15.08.2013 Views

Dipole Modelling in Cognitive Neuroscience - Low Temperature ...

Dipole Modelling in Cognitive Neuroscience - Low Temperature ...

Dipole Modelling in Cognitive Neuroscience - Low Temperature ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Dipole</strong> <strong>Modell<strong>in</strong>g</strong><br />

<strong>in</strong> <strong>Cognitive</strong> <strong>Neuroscience</strong><br />

Riitta Salmel<strong>in</strong><br />

Bra<strong>in</strong> Research Unit<br />

<strong>Low</strong> <strong>Temperature</strong> Laboratory<br />

Aalto University<br />

Espoo, F<strong>in</strong>land<br />

riitta@neuro.hut.fi


<strong>Dipole</strong> <strong>Modell<strong>in</strong>g</strong><br />

<strong>in</strong> <strong>Cognitive</strong> <strong>Neuroscience</strong><br />

What type of questions?<br />

RS/Apr10


Example: Dual-Route Model of Read<strong>in</strong>g<br />

FAMILIAR<br />

WORDS<br />

Neural markers...?<br />

UNFAMILIAR<br />

WORDS<br />

adapted from<br />

Coltheart et al. Psychol Rev 1993


Example: Dual-Route Model of Read<strong>in</strong>g<br />

TALO<br />

LAUTANEN<br />

ROKI<br />

SOIJINTO<br />

short word (F<strong>in</strong>nish)<br />

long word (F<strong>in</strong>nish)<br />

short nonword<br />

long nonword<br />

Lexical route dom<strong>in</strong>ates,<br />

length has little effect<br />

Grapheme-to-phoneme,<br />

length has strong effect<br />

100 different stimuli of each type, <strong>in</strong> a randomized order<br />

Silent read<strong>in</strong>g, prompt to read aloud (“?”) <strong>in</strong> 4% of trials<br />

Wydell et al. JoCN 2003


Sensor Signals Are Informative<br />

... Especially Planar Gradiometers<br />

Short words<br />

Long words<br />

Short nonwords<br />

Long nonwords<br />

fT/cm<br />

100<br />

0<br />

-200...800 ms<br />

0 800 ms<br />

N=1<br />

Current<br />

flow<br />

Wydell et al. JoCN 2003


<strong>Dipole</strong> <strong>Modell<strong>in</strong>g</strong><br />

<strong>in</strong> <strong>Cognitive</strong> <strong>Neuroscience</strong><br />

What type of questions?<br />

Why source analysis?<br />

RS/Apr10


Work<strong>in</strong>g at the Source Level Is Useful<br />

because...<br />

Decomposition of the sensor-level signals <strong>in</strong>to their<br />

approximate cortical source areas can markedly<br />

clarify the stimulus/task effects<br />

It is useful to have a general idea of the bra<strong>in</strong> areas<br />

<strong>in</strong>volved <strong>in</strong> the task performance<br />

--- BUT ---<br />

Do not th<strong>in</strong>k primarily <strong>in</strong> terms of location!<br />

RS/Apr10


Source Analysis Models the Source...<br />

Equivalent Current <strong>Dipole</strong><br />

ECD<br />

Activation<br />

strength<br />

(nAm)<br />

20<br />

0<br />

0 1000<br />

Time (ms)<br />

Physiologically & physically sound<br />

M<strong>in</strong>imum number of assumptions<br />

Benefits from expertise<br />

Conta<strong>in</strong>s an element of subjectivity<br />

RS/Apr10


Source Analysis Models the Source<br />

Same data & tool -> same map<br />

Subjective ROIs, thresholds<br />

Benefits from expertise<br />

or Limits the Solution<br />

Statistical test<strong>in</strong>g on ROI time<br />

courses, not maps (MEG≠fMRI)<br />

M<strong>in</strong>imum Norm Estimate<br />

MNE<br />

RS/JV/Apr10


Source Analysis Models the Source<br />

or Limits the Solution<br />

© Mika Seppä<br />

Appearance is determ<strong>in</strong>ed by<br />

the choice of analysis method,<br />

not by the structure of the active<br />

bra<strong>in</strong> areas<br />

MEG (or EEG) gives an estimate<br />

of the centre of an active area but<br />

(<strong>in</strong> typical record<strong>in</strong>gs) no direct<br />

<strong>in</strong>formation of its spatial extent<br />

RS/Apr10


Check for Artefacts at the Sensor Level<br />

Silent picture nam<strong>in</strong>g<br />

(N=1)<br />

Set 1<br />

Set 2<br />

Set 3<br />

100 fT/cm<br />

-200...1000 ms<br />

Elekta Neuromag<br />

Current<br />

flow<br />

RS/Apr10


Check for Artefacts at the Sensor Level<br />

Overt picture nam<strong>in</strong>g<br />

(N=1)<br />

Set 1<br />

Set 2<br />

Set 3<br />

100 fT/cm<br />

-200...1000 ms<br />

Elekta Neuromag<br />

Current<br />

flow<br />

RS/Apr10


Field Patterns Are Even More Tell<strong>in</strong>g<br />

Neural<br />

activation<br />

... with<br />

bl<strong>in</strong>k<br />

... with<br />

mouth movement<br />

RS/Apr10


<strong>Dipole</strong> <strong>Modell<strong>in</strong>g</strong><br />

<strong>in</strong> <strong>Cognitive</strong> <strong>Neuroscience</strong><br />

What type of questions?<br />

Why source analysis?<br />

Why dipole modell<strong>in</strong>g? How?<br />

RS/Apr10


In Source Analysis, Probably Correct<br />

Is Better than Reproducibly Wrong<br />

Measured<br />

bl<strong>in</strong>k<br />

ECD MNE<br />

artefact!<br />

...cortex?<br />

RS/JV/Apr10


ECD Solutions Are Mostly Obvious<br />

Auditory<br />

evoked<br />

responses<br />

(N=1)<br />

N100m<br />

Elekta Neuromag<br />

Current<br />

flow<br />

RS/Apr10


ECD Solutions Are Mostly Obvious<br />

60 ms 85 ms 150 ms<br />

0 100 200<br />

Time (ms)<br />

0 100 200 0 100 200<br />

Salmel<strong>in</strong> <strong>in</strong> “MEG. An Introduction to Methods”, OUP 2010


ECD Solutions Are Mostly Obvious<br />

N100m<br />

P200m<br />

Left<br />

Right<br />

Left<br />

Right<br />

80%<br />

g<br />

10 nAm<br />

0 100 200<br />

Time (ms)<br />

Salmel<strong>in</strong> <strong>in</strong> “MEG. An Introduction to Methods”, OUP 2010


ECD Solutions Are Mostly Obvious<br />

Right median nerve<br />

stimulation<br />

(N=1)<br />

100 fT/cm<br />

-50...250 ms<br />

Elekta Neuromag<br />

Current<br />

flow<br />

RS/Apr10


ECD Solutions Are Mostly Obvious<br />

20 ms 30 ms 50 ms<br />

75 ms 95 ms 105 ms<br />

Salmel<strong>in</strong> <strong>in</strong> “MEG. An Introduction to Methods”, OUP 2010


There Is No Need to Decide the<br />

Number of ECDs <strong>in</strong> Advance<br />

20 ms 30 ms 50 ms<br />

75 ms 95 ms 105 ms<br />

Salmel<strong>in</strong> <strong>in</strong> “MEG. An Introduction to Methods”, OUP 2010


There Is No Need to Decide the<br />

Number of ECDs <strong>in</strong> Advance<br />

SI<br />

PPC<br />

SII c<br />

SII i<br />

0 100 200<br />

Time (ms)<br />

20 nAm<br />

RS/Apr10


ECD Orientation Is a Useful Measure<br />

Spoken words, sentences etc.<br />

Allows to separate<br />

different functional<br />

responses<br />

0 400 800<br />

0 400 800<br />

Time (ms)<br />

Time (ms)<br />

10 nAm<br />

0 400 800<br />

0 400 800<br />

Biermann-Ruben et al. Neuroimage 2005<br />

Uusvuori et al. Cereb Cortex 2008<br />

Bonte et al. Cereb Cortex 2006


Dipolar Field Patterns Are Also<br />

Fully Identifiable <strong>in</strong> <strong>Cognitive</strong> Data<br />

Short words<br />

Long words<br />

Short nonwords<br />

Long nonwords<br />

fT/cm<br />

100<br />

0<br />

0 800 ms<br />

Current<br />

flow<br />

Wydell et al. JoCN 2003


fT/cm<br />

100<br />

0<br />

100<br />

0<br />

Dipolar Field Patterns Are Also<br />

Fully Identifiable <strong>in</strong> <strong>Cognitive</strong> Data<br />

0 800 ms<br />

0 800 ms<br />

Short<br />

words<br />

Long<br />

words<br />

Short<br />

nonwords<br />

Use the same ECD for all conditions<br />

Long<br />

nonwords<br />

Salmel<strong>in</strong> <strong>in</strong> “MEG. An Introduction to Methods”, OUP 2010


ECDs Allow Quantitative Comparison<br />

between Conditions<br />

< 200 ms<br />

200-400 ms<br />

> 400 ms<br />

Group-level cluster<strong>in</strong>g of the sources by location and tim<strong>in</strong>g<br />

... and by function:<br />

Long nonwords ><br />

short real words<br />

< 200 ms<br />

> 200 ms<br />

Wydell et al. JoCN 2003


ECDs Allow Quantitative Comparison<br />

40<br />

20<br />

0 200<br />

Time (ms)<br />

between Conditions<br />

Activation<br />

strength<br />

(nAm)<br />

40<br />

20<br />

0 200 400 600<br />

Time (ms)<br />

Short words<br />

Long words<br />

Short nonwords<br />

Long nonwords<br />

Wydell et al. JoCN 2003


ECDs Allow Quantitative Comparison<br />

nAm<br />

40<br />

20<br />

between Conditions<br />

Strength Duration<br />

ms<br />

200<br />

100<br />

Short words<br />

Long words<br />

Short nonwords<br />

Long nonwords<br />

Letter-str<strong>in</strong>g length < 200 ms <strong>in</strong> occipital cx (Tarkia<strong>in</strong>en et al. Bra<strong>in</strong> 1999)<br />

Length x lexicality > 250 ms <strong>in</strong> the left superior temporal cx<br />

... phonological analysis if we accept the dual-route model<br />

Wydell et al. JoCN 2003


<strong>Dipole</strong> <strong>Modell<strong>in</strong>g</strong><br />

<strong>in</strong> <strong>Cognitive</strong> <strong>Neuroscience</strong><br />

What type of questions?<br />

Why source analysis?<br />

Why dipole modell<strong>in</strong>g? How?<br />

Are the results reliable?<br />

RS/Apr10


Source-Level F<strong>in</strong>d<strong>in</strong>gs Must Make Sense<br />

Short w<br />

Long w<br />

Short nw<br />

Long nw<br />

Current flow<br />

<strong>in</strong> Light of Sensor-Level Effects<br />

0 800 ms<br />

9<br />

5<br />

1 4<br />

6 2<br />

8<br />

7<br />

3<br />

1<br />

2<br />

3<br />

4<br />

5<br />

6<br />

7<br />

8<br />

9<br />

0 400 ms<br />

RS/Apr10


Good-Quality Data Yields Reliable Neural<br />

Time Courses with Multiple Methods<br />

Word read<strong>in</strong>g (N=15)<br />

ECD<br />

MNE<br />

~100 ms ~150 ms ~300-400 ms<br />

Vartia<strong>in</strong>en et al., submitted


<strong>Dipole</strong> <strong>Modell<strong>in</strong>g</strong><br />

<strong>in</strong> <strong>Cognitive</strong> <strong>Neuroscience</strong><br />

What type of questions?<br />

Why source analysis?<br />

Why dipole modell<strong>in</strong>g? How?<br />

Are the results reliable?<br />

Evoked data but also oscillatory activity?<br />

RS/Apr10


ECDs Can Be Used<br />

to Localize Cortical Rhythms<br />

Time (s)<br />

Frequency<br />

spectra<br />

300 fT/cm<br />

0 20 40 Hz<br />

0 2 4 6 8<br />

Rest<br />

Motion<br />

Eyes<br />

closed<br />

Eyes<br />

open<br />

Salmel<strong>in</strong> & Hari EEG J 1994<br />

Hari & Salmel<strong>in</strong> TiNS 1997


ECDs Can Be Used<br />

to Localize Cortical Rhythms<br />

10 Hz 20 Hz<br />

Hand SI<br />

Central sulcus<br />

Salmel<strong>in</strong> & Hari Neurosci 1994<br />

Salmel<strong>in</strong> et al. Neuroimage 1995<br />

Liljeström et al. Neuroimage 2005


Read<strong>in</strong>g aloud<br />

20-Hz effects<br />

10 fT/cm<br />

0 2 4 s<br />

... and Characterize Task-Related<br />

Modulation of Oscillatory Activity<br />

-1...5 s<br />

N=1<br />

word<br />

0 300<br />

speech<br />

?<br />

800 ms<br />

0 2 4 s<br />

Salmel<strong>in</strong> et al. Bra<strong>in</strong> 2000


... and Characterize Task-Related<br />

Modulation of Oscillatory Activity<br />

w ?<br />

Source 20 Hz N=10<br />

Hand<br />

Mouth<br />

Hand<br />

Mouth<br />

2 nAm<br />

0 2 4 s<br />

Salmel<strong>in</strong> et al. Bra<strong>in</strong> 2000, HBM 2002; Saar<strong>in</strong>en et al. Cereb Cortex 2005


For Successful Source Analysis...<br />

Learn to read MEG sensor signals, know your data<br />

– dipole modell<strong>in</strong>g does a very good job there<br />

Learn to do dipole modell<strong>in</strong>g – it is worth the effort<br />

Identify clear field patterns, get well-behaved models<br />

Verify source effects aga<strong>in</strong>st the sensor-level effects<br />

Use multiple experimental conditions: function, modell<strong>in</strong>g<br />

Put effort <strong>in</strong> experimental design and data quality<br />

RS/Apr10

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!