Neural correlates underlying mental calculation in abacus experts

12 Pages·2003·0.3 MB·English

NeuroImage19(2003)296–307 www.elsevier.com/locate/ynimg
Neural correlates underlying mental calculation in abacus experts:
a functional magnetic resonance imaging study
Takashi Hanakawa,a,b Manabu Honda,a,c,d,* Tomohisa Okada,c,e
Hidenao Fukuyama,a and Hiroshi Shibasakia,c
aHumanBrainResearchCenter,KyotoUniversityGraduateSchoolofMedicine,Kyoto,Japan
bHumanMotorControlSection,NINDS,NIH,Bethesda,MD20892,USA
cNationalInstituteforPhysiologicalSciences,Okazaki,Japan
dPRESTO,JapanScienceandTechnologyCorp.,Kawaguchi,Japan
eDepartmentofNuclearMedicine,KyotoUniversityGraduateSchoolofMedicine,Kyoto,Japan
Received5August2002;accepted10December2002
Abstract
Expertsofabacusoperationdemonstrateextraordinaryabilityinmentalcalculation.Thereispsychologicalevidencethatabacusexperts
utilize a mental image of an abacus to remember and manipulate large numbers in solving problems; however, the neural correlates
underlyingthisexpertiseareunknown.Usingfunctionalmagneticresonanceimaging,wecomparedtheneuralcorrelatesassociatedwith
threemental-operationtasks(numeral,spatial,verbal)amongsixexpertsinabacusoperationsandeightnonexperts.Ingeneral,therewas
more involvement of neural correlates for visuospatial processing (e.g., right premotor and parietal areas) for abacus experts during the
numeralmental-operationtask.Activityoftheseareasandthefusiformcortexwascorrelatedwiththesizeofnumeralsusedinthenumeral
mental-operationtask.Particularly,theposteriorsuperiorparietalcortexrevealedsignificantlyenhancedactivityforexpertscomparedwith
controlsduringthenumeralmental-operationtask.Comparisonwiththeothermental-operationtasksindicatedthatactivityintheposterior
superiorparietalcortexwasrelativelyspecifictocomputationin2-dimensionalspace.Inconclusion,mentalcalculationofabacusexperts
islikelyassociatedwithenhancedinvolvementoftheneuralresourcesforvisuospatialinformationprocessingin2-dimensionalspace.
©2003ElsevierScience(USA).Allrightsreserved.
Introduction abacus players initially practice to simulate abacus opera-
tionsinmindwithactualfingermovements,asiftheycould
To perform complex calculations, most people have re- pushimaginedabacusbeads(Fig.1A).However,theyeven-
lied on physical devices such as pencils and papers, me- tually cease to use overt finger movements because it is
chanical calculators, slide rules, and more recently digital believed that finger movements rather slow down calcula-
computers. One such device, gradually downplayed in the tion speed at higher levels of skill.
computer age, is an abacus, or “soroban” in Japanese. In- Mentalarithmeticrequiresintegrationofmultiplecogni-
terestingly, however, abacus experts not only manipulate
tive functions including number recognition, retrieval of
the tool skillfully in its physical form but also gain the
arithmetic facts, temporary storage of intermediate results,
ability to mentally calculate extraordinarily large num-
and manipulation of mental representations. Much effort
bers—often more than 10 digits—with unusual speed and
has been devoted to clarify cognitive mechanisms and un-
accuracy (Stigler, 1984). To achieve this level of skill,
derlying neural correlates of number recognition and arith-
meticoperations.Linguisticprocessingissuggestedtoplay
anessentialroleinexactmentalcalculationinadultsonone
*Correspondingauthor.LaboratoryofCerebralIntegration,National
hand (Dehaene et al., 1999); it is likely that visuospatial
InstituteforPhysiologicalSciences,38NishigonakaMyodaiji-cho,Oka-
zaki444-8585,Japan.Fax:(cid:1)81-564-52-7913. processingisalsoimportant,especiallyinthedevelopmen-
E-mailaddress:[email protected](M.Honda). tal stage, for acquiring mathematical concepts on the other
1053-8119/03/$–seefrontmatter©2003ElsevierScience(USA).Allrightsreserved.
doi:10.1016/S1053-8119(03)00050-8
T.Hanakawaetal./NeuroImage19(2003)296–307 297
logical evidence (Hatano et al., 1977; Hatano and Osawa,
1983;Stigler,1984;HattaandIkeda,1988),wehypothesized
that the mental calculation of abacus experts might primarily
depend on brain areas involved in visual, visuospatial, and
visuomotorimaginationratherthanthoseinvolvedinlinguistic
informationprocessing.
To test this hypothesis, we used mental-operation tasks
in which subjects serially and mentally updated mental
representations according to external stimuli (Hanakawa et
al., 2002; Sawamoto et al., 2002). The mental-operation
tasks had advantages for the study of mental calculation in
abacus experts. First, no motor response was required dur-
ing task periods. We considered this important because
motor responses for behavioral reports would affect the
activityofbrainareasunderlyingvisuomotorimagery.Pre-
viousneuroimagingstudiesonvisuomotorimageryshowed
that the underlying neural correlates substantially over-
lapped with brain areas for high-order motor control
(GrezesandDecety,2001;Hanakawaetal.,2003).Second,
the mental-operation tasks can investigate different mental
representations(e.g.,numeral,spatial,orverbalrepresenta-
tions) by changing a rule associating a given stimulus with
Fig.1.Abacusoperationandbehavioraltasks.(A)Thebasicoperationofa
arequisitementaloperation.Third,accuracyoftaskperfor-
Japaneseabacusor“soroban.”Ithascolumnsofbeads,andeachofthemhas
a place value, corresponding to the ones, tens, hundreds, and so on (i.e., mance was appraisable after each task period. This process
base-10systemofnumeration).Onceaspecificcolumnofbeadsisarbitrarily providedabehavioralobservation,whichwasnecessaryfor
definedasthe“ones,”theothercolumnsarevaluedrelativetoit.Eachcolumn assessing experimental confounders such as task difficulty.
ofbeadshasanupperandalowersection.Thebeadintheuppersectionis
equalto5timestheunitvalueofthecolumnswhenitispusheddowntoward
thehorizontaldividingbar,andeachofthe4lowerbeadsisequalto1unit
valueofthecolumnswhenpushedup.Forexample,fortheadditionof9and Materials and methods
7,4lowerbeadsupand1upperbeaddowninacolumnrepresentthenumber
9.Toexecutetheadditionrequiringacarrying,onefirstsubtractsthecom- Subjects
plementoftheaddendto10(3here)andthenadds1tothetenscolumn.One
canaccomplishthiscalculationusingasinglefingermovement,twistingthe
The abacus-expert group consisted of one man and five
thumb(T)andtheindexfinger(I)topushup1beadinthetenscolumn(white
arrow)andtopushdown3beadsintheonescolumn(blackarrow),respec- women, ranging between 24–30 (mean (cid:2) SD (cid:3) 28 (cid:2) 2.8)
tively.(B)Inthenumeralmental-operationtask,subjectswereaskedtoadda years old. All had been qualified as Japanese abacus masters
seriesofnumberspresentedvisually.(C)Inthespatialmental-operationtask, withthehighestdegree(10thlevel)forbothmentalcalculation
subjectswererequiredtomentallymovethemarkerlocationaccordingtoan
andactualabacusoperations.Allhadexperiencedgreaterthan
arroworapairofarrows.Forexample,oneshouldmovethemarker1square
17 years of almost daily abacus training for the purpose of
totheright,asindicatedbythesinglearrowpointingtotheright,andthen
down2squares,asindicatedbythedoublearrowspointingdown,andsoforth. competition.Theywereallright-handedwithalateralityindex
(D)Intheverbalmental-operationtask,subjectsfirstrememberedthedayof (LI) of 0.8–1, except for one ambidextrous woman (LI (cid:3)
theweekspecifiedbyakanjicharacterandadvancedthedateasinstructedby (cid:4)0.2)(Oldfield,1971).Thenonexpertcontrolgroupincluded
the numbers. For example, onefirst advanced the day from Wednesday to onemanandsevenwomen,rangingbetween22–33(24(cid:2)3.6)
Fridaybythenumber2,fromFridaytoMondaybythenumber3,andsoforth.
years old. Nonexperts had negligible experience in abacus
operations. All nonexperts were right-handed (LI (cid:3) 0.8–1),
(Simon, 1999). Recently, a neuroimaging study on a non- except for one ambidextrous woman (LI (cid:3) 2.5). All subjects
abacuscalculationprodigyhasindicatedthatexpertiseuses gavewritteninformedconsentaccordingtothestudyprotocol
differentbrainregionsfromthosefornonexperts(Pesentiet approvedbytheinstitutionalethicscommittee.
al., 2001).
The neural correlates of the calculation strategy employed Task
by abacus experts remain unknown, although psychological
studieshaveshownthatvisualstrategyunderliesthisunusual We used three versions (numeral, spatial, and verbal) of
calculation ability (Hatano et al., 1977; Hatano and Osawa, mental-operationtasks(Hanakawaetal.,2002).Allmental-
1983; Stigler, 1984; Hatta and Ikeda, 1988). In the present operation tasks began with the presentation of a prime
study,weinvestigatedbrainactivityduringmentalcalculation stimulus (PS) for 2 s, followed by a task period of 30 s.
insixtop-levelJapaneseabacusexperts,usingfunctionalmag- During each task period, 15 instruction stimuli (IS) were
netic resonance imaging (fMRI). Based on previous psycho- presentedwithfixedinterstimulusintervalsof2s(stimulus
298 T.Hanakawaetal./NeuroImage19(2003)296–307
frequency, 0.5 Hz). Throughout the task periods, subjects stimulus comprising two visual stimuli side-by-side was dis-
were asked to fixate on the visual stimuli without any played for 2 s. The response stimuli were a pair of numbers,
accompanying movement. Each version of the mental-op- grids with a marker, and kanji characters for the numeral,
eration tasks was termed according to the most natural spatial, and verbal mental-operation tasks, respectively. The
strategy,althoughallmental-operationtasksinvolveatype tasksolutionwasreportedviaawristextensionmovement:a
ofcomputation,eithervectororscalaroperation,necessary leftwristextensionindicatedthattheleftfigurewascorrect;a
for updating mental representations. rightwristextensionindicatedthattherightfigurewascorrect;
In the numeral mental-operation task, each PS and IS and a bilateral wrist extension indicated that neither was cor-
wasanArabicnumber(Fig.1B).Subjectswereaskedtoadd rect. Thus, the probability of a randomly occuring correct
upallpresentednumbersandreportthefinalsumaftereach answerwas33%.Accuracyoftheresponsesservedasbehav-
task period. In a preliminary experiment outside of the ioralobservationstoassesstaskperformance.
scanner, subjects practiced the numeral mental-operation Avisualfixationtask,inwhichsubjectskeptfixatingon
task with different levels of difficulty (problems with sin- acrossflashingonandoffatthesamerateasISs,servedas
gle-, three-, or six-digit numbers as addends). All abacus a common baseline. In each experimental run (4 min), task
experts considered the task with three-digit numbers most periods were repeated four times for the same mental-
essential since they were the most frequently used number operation task version, alternated with baseline periods for
unit or abacus “width” for practice. On the other hand, all the visual fixation task, in 30-s blocks. A scanning run for
controlsubjectsconsideredthenumeraltaskwithmorethan each task condition was repeated two times.
single-digit numbers impossible. Obviously, subjective dif-
ficulty of the numeral mental-operation task was different Functional MRI
betweenthetwogroups,ifthesameproblemswereusedfor
both. Therefore, we asked abacus experts to perform the FMRIexperimentswereconductedona1.5-Tscannerwith
taskwiththreedifferentlevelsofdifficultyandthenplanned a standard head-coil (GE, Milwaukee, WI, USA). T2*-sensi-
to use accuracy of the performance to match the task diffi- tive, gradient-echo echo planar images were obtained (inter-
culty between the two groups. For the abacus experts, sin- scanintervals(cid:3)6s,echotime(cid:3)43ms,flipangle(cid:3)90°,field
gle-digit numbers were presented as PS and IS to three ofview(cid:3)22cm,matrixsize(cid:3)64(cid:5)64,38contiguousaxial
subjects of six, three-digit numbers were presented to all slices, and voxel size (cid:3) 3.5 mm3). High-resolution, T1-
subjects,andsix-digitnumberswerepresentedtoallbutone weighted, 3-dimensional structural images were acquired be-
who thought the task too difficult. These three conditions fore the fMRI acquisition for anatomic registration (fast
withinthenumeralmental-operationtaskalsoallowedusto spoiledgradient-recalledatsteady-stateimage,datamatrix(cid:3)
explorebrainareasinwhichactivitychangewascorrelated 256 (cid:5) 256 (cid:5) 124, voxel size (cid:3) 0.86 (cid:5) 0.86 (cid:5) 1.5 mm).
with the size of numbers within the task (i.e., number-size Subjects lay supine on a scanner bed. Foam cushions and
effect). For the control group, only single-digit numbers elastic tape were used to minimize head motion. subjects
were used for PS and IS. viewedvisualstimuliback-projectedontoascreen,througha
In the spatial mental-operation task, PS was a marker mirror built into a standard head coil. Each visual stimulus,
presented in one square of a grid subdivided into nine subtendinga2.5°(vertical)(cid:5)2.5–5°(horizontal)visualangle,
squares,andeachISwaseitheranarroworapairoftandem waspresentedinthecenterofview;thesubjectswereableto
arrowspointinginoneoffourdirections(up,down,right,or seethestimuliwithoutmovingtheireyes.Theabsenceofovert
left) (Fig. 1C). During the task periods after the PS disap- movements was confirmed in a preliminary training session
peared, subjects mentally moved the marker on the imag- outside of the scanner as well as during the actual fMRI
inedgridaccordingtotheISsandreportedthefinallocation acquisition,viaconstantvisualinspection.Monitoringforeye
of the marker. Visuospatial imagery was expected as a movementwasnotavailable.
natural strategy for the spatial mental-operation task. Image analysis was performed using statistical paramet-
Intheverbalmental-operationtask,PSwasakanjichar- ricmapping(SPM99,http://www.fil.ion.ucl.ac.uk/spm)im-
acter indicating one of the days of the week, and each IS plemented on MATLAB (MathWorks, Inc., Natick, MA,
was a single-digit number (1, 2, or 3) (Fig. 1D). Starting USA). After the first two images from each run were dis-
fromthedayindicatedbythePS,subjectsseriallyadvanced carded, the remaining functional images were spatially
the day of the week according to the ISs (e.g., the number aligned and resliced using sinc interpolation. The anatomi-
2 meant 2 days later) and reported the day they finally cal images were individually coregistered onto the mean
reached. The days of the week constitute one of the most functional images from each subject and spatially normal-
familiar string of words to Japanese people, regardless of izedtofitaMontrealNeurologicalInstitutetemplate(Evans
abacus expertise. Therefore, the natural strategy was se- et al., 1993) based on the standard stereotaxic coordinate
quentially advancing the days in mind based on verbal– system (Talairach and Tournoux, 1988). The functional
phonological rehearsal. images were spatially normalized using the same transfor-
SubjectsreportedthefinalimageofthetaskafterthelastIS mationmatrixandsmoothedwitha3-dimensionalGaussian
forbehavioralassessment.Aftereachtaskperiod,theresponse kernel of 10-mm full width at half maximum.
T.Hanakawaetal./NeuroImage19(2003)296–307 299
A first-level multiregression analysis was performed to Brain activity: general patterns
test the correlation between MRI signal changes and a
boxcar function convolved with the canonical hemody- Brain activity associated with each mental-operation
namicresponsefunction.Inthisprocess,aglobaldifference task, relative to the visual fixation task, was examined as a
insignalintensitywasremoved,usingproportionalscaling. conjunction analysis across subjects for each group (Table
Planned linear contrasts were applied to the parameter es- 1,Fig. 2A). For the abacus experts, the numeral mental-
timates from the multiregression analysis, which yielded t operation task induced activity in the frontal operculum,
statistic maps (Friston et al., 1995). The effects of interest superior precentral sulcus (SPcS), posterior parietal cortex
within each group were tested in terms of the conjoint including intraparietal sulcus areas (IPS), and posterior su-
effects across subjects (Friston et al., 1999). A statistical perior parietal cortex/precuneus, fusiform gyrus, and cere-
threshold was set at the height threshold of P (cid:6) 0.05 bellar hemisphere. This activity was bilaterally symmetri-
(corrected for multiple comparisons). cal. For the controls, the numeral mental-operation task
For the between-group comparisons of activity during the activatedtheprefrontalcortex,Broca’sarea,medialfrontal
numeraltask,thetaskinvolvingsix-digitnumbersforexperts areasincludingtheanteriorcingulatecortexandpresupple-
was compared with the task for nonexperts because the task mentary motor area (pre-SMA), and lateral parietal area.
performance was comparable only between these two condi- Nonexpertsalsoshowedactivityintheareasobserveddur-
tions across the groups (see Results). Mean effect images ing the numeral task for experts (i.e., frontal operculum,
reflectingaslopeofaparticularregressorwerecomputedfrom SPcS, IPS, posterior parietal cortex, fusiform gyrus, and
the first-level multiregression analysis and were used in a cerebellar hemisphere); however, this calculation-related
second-level two-sample t test based on a random effects activity for controls was strongly lateralized to the left
model. This analysis allowed population level inference be- hemisphere. This pattern of activity during mental calcula-
tween abacus experts and nonexperts. For the between-group tionwasconsistentwithpreviousneuroimagingstudies(de
comparisons, an explicit mask image was used for each con- Jong et al., 1996; Dehaene et al., 1996; Rueckert et al.,
trastinthetwo-samplettest(expertsminuscontrolsorcontrols 1996; Burbaud et al., 1999; Rickard et al., 2000; Gruber et
minusexperts)tolimitthesearchvolumetobrainareasshow- al., 2001; Pesenti et al., 2001; Zago et al., 2001).
ingtask-relatedactivityineachgroup(fromsecond-levelone- The spatial mental-operation task induced symmetrical
sample t test, P (cid:6) 0.005, uncorrected). This was to avoid bilateral activity of the frontal operculum, SPcS, IPS, pos-
detecting between-group differences due to “deactivations” terior superior parietal cortex, precuneus, and visual asso-
greaterinacontrolgroupthanagroupofinterestsincethere ciationareas.Thisactivitywasverysimilarbetweenthetwo
are still many unsolved issues in interpreting deactivations in groups; moreover, the activity in the premotor and parietal
neuroimaging studies (Gusnard and Raichle, 2001). Again, a cortices during the spatial task for both groups was almost
statisticalthresholdwassetattheheightthresholdofP(cid:6)0.05 identical to that observed during mental calculation for
(corrected for multiple comparisons). Similarly, the between- experts. During the verbal mental-operation task, experts
groupdifferenceinactivitywastestedfortheothertwomen- andnonexpertsbothexhibitedactivityprimarilyintheme-
tal-operationtasks. dial frontal areas (anterior cingulate cortex and pre-SMA),
SPcS, lateral parietal areas, and IPS. This activity was
predominantly in the left hemisphere in the two groups,
Results although the experts tended to exhibit more activity of the
right hemisphere.
Task performance It appeared that abacus experts activated both hemi-
spheresregardlessoftheversionsofmental-operationtasks
For the numeral mental-operation task, accuracy was sta- (see Fig. 2A). For a more quantitative analysis, laterality
tistically not different between the experts during six-digit indexwascalculatedfortask-relatedactivitychangesinthe
numberaddition(87%correct)andthenonexpertsduringsin- SPcS and IPS regions, using parameter estimates from the
gle-digitnumberaddition(77%correct,P(cid:3)0.36byUtest). first-level multiregression analysis in each individual (see
However, the abacus experts during the numeral task for the legend to Fig. 2B). Nonparametric statistics revealed that
other number sizes (100% correct for both single- and three- the laterality of activity was different between the two
digit numbers) demonstrated significantly superior perfor- groups only during the numeral mental-operation task (U
mance in task accuracy over the nonexperts (P (cid:3) 0.01). Ac- test,P(cid:3)0.039forSPcSandP(cid:3)0.029forIPS)(Fig.2B).
curacy was comparable between the two groups both for the
spatial mental-operation task (experts, 97% correct; controls, Brain activity during mental calculation: effect of number
90% correct, P (cid:3) 0.44) and for the verbal mental-operation size for abacus experts
task (experts, 94% correct; controls, 91% correct, P (cid:3) 0.61).
Constantvisualinspectionduringscanningdidnotdetectany We investigated the effect of number size on mental
movement during the task periods, except for that necessary abacus operations (Fig. 3). Left IPS and posterior superior
forthebehavioralreport. parietal cortex demonstrated significant activity during the
300 T.Hanakawaetal./NeuroImage19(2003)296–307
Table1
Activityduringmentaloperationtasksasrevealedbyconjointanalysisacrosssubjectsineachgroup
Regions(Brodmannarea) Clustersize Coordinates Z Pcorrected
value
x y z
A.Numeralmental-operationtask
Abacusexperts
1. Lprecuneus(7) 1311 (cid:4)18 (cid:4)66 60 Infinite 0.000
Lintraparietalsulcus(40/7) (cid:4)34 (cid:4)48 54 Infinite 0.000
2. Rintraparietalsulcus(40/7) 363 42 (cid:4)52 56 Infinite 0.000
3. Rprecuneus(7) 60 14 (cid:4)66 64 Infinite 0.000
4. Rsuperiorprecentralsulcus(6) 103 24 (cid:4)6 54 Infinite 0.000
5. Lfrontaloperculum(6) 73 (cid:4)48 6 28 6.89 0.000
6. Lsuperiorprecentralsulcus(6) 34 (cid:4)32 (cid:4)6 52 6.55 0.000
7. Lcerebellarhemisphere 14 (cid:4)36 (cid:4)74 (cid:4)28 6.28 0.000
8. Cerebellarvermis 11 6 (cid:4)76 (cid:4)30 6.00 0.000
9. Rfrontaloperculum(6) 13 50 10 36 5.99 0.000
10. Lfusiformgyrus(37) 8 (cid:4)52 (cid:4)62 (cid:4)16 5.88 0.000
11. Rcerebellarhemisphere 10 40 (cid:4)74 (cid:4)24 5.83 0.000
Nonexperts
1. Lprecuneus(7) 810 (cid:4)24 (cid:4)62 56 Infinite 0.000
Llateralparietalarea(40) (cid:4)46 (cid:4)40 54 Infinite 0.000
Lintraparietalsulcus(40/7) (cid:4)40 (cid:4)48 58 Infinite 0.000
2. Rcerebellarhemisphere 677 30 (cid:4)63 (cid:4)32 Infinite 0.000
3. Lsuperiorprecentralsulcus(6) 704 (cid:4)32 0 54 Infinite 0.000
LBroca’sarea(44) (cid:4)50 10 26 Infinite 0.000
4. Lcerebellarhemisphere 211 (cid:4)38 (cid:4)62 (cid:4)30 Infinite 0.000
5. Ranteriorcingulatecortex(24) 191 4 14 45 Infinite 0.000
Lmedialfrontalgyrus(6) (cid:4)2 1 61 7.31 0.000
6. Rsuperiorprecentralsulcus(6) 86 34 0 54 Infinite 0.000
7. Lprefrontalcortex(46) 97 (cid:4)52 34 16 Infinite 0.000
8. Lfusiformgyrus(37/19) 19 (cid:4)48 (cid:4)58 (cid:4)12 6.91 0.000
B.Spatialmental-operationtask
Abacusexperts
1. Rintraparietalsulcus(40/7) 933 40 (cid:4)46 44 Infinite 0.000
2. Rprecuneus(7) 470 14 (cid:4)64 62 Infinite 0.000
3. Lprecuneus(7) 1668 (cid:4)12 (cid:4)70 58 Infinite 0.000
Lintraparietalsulcus(40/7) (cid:4)38 (cid:4)48 56 Infinite 0.000
4. Lsuperiorprecentralsulcus(6) 223 (cid:4)32 (cid:4)8 52 Infinite 0.000
5. Rsuperiorprecentralsulcus(6) 89 28 (cid:4)4 62 Infinite 0.000
6. Rfrontaloperculum(6) 21 56 8 32 6.68 0.000
7. Lmiddletemporalgyrus(39) 43 38 (cid:4)76 28 6.50 0.000
8. Rmiddletemporalgyrus(37) 5 (cid:4)46 (cid:4)62 4 6.05 0.000
9. Rmiddleoccipitalgyrus(19) 6 52 (cid:4)66 (cid:4)8 5.90 0.000
Nonexperts
1. Lprecuneus(7) 639 (cid:4)16 (cid:4)70 60 Infinite 0.000
Lintraparietalsulcus(40/7) (cid:4)42 (cid:4)46 60 Infinite 0.000
2. Rsuperiorprecentralsulcus(6) 334 32 0 52 Infinite 0.000
3. Lsuperiorprecentralsulcus(6) 701 (cid:4)28 (cid:4)6 50 Infinite 0.000
4. Rintraparietalsulcus(40/7) 527 42 (cid:4)42 56 Infinite 0.000
Rprecuneus(7) 20 (cid:4)64 56 Infinite 0.000
5. Lmiddleoccipitalgyrus(19) 113 (cid:4)52 (cid:4)72 (cid:4)10 Infinite 0.000
6. Rcerebellarhemisphere 55 40 (cid:4)48 (cid:4)38 Infinite 0.000
7. Rsupramarginalgyrus(40) 15 56 (cid:4)24 36 6.69 0.000
8. Lfrontaloperculum(6) 48 (cid:4)56 8 32 6.51 0.000
9. Lsupramarginalgyrus(40) 10 (cid:4)60 (cid:4)26 34 6.47 0.000
10. Anteriorcingulatecortex(24) 17 0 10 50 5.89 0.000
numeral mental-operation task with every number size sphere. There was a number-size-dependent, monotonous
tested. The task on larger number size gradually induced increase in activity in IPS (x, y, z (cid:3) (cid:4)34, (cid:4)52, 56 on the
increasedactivityinadditionalbrainareas,inwhichactivity left; x, y, z (cid:3) 40, (cid:4)58, 56 on the right), posterior superior
was otherwise subthreshold, especially in the right hemi- parietal cortex/precuneus (x, y, z (cid:3) (cid:4)18, (cid:4)66, 60 on the
T.Hanakawaetal./NeuroImage19(2003)296–307 301
Table1(continued)
Regions(Brodmannarea) Clustersize Coordinates Zvalue Pcorrected
x y z
C.Verbalmental-operationtask
Abacusexperts
1. Llateralparietalarea(40) 806 (cid:4)52 (cid:4)46 48 Infinite 0.000
Lintraparietalsulcus(40/7) (cid:4)38 (cid:4)52 50 Infinite 0.000
2. Rlateralparietalarea(40) 367 54 (cid:4)40 50 7.80 0.000
Rintraparietalsulcus(40/7) 42 (cid:4)46 48 7.13 0.000
3. Lfrontaloperculum(6) 87 (cid:4)48 (cid:4)2 36 7.50 0.000
4. Rsuperiorprecentralsulcus(6) 38 24 2 58 6.73 0.000
5. Ranteriorcingulatecortex(6) 61 4 10 54 6.55 0.000
6. LBroca’sarea(44) 29 (cid:4)48 10 38 6.36 0.000
7. Rintraparietalsulcus(40/7) 28 34 (cid:4)62 46 6.27 0.000
8. Lsuperiorprecentralsulcus(6) 15 (cid:4)30 (cid:4)12 48 6.19 0.000
9. Lmedialfrontalgyrus(6) 6 (cid:4)2 2 62 6.04 0.001
10. Lprecuneus(7) 16 (cid:4)18 (cid:4)64 50 6.02 0.001
Nonexperts
1. Lmedialfrontalcortex(6) 191 (cid:4)2 0 66 Infinite 0.000
2. Llateralparietalarea(40) 207 (cid:4)50 (cid:4)42 50 7.74 0.000
3. Lsuperiorprecentralsulcus(6) 86 (cid:4)44 0 54 7.53 0.000
4. Lintraparietalsulcus(40/7) 117 (cid:4)28 (cid:4)62 56 7.43 0.000
5. Rcerebellarhemisphere 67 32 (cid:4)64 (cid:4)30 7.27 0.000
left; x, y, z (cid:3) 14, (cid:4)66, 64 on the right), left frontal oper- Discussion
culum(x,y,z(cid:3)(cid:4)48,6,28),SPcS(x,y,z(cid:3)(cid:4)28,(cid:4)4,46
on the left; x, y, z (cid:3) 24, (cid:4)4, 50 on the right), and left General discussion
fusiform gyrus (x, y, z (cid:3) (cid:4)46, (cid:4)62, (cid:4)18).
The present study, for the first time to our knowledge,
Brain activity during mental-operation tasks: reported the neural correlates during mental calculation in
between-group comparisons abacusexperts.Themostsignificantfindingwasthattheleft
posteriorsuperiorparietalcortex/precuneus,wherecontrols
Whencalculation-relatedactivitywascomparedbetween
also revealed some calculation-related activity, showed
the two groups, only the left posterior superior parietal
muchenhancedactivityduringmentalarithmeticofabacus
cortex/precuneus showed significantly greater activity for
experts. Additionally, there was no significant between-
abacusexpertsthancontrols(Fig.4A,Table2).Ontheother
group difference in activity of this area for the two control
hand, no hyperactive area was detected for the controls
tasks. We tested between-group differences using a statis-
during the numeral mental-operation task.
tical method that allowed population-level inference; this
When activity during the spatial mental-operation task
analysis gave a significant result despite a relatively small
was compared, the expert group revealed hyperactivity in
number of subjects for this type of analysis, probably be-
the right IPS while the control group did not show any
cause of the prominent difference in this area. We also
hyperactivityovertheexpertgroup(Table2).Fortheverbal
found more involvement of right SPcS and IPS during
mental-operation task, there was no significant between-
mentalcalculationinabacusexperts,althoughthebetween-
group difference.
groupdifferencewassignificantonlyinthelateralityindex
In the left posterior superior parietal cortex/precuneus,
analysis. These areas, the posterior superior parietal cortex
brainactivityexhibitedsignificantgroup-by-taskinteraction
(P(cid:3)0.007,repeated-measuresANOVAwithGreenhouse– and right frontoparietal areas, have been often indicated as
neural substrates subserving visuospatial/visuomotor pro-
Geisser correction), reflecting overactivity for experts dur-
ingthenumeralmental-operationtask.AconfirmatoryStu- cessing. Hence, our main conclusion is that mental abacus
dentttestshowedthattheposteriorsuperiorparietalactivity operation is associated with enhanced activity in the areas
wasdifferentonlyduringthenumeralmental-operationtask relevant for visuospatial/visuomotor information process-
betweenthetwogroups(Fig.4B).Awithin-groupANOVA ing, which is consistent with our a priori hypothesis. One,
followedbyaposteriortest(Student–Newman–Keuls,P(cid:6) however, needs caution to conclude that this visuospatial
0.05) showed significantly greater activity during the nu- processing during mental calculation is truly exclusive for
meral task over the verbal task for abacus experts and abacus experts because activity in those visuospatial areas
greateractivityduringthenumeralandspatialtasksoverthe hasalsobeenobservedfornonexpertsduringmentalcalcu-
verbal task for nonexperts. lation. We therefore consider that the difference in brain
302 T.Hanakawaetal./NeuroImage19(2003)296–307
Fig.2.Generalpatternofactivityduringeachmental-operationtask.(A)Activityduringthethreemental-operationtasksrelativetoavisualfixationtask
for abacus experts and nonexperts (within-group conjunction analysis, P (cid:6) 0.05 corrected for multiple comparisons), rendered onto a standard brain.
Significantactivityonlyforabacusexpertsisshowninred,activityonlyfornonexpertsingreen,andthespatialoverlapofactivityinyellow.(B)Laterality
ofactivityforthesuperiorprecentralsulcus(SPcS)andintraparietalsulcus(IPS),wherethethreemental-operationtaskscommonlyinducedactivity.Aliberal
thresholdofP(cid:6)0.001(uncorrected)wasusedonlytolocatetheseactivitiesineachindividual;allsubjectsshowedsuprathresholdactivityinallfourregions
at this threshold. Estimated mean signal increase (SI, arbitrary unit) at the most significantly activated single voxel was computed in each area in each
individualandthenwasusedtocalculatealateralityindex(LI),thatisSIforleftminusSIforright,dividedbythesumofSIforbothsides.Themore
involvementoftherighthemispherewasevidentforexpertscomparedwithnonexpertsonlyduringthenumeralmental-operationtask(*P(cid:3)0.039forSPcS
and*P(cid:3)0.029forIPSbyUtest).
activity between the two groups primarily reflects different of motor responses such as verbalization) can be removed
weight on calculation strategy employed in each group: from the tasks, by subtracting out activity evoked by a
more involvement of visuospatial strategy for abacus ex- control sensorimotor task. For example, more complex
perts and more involvement of linguistic strategy for non- arithmetic problems usually result in more complex solu-
experts. tions to verbalize or retrieve from working memory; there-
The areas in which we found enhanced activity for aba- fore, the complex vs simple arithmetic contrast reflects
cusexpertsaredifferentfromtheareaswherePesentietal. neural activity not only for complex computation but also
found increased activity for a prodigy of nonabacus calcu- for complex responses, especially when a task includes
lation (Pesenti et al., 2001). There are several possible multipleresponses.Ontheotherhand,onemustbecautious
sources that can explain this discrepancy. First, we exam- about how to interpret the present results, because the all
ined six experts who had gone through similar training mental-operationtasksinvolvesomecomputation,eitherin
processesforyears,asopposedtooneprodigy.Second,itis 1dimension(numeralandverbal)or2dimensions(spatial).
possible that the discrepancy truly reflects a difference in Another issue in the experimental design is how to con-
calculation strategy, since the well-trained subject studied trol task difficulty between experts and nonexperts; an ap-
byPesentietal.haddevelopedanoriginalcalculationstrat- propriate calculation task for experts would be too difficult
egy,whichstronglyreliedonepisodicmemory.Third,there for nonexperts. The present task design forced subjects to
are differences in the behavioral tasks between the two solve problems at the same speed between the two groups.
studies.Weusedmental-operationtasksthatdidnotrequire A single variable for the numeral task (i.e., number size)
any motor response during the task periods (Hanakawa et wasmodulatedfortheexpertgrouptoexperiencethesame
al., 2002; Sawamoto et al., 2002). Cognitive tasks usually difficulty as for the nonexperts group. A comparison was
requiremultipleresponsesforbehavioralreports.However, thenmadefortheconditionsinwhichaccuracywassimilar
it is not entirely reasonable to assume that motor compo- between the two groups, assuming that accuracy would
nents of cognitive tasks (i.e., preparation for and execution reflect subjective difficulty of the tasks. The present exper-
T.Hanakawaetal./NeuroImage19(2003)296–307 303
Fig.3.Numbersizeeffectduringmentalcalculationofabacusexperts.(A)Activityduringthenumeralmental-operationtaskofabacusexpertsshownfor
each number size (within-group conjunction analysis, P (cid:6) 0.05 corrected for multiple comparisons). Abacus experts exhibited activity that increased
monotonouslyinassociationwithnumbersizeofthecalculationproblems.Thistypeofactivitywaspresentinthebilateralsuperiorprecentralsulcus(SPcS),
leftfrontaloperculum(FO),bilateralintraparietalsulcus(IPS),bilateralposteriorsuperiorparietalcortex(PSPC),andleftfusiformgyrus(FG).Activityis
shownonasurface-renderedimage(left)andacoronalsection(right)oftheanatomicMRIaveragedoversixabacusexperts.(B)Themeaneffectsizein
arbitraryunit(verticalaxis)isplottedagainstthenumberofdigitsinproblems(horizontalaxis).Errorbarsshowstandarderrorofthemean.
imentaldesignalsoprovidedanopportunitytoexaminethe abacus operation, yet the direct comparisons of the task-
effect of number size on mental abacus operation. The related activity between the two groups did not yield sig-
results revealed the brain areas in which activity was in- nificant differences in these areas. This discrepancy re-
creased monotonously as number size in arithmetic prob- flectedthefactthatnonexpertsalsoshowedbrainactivityin
lems was increased. To our knowledge, only one has re- the right frontoparietal areas during the numeral mental-
ported number size effect in mental calculation, although operation task (see legend to Fig. 2). Nevertheless, the
the comparison was very limited (problems with operands observationofrightfrontoparietaloveractivityisconsistent
ranged from 1 to 5 versus operands ranged from 5 to 9) with a study claiming the relative significance of right
(Stanescu-Cossonetal.,2000),whichrevealednumbersize hemisphere for mental abacus operation (Hatta and Ikeda,
effectinthefrontaloperculumandthemiddlepotionofIPS. 1988). The participation of the right hemisphere supports a
In addition to these areas shown in calculation nonexperts, role of visuospatial processing in mental abacus operation
abacus experts in the present study revealed remarkable since there is evidence that spatial information is predomi-
numbersizeeffectintheSPcSregionandposteriorsuperior nately processed in the right hemisphere (Kohler et al.,
parietal cortex/precuneus. This finding underscores signifi- 1998).Veryrecently,Formisanoandcolleagues(Formisano
cance of these two areas in mental abacus operation. et al., 2002) reported neuroimaging findings that indicate
The semiquantitative, laterality index analysis for the functional segregation between the left and right parietal
threemental-operationtaskssuggestedmoreinvolvementof cortices. They suggested that the left parietal cortex would
therightfrontoparietalareas(SPcSandIPSareas)inmental beinvolvedinthegenerationofspatialmentalimageswhile
304 T.Hanakawaetal./NeuroImage19(2003)296–307
hypothesis that neural substrates for visuospatial informa-
tion processing including spatial working memory (Tanaka
et al., 2002) underlie mental calculation ability of abacus
experts. The results also suggested that these visuospatial
areas might play some roles in standard mental calculation
strategy (Simon, 1999), which could explain the mild ac-
tivity in the right frontoparietal areas observed for nonex-
perts.
Parietal cortex
The parietal cortex has been the primary focus of atten-
tion as neural substrates underlying calculation because
impairmentoftheparietalcortex,especiallyontheleft,can
cause dyscalculia or anarithmetia (Kahn and Whitaker,
1991). More recently, growing evidence has suggested that
the parietal cortex is composed of multiple anatomofunc-
tional subdivisions, and impairment of each subdivision
may relate to dysfunction of a specific component of cal-
culationability.Forexample,alesionintheleftperisylvian
part of the inferior parietal cortex can impair the ability to
manipulatenumbersinaverbalformatbutsparetheability
to manipulate nonverbal number representations, possibly
representing “verbal anarithmetia” (Cohen et al., 2000a).
The lateral parietal area shown for the verbal task for both
Fig. 4. Between-group comparisons. (A) Left posterior superior parietal experts and nonexperts is consistent with this perisylvian
cortex showing calculation-related overactivity for the abacus experts is
parietal area. This is reasonable because the verbal mental-
projected onto a surface-rendered MRI. (B) Estimated mean effect size
operation task supposedly involves more verbal aspects of
(arbitraryunit)foreachmental-operationtaskintheleftposteriorsuperior
parietalcortex.Onlythenumeralmental-operationtaskinducedasignif- mathematic operation than nonverbal aspects.
icantlydifferentbrainactivityinthisareabetweenthetwogroups. Lesions in more dorsal and caudal parts of the parietal
cortex, including the posterior angular gyrus and IPS, can
cause a more general deficit in number processing or se-
the right parietal cortex would be engaged in the manipu- mantic aspects of quantity processing (Takayama et al.,
lationandcomparisonofsuchimages.Thisproposedroleof 1994), possibly representing “semantic anarithmetia” (Co-
the right parietal cortex is consistent with the prominent hen et al., 2000a). Neuroimaging experiments have sup-
right parietal activity during mental abacus operation as ported the significance of this area for number quantity
well as the spatial mental-operation task for both groups. manipulation(Dehaeneetal.,1999;Stanescu-Cossonetal.,
Also, there is abundant evidence indicating that the poste- 2000; Pinel et al., 2001). In the present study, both experts
rior superior parietal cortex/precuneus, the most empha- andnonexpertsrevealedmarkedactivityinareassurround-
sized neural substrate for mental abacus here, is important ing the left IPS, supporting the general importance of this
for visuospatial processing as discussed below. Therefore, area in quantity manipulation regardless of arithmetic ex-
the results from the present study generally confirmed our pertise (Pesenti et al., 2001). By contrast, the right IPS is
Table2
Between-groupcomparisons
Regions(Brodmannarea) Clustersize Coordinates Zvalue Pcorrected
x y z
Numeralmental-operationtask
Experts-nonexperts
Lposteriorsuperiorparietalcortex(7) 12 (cid:4)22 (cid:4)62 58 5.19 0.041
Spatialmental-operationtask
Experts-nonexperts
Rintraparietalsulcus(40) 12 40 (cid:4)48 44 5.38 0.036
Note.OtherpairsofcomparisonsdidnotshowanysignificantdifferenceinactivitybetweenthetwogroupswithathresholdofP(cid:6)0.05correctedfor
multiplecomparisons.
T.Hanakawaetal./NeuroImage19(2003)296–307 305
important in the operation of visual images (Formisano et withotherbrainareasdedicatedtovisuospatialinformation
al., 2002), which is very likely essential for mental abacus processing. The numeral mental-operation task for abacus
operation. experts and spatial task for both groups induced symmetri-
The posterior superior parietal cortex/precuneus was calbilateralactivationoftheSPcSregion.TheSPcSregion
moreactiveduringthenumeraltaskforabacusexpertsthan also revealed number size effect for mental abacus opera-
nonexperts. This zone also exhibited marked number size tions. Importantly, the parietal area V6a has an anatomic
effect for abacus experts, which has not been reported for connection with the dorsal premotor cortex (F2 and F7)
nonexperts (Stanescu-Cosson et al., 2000). Although activ- (Matelli et al., 1998). The SPcS region probably corre-
ityintheposteriorsuperiorparietalcortex/precuneuscanbe sponds to F7, the rostral part of the dorsal premotor cortex
observedformentalcalculationinnonexperts,thisareahas (PMdr) (Geyer et al., 2000), or “pre-PMd” (Picard and
seldom been a focus of attention as the underlying neural Strick,2001).Thisareaisdistinctfromthefrontaleyefields
resources for mental calculation. Given that all mental- occupying the ventrolateral part of the SPcS (Courtney et
operationtasksinvolvemathematicaloperations,thediffer- al.,1998;Hanakawaetal.,2002);hence,itislesslikelythat
entialinvolvementofthisareamaygiveusahinttoclarify the whole SPcS activity for abacus arithmetic entirely re-
a functional role of this area. This area was also highly sulted from eye movements. The dorsal premotor cortex is
active during the spatial mental-operation task for both of important in motor imagery (Gerardin et al., 2000;
the groups but not during the verbal mental-operation task. Hanakawa et al., 2003) as well as visuomotor association
One hypothesis is that the posterior superior parietal (Wise et al., 1997). Furthermore, along with the posterior
cortexplaysasignificantpartincomputationaloperationin superiorparietalcortex,PMdrhasbeenimplicatedinspatial
2- and higher-dimensional space (i.e., vector-type opera- information processing, spatial working memory, or visuo-
tion).Itisverylikelythatoneneedsavector-ormatrix-type spatial imagery not necessarily associated with movement
of computational operations to work on a putative visual (Mellet et al., 1996; Courtney et al., 1998; Richter et al.,
formofmentalabacusthatmustberepresentedin2-dimen- 2000). To summarize, the SPcS and posterior superior pa-
sional space. This idea can explain high activity during the rietalregionsseemtoconjointlyfunctioninbothnonmotor
spatial mental-operation task that also requires computa- mentaloperationsandvisuomotorcontrol.Thismayexplain
tional operations in 2-dimensional space as opposed to the the fact that simultaneous performance of motor tasks in-
verbaltaskthatprimarilyrequiresascalar-orsequence-type terferes with mental abacus operations (Stigler, 1984),
of operations in 1-dimensional space. This hypothesis is whichmaymostvividlycharacterizethedifferencebetween
consistent with evidence that the posterior superior parietal abacus and nonabacus calculation strategies.
cortex is dedicated to visuospatial information processing, The left prefrontal cortex, frontal operculum including
which primarily deals with 2- or higher-dimensional space Broca’s area, and pre-SMA appeared to be more active
(Melletetal.,1996,1998;Simonetal.,2002).Judgingfrom duringthenumeraltaskfornonexpertsthanforexperts;yet,
this property and anatomic location, this parietal subdivi- there was no significant population-level difference there
sionmightcorrespondtotheparietalareaV6ainnonhuman possibly because of low statistical power. The part of the
primates(RosaandTweedale,2001).Itissuggestedthat,as lateral frontal area is suggested to play a role in nonabacus
opposedtoobject-motiondetectedbyV5,V6afunctionsto mental arithmetic (Burbaud et al., 1995; Rickard et al.,
detect self-motion. This is consistent with an a priori hy- 2000). Also, there is a patient report of acalculia after a
pothesis that enhanced motor imagery subserves mental lesion involving lateral frontal cortex (Tohgi et al., 1995).
arithmetic for abacus experts. Area V6a also has a putative Medial frontal areas including the pre-SMA and anterior
role in directing skeletomotor activity to extrapersonal cingulatecortexarealsosuggestedtohavearoleformental
space, which again requires vector-type computation. calculation(LucchelliandDeRenzi,1993;Hanakawaetal.,
We propose a hypothesis that multiple parietal cortical 2001).
modulesformentalcalculationareorganizedinsuchaway
thattheanterolateralparietalareasprimarilyconcernverbal
Other areas
aspects of numerals with emphasis on sequence-type of
operations, the middle parietal areas concern general con-
Although there was no overactivity in this area for
ceptual aspects of calculation, and the posteromedial pari-
abacus experts over nonexperts, we found number size
etal areas concern vector-type operation in 2- and higher-
effect in the left fusiform cortex. The parietal cortex
dimensional space.
functionally interconnects with the fusiform gyrus, part
of the ventral visual pathway (Buchel et al., 1999). The
Frontal cortex fusiform gyrus is shown to represent visual word form
(Cohen et al., 2000b). One interesting question regards
It is less likely that only the parietal cortex plays a role whether this area also represents visual form of an aba-
inmentalcalculationofabacusexperts.Weratherthinkthat cus, which will need future testing using nonvisual stim-
the posterior superior parietal cortex functions in concert uli for behavioral tasks.