o g/10.1016/j.clinph.2020.06.031
2020
Jou nal P e-p oo s
Func ional connec i i y o majo dep ession diso de using ongoing EEG du -
ing music pe cep ion
Wenya Liu, Chi Zhang, Xiaoyu Wang, Jing Xu, Yi Chang, Tapani Ris aniemi,
Fengyu Cong
PII: S1388-2457(20)30416-8
DOI: h ps://doi.o g/10.1016/j.clinph.2020.06.031
Re e ence: CLINPH 2009314
To appea in: Clinical Neu ophysiology
Recei ed Da e: 25 No embe 2019
Re ised Da e: 7 May 2020
Accep ed Da e: 29 June 2020
Please ci e his a icle as: Liu, W., Zhang, C., Wang, X., Xu, J., Chang, Y., Ris aniemi, T., Cong, F., Func ional
connec i i y o majo dep ession diso de using ongoing EEG du ing music pe cep ion, Clinical Neu ophysiology
(2020), doi: h ps://doi.o g/10.1016/j.clinph.2020.06.031
This is a PDF ile o an a icle ha has unde gone enhancemen s a e accep ance, such as he addi ion o a co e
page and me ada a, and o ma ing o eadabili y, bu i is no ye he de ini i e e sion o eco d. This e sion
will unde go addi ional copyedi ing, ypese ing and e iew be o e i is published in i s inal o m, bu we a e
p o iding his e sion o gi e ea ly isibili y o he a icle. Please no e ha , du ing he p oduc ion p ocess, e o s
may be disco e ed which could a ec he con en , and all legal disclaime s ha apply o he jou nal pe ain.
© 2020 Published by Else ie B.V. on behal o In e na ional Fede a ion o Clinical Neu ophysiology.
Func ional connec i i y o majo dep ession diso de using
ongoing EEG du ing music pe cep ion
Wenya Liua,b, Chi Zhanga, Xiaoyu Wanga, Jing Xuc, Yi Changc, Tapani Ris aniemib, Fengyu Conga,b,d,e
aSchool o Biomedical Enginee ing, Facul y o Elec onic and Elec ical Enginee ing, Dalian
Uni e si y o Technology, 116024, Dalian, China
bFacul y o In o ma ion Technology, Uni e si y o Jy äskylä, 40014, Jy äskylä, Finland
cDepa men o Neu ology and Psychia y, Fi s A ilia ed Hospi al, Dalian Medical Uni e si y,
116011, Dalian, China
dSchool o A i icial In elligence, Facul y o Elec onic In o ma ion and Elec ical Enginee ing,
Dalian Uni e si y o Technology, 116024, Dalian, China
eKey Labo a o y o In eg a ed Ci cui and Biomedical Elec onic Sys em, Liaoning P o ince. Dalian
Uni e si y o Technology, 116024, Dalian, China
Co esponding au ho :
1. Fengyu Cong, School o Biomedical Enginee ing, Facul y o Elec onic and Elec ical Enginee ing,
Dalian Uni e si y o Technology, 116024, Dalian, China
Email: [email p o ec ed]
2. Jing Xu, Depa men o Neu ology and Psychia y, Fi s A ilia ed Hospi al, Dalian Medical
Uni e si y, 116011, Dalian, China
Email: [email p o ec ed]
3. Yi Chang, Depa men o Neu ology and Psychia y, Fi s A ilia ed Hospi al, Dalian Medical
Uni e si y, 116011, Dalian, China
Email: [email p o ec ed]
Highligh s:
Majo dep ession causes al e ed connec i i y in del a and be a bands du ing music pe cep ion.
Be a band connec i i y is a p omising bioma ke o he diagnosis o majo dep ession diso de .
Na u alis ic music s imuli lead o equency-speci ic unc ional connec i i y.
Abs ac
Objec i e: The unc ional connec i i y (FC) o majo dep ession diso de (MDD) has no been well
s udied unde na u alis ic and con inuous s imuli condi ions. In his s udy, we in es iga ed he
equency-speci ic FC o MDD pa ien s exposed o condi ions o music pe cep ion using ongoing
elec oencephalog am (EEG).
Me hods: Fi s , we applied phase lag index (PLI) me hod o calcula e he connec i i y ma ices and
g aph heo y-based me hods o measu e he opology o b ain ne wo ks ac oss di e en equency bands.
Then, classi ica ion me hods we e adop ed o iden i y he mos disc imina e equency band o he
diagnosis o MDD.
Resul s: Du ing music pe cep ion, MDD pa ien s exhibi ed a dec eased connec i i y pa e n in he del a
band bu an inc eased connec i i y pa e n in he be a band. Heal hy people showed a le hemisphe e-
dominan phenomenon, bu MDD pa ien s did no show such a la e alized e ec . Suppo ec o
machine (SVM) achie ed he bes classi ica ion pe o mance in he be a equency band wi h an
accu acy o 89.7%, sensi i i y o 89.4% and speci ici y o 89.9%.
Conclusions: MDD pa ien s exhibi ed an al e ed FC in del a and be a bands, and he be a band showed
a supe io i y in he diagnosis o MDD.
Signi icance: Ou s udy p o ided a p omising e e ence o he diagnosis o MDD, and e ealed a new
pe spec i e o unde s anding he opology o MDD b ain ne wo ks du ing music pe cep ion.
Keywo ds: unc ional connec i i y, ongoing EEG, majo dep ession diso de , music pe cep ion,
na u alis ic s imuli.
1. In oduc ion
Majo dep ession diso de (MDD) is cu en ly one o he mos p e alen psychia ic diso de s, and i
subs an ially dis up s pa ien s’ li es. MDD pa ien s a e usually cha ac e ized by de ici s o a ec i e
and cogni i e unc ions (Kaise e al. 2015; Li e al. 2018; Xia e al. 2018). Al hough many esea che s
ha e dedica ed hemsel es o he explo a ion o he pa hophysiology o MDD, he neu al mechanisms
o i s e iology and pa hogenesis a e s ill no ully unde s ood. Cu en ly, he e a e no bioma ke s o he
clinical diagnosis o MDD (Fingelku s and Fingelku s 2015; Gao e al. 2018; Nugen e al. 2019).
Con en ionally, he clinical diagnosis o MDD equen ly depends on some public c i e ia, such as
Diagnos ic and S a is ical Manual o Men al Diso de s V (DSM-5), which makes he diagnosis o MDD
e y subjec i e due o human ac o s and causes aul y diagnos ic esul s (Mum az e al. 2015; Nugen
e al. 2019). Fo his eason, nonin asi e neu oimaging echniques, such as elec oencephalog am
(EEG), magne oencephalog aphy (MEG) and unc ional magne ic esonance imaging ( MRI), a e
u gen ly needed as mo e e ec i e and in elligen diagnos ic ools. EEG is an inexpensi e echnique ha
bene i s om high empo al esolu ion. EEG is able o eco d elec ical ac i i y a equencies ela ed
o neu onal ac i i y and o cap u e he dynamic changes a a millisecond scale. These ad an ages make
EEG a e y p omising echnique o commonly use in he diagnosis o MDD (Baska an e al. 2012;
Mum az e al. 2015, 2017).
Many MRI s udies ha e demons a ed ha he pa hogenesis o MDD is he abno mali y o la ge-scale
b ain ne wo ks, such as de aul mode ne wo k (DMN) (Zhu e al. 2012; Wu e al. 2013) and a ec i e
ne wo k (AN) (A e y e al. 2014), o he dysconnec i i y o some b ain egions, such as co icolimbic
pa hways (Nugen e al. 2019), a he han he dys unc ion o an indi idual b ain egion. So, unc ional
connec i i y (FC) has p o en o be e ec i e o in es iga e ne wo k dys unc ion in MDD. FC p o ides
a new line o hough o he diagnosis o MDD pa ien s, and many s udies, especially MRI and EEG
s udies, ha e ocused on he classi ica ion o MDD based on FC analysis (Wang e al. 2017; Gao e al.
2018; Sakai and Yamada 2019). Howe e , FC analysis and MDD classi ica ion always ocus on es ing-
s a e o highly con olled and epea ed s imuli, bu he di e ences in FC unde na u alis ic and
con inuous s imuli be ween heal hy people and MDD pa ien s ha e no been well s udied. Compa ed
wi h es ing s a e, lis ening o con inuous music is mo e closely ela ed o eal-wo ld expe ience (Wang
e al. 2020), and emo ional a ousal can be induced o a ec i e p ocessing (Miku a e al. 2012). Music
he apy has become an a ac i e ool o MDD ea men , so unde s anding he mechanism o he b ain
esponse du ing lis ening o music is he basis o he diagnosis and ea men o MDD (Michael e al.
2005; Ma a os e al. 2008). An inc easing amoun o li e a u e has demons a ed ha human b ain
ne wo ks a e di e en ac oss equency bands in bo h es ing-s a e and ask condi ions, and ne wo ks
in speci ic equency bands may e eal di e en b ain unc ions (B ookes e al. 2012, 2016; Hilleb and
e al. 2012, 2016). P e ious s udies ha e demons a ed al e ed FC in MDD in di e en equency bands,
so FC analysis ac oss di e en equency bands is impo an o he diagnosis o MDD (Mum az e al.
2015; Kno e al. 2001; Whi on e al. 2018). Some s udies ha e ound ha equency-speci ic and
la ge-scale b ain ne wo ks will eme ge du ing music pe cep ion o sus ain ongoing cogni i e asks
(Allu i e al. 2012; Cong e al. 2013; Wang e al. 2020). A e iew by Ma a os e al. emphasized ha
music he apy was associa ed wi h imp o emen s in mood o ea dep ession (Ma a os e al. 2008).
Some esea che s ha e al eady ocused on equency-speci ic b ain esponses o music in dep ession
pa ien s and o he psychia ic diso de s and ha e ound ha music he apy can al e FC and modula e
b ain esponses (Michael e al. 2005; Rami ez e al. 2015; Dha madhika i e al. 2018). These p e ious
s udies suppo ou assump ion ha al e ed FC exis s in di e en equency bands du ing music
pe cep ion in MDD pa ien s. Howe e , ew s udies ha e in es iga ed he mechanism o dysconnec i i y
and b ain esponses o MDD pa ien s du ing music pe cep ion.
Fo elec ophysiological neu oimaging echniques, like EEG, he collec ed signals om one scalp
senso a e ac ually om he whole b ain due o he olume conduc ion e ec (Van Den B oek e al.

1998; Scho elen and G oss 2009; B unne e al. 2016). B ain connec i i y in senso space is usually
con ounded by olume conduc ion, and e en wi h he conduc ion o sou ce econs uc ion me hods,
sou ce leakage s ill exis s due o he ill-posed na u e o he in e se p oblem (O’Neill e al. 2018). An
inc easing numbe o s udies ha e demons a ed ha he communica ion o b ain egions o neu al
popula ions depends on phase in e ac ions (Womelsdo e al. 2007; Pal a and Pal a 2012; He e al.
2019). A ze o-lag in e ac ion is conside ed o be he consequence o olume conduc ion because signal
leakage is ins an aneous. Among he phase synch oniza ion me hods, phase lag index (PLI) disca ds
he in e ac ions esul ing om phase di e ences o ze o, so PLI is no sensi i e o he olume
conduc ion e ec ; hus, i is commonly used in he FC analysis o EEG and MEG s udies (S am e al.
2007; Vinck e al. 2011; Wu e al. 2012). Ruiz-Gómez e al ha e demons a ed ha PLI could educe
he bias in oduced by he spu ious in luence o olume conduc ion and was supe io o he o he se en
FC synch oniza ion measu es (Ruiz-Gómez e al. 2019).
Ne wo k analysis me hods based on g aph heo y a e widely used o e eal he opology o b ain
ne wo ks (Spo ns 2018; Ren e al. 2019). In EEG senso space, he b ain ne wo ks a e cons i u ed by
nodes ep esen ing elec odes and edges ep esen ing FC s eng h be ween e e y pai o nodes. The
a ious ne wo k p ope ies a e e icien measu es used o quan i y b ain unc ional in eg a ion and
unc ional seg ega ion (Rubino and Spo ns 2010; Liao e al. 2017). Deg ee, which is a measu e o
in luence, clus e ing coe icien , which is a measu e o unc ional seg ega ion, and cha ac e is ic pa h
leng h, which is a measu e o unc ional in eg a ion, a e ne wo k p ope ies ha a e commonly used o
quan i y he e iciency o in o ma ion p ocessing (Acha d e al. 2006; He e al. 2007; Gong and He
2015). In his s udy, we applied deg ee, clus e ing coe icien and cha ac e is ic pa h leng h o quan i y
he di e ences be ween heal hy people and MDD pa ien s.
In his s udy, we collec ed EEG da a om heal hy people and MDD pa ien s unde condi ions o music
pe cep ion, and used he PLI me hod o calcula e FC ac oss i e ypically analyzed equency bands:
del a, he a, alpha, be a and gamma bands. A e s a is ical analysis using he ne wo k-based-s a is ic
(NBS) me hod, we compa ed he wo g oups h ough connec i i y ma ices and g aph- heo y based
ne wo k p ope ies in del a and be a equency bands, which exhibi ed signi ican di e ences. Finally,
machine lea ning me hods we e used o pe o m he classi ica ion.
2. Me hods
2.1 Da a acquisi ion
Nine een heal hy adul s ( ou een emales and i e males) aged 24 o 65 yea s in he con ol (CON)
g oup and wen y adul s ( ou een emales and six males) wi h MDD aged 23 o 58 yea s in he MDD
g oup we e ec ui ed o his expe imen . All he pa ien s we e om he Fi s A ilia ed Hospi al o
Dalian Medical Uni e si y in China. This s udy was app o ed by he e hics commi ee o he hospi al,
and all he pa icipan s signed he in o med consen be o e hei en ollmen . None o he pa icipan s
epo ed hea ing loss o o mal aining in music. MDD pa ien s we e p ima ily diagnosed by a clinical
expe , and he cou se o he disease a ied om 2 and 36 mon hs. All he pa icipan s we e es ed
acco ding o Hamil on Ra ing Scale o Dep ession (HRSD), Hamil on Anxie y Ra ing Scale (HAMA)
and Mini-Men al S a e Examina ion (MMSE). The means and s anda d de ia ions (SD) o age, gende ,
educa ion and clinical measu es o bo h g oups a e lis ed in Table 1. Du ing he expe imen ,
pa icipan s we e old o si com o ably in a chai and lis en o a piece o music. An 8.5-minu e long
musical piece o mode n ango by As o Piazzolla was used as he s imulus due o i s ich musical
s uc u e and high ange o a ia ion in musical ea u es, such as dynamics, imb e, onali y and hy hm
(Allu i e al. 2012, 2013).
The EEG da a we e eco ded by he Neu oscan Quik-cap de ice wi h 64 elec odes a anged acco ding
o he in e na ional 10-20 sys em. Elec odes placed a he le and igh ea lobes we e used as he
e e ences. The da a we e down-sampled o 256 Hz o u he p ocessing and isually checked o
emo e ob ious a i ac s om head mo emen s. Eye mo emen s a i ac s we e ejec ed by independen
componen analysis (ICA), and 50-Hz a i ac s we e emo ed by sho ime Fou ie ans o m (STFT).
STFT was applied o il e he da a in o i e ypically analyzed equency bands, namely, he del a
(0.5-4 Hz), he a (4-8 Hz), alpha (8-13 Hz), be a (13-30 Hz) and gamma (30-80 Hz) bands, o u he
analysis.
2.2 Phase synch oniza ion
In his s udy, phase synch oniza ion was measu ed be ween all he pai s o channels by he PLI me hod,
which is an asymme y index ha measu es he dis ibu ion o phase di e ences (S am e al. 2007;
Vinck e al. 2011). Due o he ins an aneous sp ead o cu en , he same sou ces collec ed by wo
elec odes a e conside ed o cause a ze o-lag phase di e ence, which is ejec ed by PLI. The e o e, PLI
is less sensi i e o he olume conduc ion e ec , and i can e eal he ue coupling s eng h be ween
pai s o channels.
Fo an EEG signal om one channel, he analy ical signal can be cons uc ed
𝑥
(
𝑡
)
, 𝑡
=
1,2,3,⋯,𝑇
𝑧(𝑡)
by Hilbe ans o m,
𝑧
(
𝑡
)
=
𝑥
(
𝑡
)
+
𝑖
𝑥
(
𝑡
)
=
1
𝜋
𝑃𝑉
∫
∞
―
𝑥
𝑥
(
𝜏
)
𝑡
―
𝜏
𝑑𝜏, #
(
1
)
whe e is he imagina y pa , and e e s o he Cauchy p incipal alue. Then, he ins an aneous
𝑥
(
𝑡
)
𝑃𝑉
ampli ude and he ins an aneous phase can be compu ed as ollows:
𝐴(𝑡)
φ(
𝑡
)
{
𝐴
(
𝑡
)
=
[
𝑥
(
𝑡
)
]
2
+
[
𝑥
(
𝑡
)
]
2
𝜑
(
𝑡
)
=
𝑎𝑟𝑐𝑡𝑎𝑛
𝑋
(
𝑡
)
𝑥
(
𝑡
)
.
#
(
2
)
The e o e, he phase di e ence o wo signals and a ime can be o mula ed as:
∆𝜑(𝑡)
𝑥
𝑎
(𝑡)
𝑥
𝑏
(𝑡)
𝑡
∆𝜑
(
𝑡
)
=
𝜑
𝑎
(
𝑡
)
―
𝜑
𝑏
(
𝑡
)
. #
(
3
)
Then, he PLI index can be de ined ia
𝑃𝐿𝐼
=
|
<
𝑠𝑖𝑔𝑛[∆𝜑(𝑡)]
>
|, 𝑡
=
1…𝑇. #
(
4
)
The alue o PLI index a ies be ween 0 and 1. A alue o 0 indica es no coupling o coupling wi h a
phase di e ence cen e ed a ound 0 mod , and a alue o 1 indica es pe ec phase synch oniza ion
𝜋
be ween wo signals a a cons an lag excep 0 o .
𝜋
In his s udy, o he 8.5-minu e EEG da a wi h a sampling equency o 256 Hz, we i s emo ed ou
unusable elec odes. Then, we emo ed he i s and las 10 seconds o he EEG da a o a oid ansi ion
e ec s, and we segmen ed he EEG da a in o non-o e lapping epochs by a ime window o 10 seconds,
so he e we e a o al o 49 epochs. Then, an adjacency ma ix o was calcula ed by PLI o
60
×
60
each epoch and each equency band.
2.3 Ne wo k analysis
G aph heo y is no mally used a e he calcula ion o he adjacency ma ix o quan i y he opology o
b ain ne wo ks. In his s udy, we used h ee commonly used ne wo k measu es o quan i y in luence,
unc ional seg ega ion and unc ional in eg a ion, including deg ee, clus e ing coe icien and
cha ac e is ic pa h leng h. All he ne wo k measu es men ioned abo e we e compu ed using he B ain
Connec i i y Toolbox (Rubino and Spo ns 2010) (h p://www.b ain-connec i i y- oolbox.ne ).
Fo an adjacency ma ix , wi h nodes, ep esen s he connec ion s eng h be ween node and
𝑮
𝑁
𝑤
𝑖𝑗
𝑖
node , whe e . The diagonal elemen s mean sel -connec ions o nodes, so
𝑗
0
≤
𝑤
𝑖𝑗
≤
1
𝑤
𝑖𝑖
=
0, 𝑖
=
1
,
.
2,⋯,𝑁
2.3.1 Deg ee
Deg ee is conside ed an impo an ma ke o ne wo k de elopmen and esilience, and o a weigh ed
ne wo k, he deg ee o node can be de ined as ollows:
𝑖
𝑘
𝑖
=
∑
𝑗
∈
𝑁
𝑤
𝑖𝑗
#
(
5
)
2.3.2 Clus e ing coe icien
Clus e ing coe icien is a measu e o unc ional seg ega ion which is a e lec ion o he local
o ganiza ion o a ne wo k by depic ing he endency o a node o ming local iangles (Rubino and
Spo ns 2010), and i s de ini ion o a weigh ed ne wo k o node is desc ibed as ollows:
𝑖
𝐶
𝑖
=
2
𝑡
𝑖
𝑘
𝑖
(
𝑘
𝑖
―
1
)
,
#
(
6
)
whe e is he geome ic mean o iangles a ound . The clus e ing coe icien
𝑡
𝑖
=
1
2
∑
𝑗,ℎ
∈
𝑁
(
𝑤
𝑖𝑗
𝑤
𝑖ℎ
𝑤
𝑗ℎ
)
1
3
𝑖
o he whole ne wo k is de ined as he mean o clus e ing coe icien o all nodes,
𝐶
=
1
𝑁
∑
𝑖
∈
𝑁
𝐶
𝑖
. #
(
7
)
2.3.3 Cha ac e is ic pa h leng h
Cha ac e is ic pa h leng h is he a e age o sho es pa h leng h be ween all pai s o nodes and is
commonly used o measu e unc ional in eg a ion. Cha ac e is ic pa h leng h is a e lec ion o he
e iciency o a ne wo k (Bullmo e and Spo ns 2009). The de ini ion is desc ibed as ollows:
𝐿
=
1
𝑁
∑
𝑖
∈
𝑁
∑
𝑗
∈
𝑁, 𝑗
≠
𝑖
𝑑
𝑖𝑗
𝑁
―
1
, #
(
8
)
whe e is he sho es pa h leng h be ween node and node , and is he sho es
𝑑
𝑖𝑗
=
∑
𝑎
𝑢𝑣
∈
𝑔
𝑤
𝑖⟷𝑗
1
𝑤
𝑢𝑣
𝑖
𝑗
𝑔
𝑤
𝑖⟷𝑗
weigh ed pa h be ween and .
𝑖
𝑗
2.4 S a is ical analysis
To de e mine in which equency band a signi ican di e ence exis s be ween he CON g oup and MDD
g oup, Ne wo k Based S a is ic Toolbox was applied in his s udy (Zalesky e al. 2010). The NBS
me hod can con ol he amily-wise e o when mul iple uni a ia e es ing is pe o med a each
connec ion o a ne wo k. NBS me hod is used o iden i y signi ican b ain ne wo k subs uc u es o med
by some sup a h eshold links bu no o iden i y indi idual links as being signi ican . The h eshold is
used on he es s a is ic compu ed o each pai wise connec ion, and di e en h esholds can cons uc
di e en le el o spa se g aphs. A e a e aging he adjacency ma ices ac oss ime windows o each
subjec , s a is ical analysis was pe o med be ween he CON g oup and MDD g oup o each equency
band. A signi icance le el o co ec ed and a nonpa ame ic pe mu a ion es o 5000
𝑃
<
0.05
pe mu a ions we e used in his s udy. T- es was selec ed o he s a is ical es , and di e en es s a is ic
h esholds ( -s a is ic) we e es ed o iden i y he mos signi ican b ain ne wo k subs uc u es.
2.5 Classi ica ion
Conside ing he limi a ions o using sliding windows wi hou o e lapping, which will lead o he
p oblem ha FC opology may no been well desc ibed wi hin one ixed ime window (Liuzzi e al.
2019), we a e aged e e y six ime windows ( he connec i i y ma ices wi hin one minu e) o gene a e
one classi ica ion sample o highligh he main connec i i y pa e ns du ing music pe cep ion. To
imp o e classi ica ion pe o mances, we cons uc ed spa se ne wo ks based on he no ion o connec ed
g aphs o emo e edundan in o ma ion, which can ensu e ha e e y node has a connec ion o ano he
node o a spa se ne wo k. The de ailed me hod o h eshold selec ion can be ound in e e ence (A ay
and Biyikoǧlu 2005).
In his s udy, we used he adjacency ma ices ob ained by PLI o pe o m classi ica ion, and we
compa ed he classi ica ion pe o mance using o iginal ne wo ks and spa se ne wo ks be ween del a
and be a equency bands and six classi ie s, including decision ee (DT), Gaussian mix u e model
(GMM), k-nea es neighbo (KNN), naï e Bayes (NB), andom o es (RF) and suppo ec o machine
(SVM). We un olded he adjacency ma ix o a ec o as one sample. Because o he symme y p ope y
o he adjacency ma ix, we can ob ain a iables o each sample.
𝑁
(
𝑁
―
1)/2
=
60(60
―
1)/2
=
1770
The e o e, we can ge 152 samples o he CON g oup and 160 samples o he MDD g oup. To a oid
o e i ing, p incipal componen analysis (PCA) was applied o dimension educ ion be o e
classi ica ion.
To assess he pe o mance o classi ica ion, we calcula ed some s a is ical e alua ion measu emen s
including accu acy, sensi i i y, and speci ici y (Yan e al. 2019), which can be calcula ed by:
accu acy
=
𝑇𝑃
+
𝑇𝑁
𝑇𝑃
+
𝐹𝑃
+
𝑇𝑁
+
𝐹𝑁
, #
(
9
)
sensi i i y
=
𝑇𝑃
𝑇𝑃
+
𝐹𝑁
,
#
(
10
)
speci ici y
=
𝑇𝑁
𝑇𝑁
+
𝐹𝑃
, #
(
11
)
whe e TP, TN, FP and FN ep esen ue posi i e, ue nega i e, alse posi i e and alse nega i e,
espec i ely. To ob ain a eliable classi ica ion esul , we shu led he da a o de , used 10- old c oss
alida ion, and an 10 imes o each classi ie . Then, we a e aged he classi ica ion esul s o calcula e
he inal pe o mance o each classi ie and each equency band.
3. Resul s
3.1 Phase synch oniza ion
A e s a is ical analysis by NBS o each equency band, a signi ican di e ence only exis ed in wo
equency bands: del a and be a bands (del a: , he a: , alpha: , be a:
𝑃
=
0.0450
𝑃
=
0.
2386
𝑃
=
0.
3447
, gamma: ). The adjacency ma ices o hese wo equency bands o he CON
𝑃
=
0.03
44
𝑃
=
0
.0649
g oup and MDD g oup a e shown in Figu e 1. Fo he del a equency band, he connec i i y s eng h
inc eased in he MDD g oup (del a: , ) compa ed o he CON g oup (del a:
mean
=
0.0867
SD
=
0.0197
, ;). Howe e , o he be a equency band, he connec i i y s eng h o
mean
=
0.0853
SD
=
0.0178
he MDD g oup ( , ) dec eased compa ed wi h ha o he CON g oup
mean
=
0.0408
SD
=
0.0133
( , ). F om Figu e 1, we can see ha sho -dis ance synch oniza ion was
mean
=
0.0485
SD
=
0.0143
s onge han long-dis ance synch oniza ion, and he whole b ain connec i i y was o med by many
small modules.
The signi ican b ain ne wo k connec ions be ween he CON g oup and he MDD g oup in del a and
be a equency bands a e shown in Figu e 2. We can see ha he e we e 13 signi ican connec ions in
he del a band dis ibu ed wi hin igh cen al b ain a eas and be ween igh empo al and le pa ie al
b ain egions. While in he be a band, he e we e 43 signi ican connec ions cha ac e ized mos ly by
long-dis ance edges, which we e dis ibu ed mos ly wi hin on al b ain a eas and be ween on al and
pa ie a-occipi al b ain a eas. The subs uc u es we e conside ed o be impo an indica o s o he
Mum az W, Xia L, Ali SSA, Yasin MAM, Hussain M, Malik AS. Elec oencephalog am (EEG)-based compu e -
aided echnique o diagnose majo dep essi e diso de (MDD). Biomed Signal P ocess Con ol.
2017;31:108–115.
Nugen AC, Fa me C, E ans JW, Snide SL, Bane jee D, Za a e CA. Mul imodal imaging e eals a complex
pa e n o dys unc ion in co icolimbic pa hways in majo dep essi e diso de . Hum B ain Mapp.
2019;40:3940–3950.
O’Neill GC, Tewa ie P, Vidau e D, Liuzzi L, Wool ich MW, B ookes MJ. Dynamics o la ge-scale
elec ophysiological ne wo ks: A echnical e iew. Neu oimage. 2018;180:559–576.
Ohnishi T. Func ional Ana omy o Musical Pe cep ion in Musicians. Ce eb Co ex. 2001;11:754–760.
Olb ich S, T änkne A, Chi ka T, Hege l U, Schönknech P. Func ional connec i i y in majo dep ession:
Inc eased phase synch oniza ion be ween on al co ical EEG-sou ce es ima es. Psychia y Res -
Neu oimaging. 2014;222:91–99.
Pal a S, Pal a JM. Disco e ing oscilla o y in e ac ion ne wo ks wi h M/EEG: Challenges and b eak h oughs.
T ends Cogn. Sci. 2012;16:219–230.
Pa k H-J, F is on K. S uc u al and unc ional b ain ne wo ks: om connec ions o cogni ion. Science.
2013;342:1238411.
Punkanen M, Ee ola T, E kkilä J. Biased emo ional ecogni ion in dep ession: Pe cep ion o emo ions in music
by dep essed pa ien s. J A ec Diso d. 2011;130:118–126.
Rami ez R, Palencia-Le le M, Gi aldo S, Vam akousis Z. Musical neu o eedback o ea ing dep ession in
elde ly people. F on Neu osci. 2015;9:354.
Ren Y, Cong F, Ris aniemi T, Wang Y, Li X, Zhang R. T ansien seizu e onse ne wo k o localiza ion o
epilep ogenic zone: e ec i e connec i i y and g aph heo y-based analyses o ECoG da a in empo al lobe
epilepsy. J Neu ol. 2019;266:844–859.
Rubino M, Spo ns O. Complex ne wo k measu es o b ain connec i i y: Uses and in e p e a ions. Neu oimage.
2010;52:1059–1069.
Ruiz-Gómez SJ, Ho ne o R, Poza J, Ma u ana-Candelas A, Pin o N, Gómez C. Compu a ional modeling o he
e ec s o EEG olume conduc ion on unc ional connec i i y me ics. Applica ion o Alzheime ’s disease
con inuum. J Neu al Eng. 2019;16: 066019.
Ruiz MH, Koelsch S, Bha acha ya J. Dec ease in ea ly igh alpha band phase synch oniza ion and la e gamma
band oscilla ions in p ocessing syn ax in music. Hum B ain Mapp. 2009;30:1207–1225.
Sakai K, Yamada K. Machine lea ning s udies on majo b ain diseases: 5-yea ends o 2014–2018. Jpn J Radiol.
2019;37:34–72.
Scho elen JM, G oss J. Sou ce connec i i y analysis wi h MEG and EEG. Hum B ain Mapp. 2009;30:1857–1865.
Singh MK, Kesle SR, Hadi Hosseini SM, Kelley RG, Ama ya D, Hamil on JP, e al. Anomalous g ay ma e
s uc u al ne wo ks in majo dep essi e diso de . Biol Psychia y. 2013;74:777–785.
Spo ns O. G aph heo y me hods: applica ions in b ain ne wo ks. Dialogues Clin Neu osci. 2018;20:111–121.
S am CJ, Nol e G, Da e sho e A. Phase lag index: Assessmen o unc ional connec i i y om mul i channel
EEG and MEG wi h diminished bias om common sou ces. Hum B ain Mapp. 2007;28:1178–1193.
Toi iainen P, Allu i V, B a ico E, Wallen in M, Vuus P. Cap u ing he musical b ain wi h Lasso: Dynamic
decoding o musical ea u es om MRI da a. Neu oimage. 2014;88:170–180.
Uy denhoe P, Po elange P, Jacquy J, Cha les G, Linkowski P, Mendlewicz J. Regional ce eb al blood low and
la e alized hemisphe ic dys unc ion in dep ession. B J Psychia y. 1983;143:128–132.
Van Den B oek SP, Reinde s F, Donde winkel M, Pe e s MJ. Volume conduc ion e ec s in EEG and MEG.
Elec oencephalog Clin Neu ophysiol. 1998;106:522–534.
Vee IM, Beckmann C, Van Tol M-J, Fe a ini L, Milles J, Vel man D, e al. Whole b ain es ing-s a e analysis
e eals dec eased unc ional connec i i y in majo dep ession. F on Sys Neu osci. 2010;4:41.

Vinck M, Oos en eld R, Van Winge den M, Ba aglia F, Penna z CMA. An imp o ed index o phase-
synch oniza ion o elec ophysiological da a in he p esence o olume-conduc ion, noise and sample-size
bias. Neu oimage. 2011;55:1548–1565.
Wang X, Liu W, Toi iainen P, Ris aniemi T, Cong F. G oup analysis o ongoing EEG da a based on as double-
coupled nonnega i e enso decomposi ion. J Neu osci Me hods. 2020;330: 108502.
Wang X, Ren Y, Zhang W. Dep ession Diso de Classi ica ion o MRI Da a Using Spa se Low-Rank Func ional
B ain Ne wo k and G aph-Based Fea u es. Compu Ma h Me hods Med. 2017;2017:1–11.
Wa anabe T, Yagishi a S, Kikyo H. Memo y o music: Roles o igh hippocampus and le in e io on al gy us.
Neu oimage. 2008;39:483–491.
Wheele RE, Da idson RJ, Toma ken AJ. F on al b ain asymme y and emo ional eac i i y: A biological
subs a e o a ec i e s yle. Psychophysiology. 1993;30:82–89.
Whi on AE, Deccy S, I onside ML, Kuma P, Bel ze M, Pizzagalli DA. Elec oencephalog aphy Sou ce
Func ional Connec i i y Re eals Abno mal High-F equency Communica ion Among La ge-Scale
Func ional Ne wo ks in Dep ession. Biol Psychia y Cogn Neu osci Neu oimaging. 2018;3:50–58.
Womelsdo T, Scho elen JM, Oos en eld R, Singe W, Desimone R, Engel AK, e al. Modula ion o neu onal
in e ac ions h ough neu onal synch oniza ion. Science. 2007;316:1609–1612.
Wu D, Yuan Y, Bai F, You J, Li L, Zhang Z. Abno mal unc ional connec i i y o he de aul mode ne wo k in
emi ed la e-onse dep ession. J A ec Diso d. 2013;147:277–287.
Wu J, Zhang J, Liu C, Liu D, Ding X, Zhou C. G aph heo e ical analysis o EEG unc ional connec i i y du ing
music pe cep ion. B ain Res. 2012;1483:71–81.
Xia MR, Si TM, He Y. Imaging connec omics in dep ession. CNS Neu osci The . 2018;24:991.
Yan R, Zhang C, Sp uy K, Wei L, Wang Z, Tian L, e al. Mul i-modali y o polysomnog aphy signals’ usion o
au oma ic sleep sco ing. Biomed Signal P ocess Con ol. 2019;49:14–23.
Ye M, Yang T, Qing P, Lei X, Qiu J, Liu G. Changes o unc ional b ain ne wo ks in majo dep essi e diso de :
A g aph heo e ical analysis o es ing-s a e MRI. PLoS One. 2015;10:e0133775.
Zalesky A, Fo ni o A, Bullmo e ET. Ne wo k-based s a is ic: Iden i ying di e ences in b ain ne wo ks.
Neu oimage. 2010;53:1197–1207.
Zhu X, Wang X, Xiao J, Liao J, Zhong M, Wang W, e al. E idence o a dissocia ion pa e n in es ing-s a e
de aul mode ne wo k connec i i y in i s -episode, ea men -nai e majo dep ession pa ien s. Biol
Psychia y. 2012;71:611–617.
Figu e legends
Figu e 1. A e aged adjacency ma ices o he CON g oup and MDD g oup ac oss ime windows o
del a and be a equency bands. Each adjacency ma ix is o med by a ma ix wi h ze o alues
60
×
60
in he diagonal. CON, con ol; MDD, majo dep ession diso de .
Figu e 2. The signi ican b ain ne wo k connec ions in del a and be a equency bands o he CON
g oup and MDD g oup. The esul s we e conduc ed by he NBS me hod using 5000 pe mu a ions,
co ec ed alue o , and maximum componen h eshold o he del a band and
𝑝
𝑝
<
0.05
𝑡
>
3.1
o he be a band. The e a e 13 signi ican connec ions in he del a band and 43 signi ican
𝑡
>
2.3
connec ions in he be a band. CON, con ol; MDD, majo dep ession diso de ; NBS, ne wo k based
s a is ic.
Figu e 3. The deg ee o each node in del a and be a equency bands o he CON g oup and MDD
g oup. CON, con ol; MDD, majo dep ession diso de .
Figu e 4. Boxplo o clus e ing coe icien and cha ac e is ic pa h leng h o he CON g oup and MDD
g oup in del a and be a equency bands. The uppe and lowe black lines ep esen he maximum alue
and he minimum alue, espec i ely, and he ed c oss indica es ou lie s. The bo om and op edges o
he blue box indica e he 25 h and 75 h pe cen iles, and he ed line and hombus in he box indica e he
median alue and he mean alue, espec i ely. CON, con ol; MDD, majo dep ession diso de .
Table 1. Means and s anda d de ia ions o age, gende , educa ion and clinical measu es o he CON
g oup and MDD g oup.
CON g oup
MDD g oup
Analysis
Mean
SD(Range)
Mean
SD(Range)
- alue
𝑝
Age
38.4
11.8(24-65)
42.9
11.0(23-58)
>0.05
Educa ion
13.6
3.8(6-20)
12.8
3.4(6-16)
>0.05
HRSD
2.4
1.3(0-4)
23.3
3.6(16-28)
<0.01
HAMA
2.4
1.3(0-5)
19.2
3.0(15-25)
<0.01
MMSE
28.2
0.9(27-30)
28.1
1.1(26-30)
>0.05
Du a ion
0
0
12.8
8.5(2-36)
-
Gende
14 emales, 5 males
14 emales, 6 males
-
Abb e ia ions: CON, con ol; MDD, majo dep ession diso de ; SD, s anda d de ia ions; HRSD,
Hamil on Ra ing Scale o Dep ession; HAMA, Hamil on Anxie y Ra ing Scale; MMSE, Mini-Men al
S a e Examina ion.
Table 2. The classi ica ion accu acy o six classi ie s in del a and be a equency bands.
Ne wo k
DT
GMM
KNN
NB
RF
SVM
O iginal
54.7%
49.7%
65.7%
61.2%
62.0%
66.9%
Del a
Spa se
60.9%
53.2%
68.7%
62.5%
69.7%
72.9%
O iginal
61.4%
49.5%
77.4%
55.2%
70.9%
78.2%
Be a
Spa se
68.5%
54.8%
85.6%
56.7%
82.3%
89.7%
Abb e ia ions: DT, decision ee; GMM, Gaussian mix u e model; KNN, k-nea es neighbo ; NB, naï e
Bayes; RF, andom o es ; SVM, suppo ec o machine.