[ERP Decoding] The step toward mind reading

2023. 2. 10. 00:15
๐Ÿง  Key Words

fMRI
ERP scalp map
orientations

 

์–ด๋–ป๊ฒŒ ์‚ฌ๋žŒ์˜ ๋งˆ์Œ์„ ์ฝ์„ ๊ฒƒ์ธ๊ฐ€?

 

  • ์ฒ˜์Œ์œผ๋กœ ์ ‘๊ทผํ•œ ๊ฒƒ์€ mind reading์„ ์œ„ํ•ด visual orientation์„ ์‚ดํ”ผ๋Š” ๊ฒƒ์ด๋‹ค.
    • Different orientations maximally stimulate different fMRI voxels in visual cortex.
    • ๋‹ค๋ฅธ orientation ์ž๊ทน์— ๋Œ€ํ•ด fMRI๋Š” V1 ์˜์—ญ์—์„œ ๋‹ค๋ฅธ ๋ถ€๋ถ„์„ ๋‚˜ํƒ€๋‚ธ๋‹ค.
    • We might therefore expect slightly different ERP scalp distributions(i.e., voltage pattern over the scalp) for different orientations.
    • ๊ทธ๋Ÿฌ๋ฏ€๋กœ ์šฐ๋ฆฌ๋Š” ๋‹ค๋ฅธ orientation ์ž๊ทน์— ๋Œ€ํ•ด ERP๋„ ๋ถ„์„์—์„œ๋„ ERP scalp map์ด ๋‚˜๋ฅด๊ฒŒ ๋‚˜ํƒ€๋‚˜๋Š” ๊ฒƒ์„ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋‹ค.
  • ์‹คํ—˜๋‚ด์šฉ.
    • 16๊ฐœ์˜ orientations, Scalp distribution of ERP for 16 orientations.
    • EEG recording.
    • ํ•ด๋‹น orientation์„ ์ฃผ๊ณ  ์ž ์‹œ ๋’ค ๊ทธ orientation์„ ํ”ผ์‹คํ—˜์ž๊ฐ€ ์ฐ๋„๋ก ํ•˜๋Š” ๊ฒƒ. 
  • next 
    • ERP scalp distrubution with orientation label 16๊ฐœ์— ๋Œ€ํ•ด ์ƒˆ๋กœ ๊ตฌํ•œ dataset์ธ scalp map์„ ๋น„๊ตํ•˜์—ฌ ์–ด๋–ค scalp map์ด ์ƒˆ๋กœ ๊ตฌํ•œ ์ด data์™€ ๊ฐ€์žฅ ๋น„์Šทํ•œ๊ฐ€๋ฅผ ๋ณธ๋‹ค.
    • ์ฐพ์€ ๊ฒƒ์€ label์˜ orientation์ด 60๋„ ์ด๋ฉด ์ƒˆ๋กœ ๊ตฌํ•œ data set์˜ orientation๋„ 60๋„ ์ผ ๊ฒƒ์ด๋‹ค.
    • If the scalp distributions for different orientations are distinguishable,
    • and if the differences in the scalp distributions are systematic,
    • then we should be able to guess the orientation associated with this new scalp distributon.

 

 

Decoding orientation from the scalp distribution of ERP

์ถœ์ฒ˜ : https://erpinfo.org/blog/2018/9/16/decoding

 

 

  • y์ถ•์€ Decoding Accuracy, x์ถ•์€ chance level๋กœ time.
    • x์ถ•์—์„œ,
    • Stimulus Encoding
    • Working memory maintenance
    • Orientation report
    • ์™€ ๊ฐ™์€ ์ˆœ์„œ๋กœ ์ด๋ฃจ์–ด์งˆ ์ˆ˜ ์žˆ๋‹ค.
  • If the scalp ERP contains no information about the orientation,
  • then decoding accuracy should be at chance.
  • ERP scalp๊ฐ€ orientation ์ •๋ณด๋ฅผ ํฌํ•จํ•˜๋ฉด chance ์ด์ƒ์ด๊ณ , ์•„๋‹ˆ๋ผ๋ฉด decoding accuracy๊ฐ€ chance level์ผ ๊ฒƒ์ด๋‹ค.
  • Gray area : Statistically significant time points.
  • ๊ทธ๋ฆฌ๊ณ  ์ž๊ทน์„ ๋ณด๋Š” perceptual ๊ธฐ๊ฐ„๊ณผ ์ž๊ทน์„ ๊ธฐ์–ตํ•˜๋Š” working memory ๊ธฐ๊ฐ„์œผ๋กœ ๋‚˜๋‰œ๋‹ค.
    • ์œ„ gray area์—๋Š” working memory ๊ธฐ๊ฐ„๋งŒ ํ†ต๊ณ„์ ์ธ ๋ถ„์„์„ ํ•œ๋‹ค.
    • ๋‡ŒํŒŒ๋ฅผ ํ†ตํ•ด ์–ด๋–ค orientation์„ ๊ธฐ์–ตํ•˜๊ณ  ์žˆ๋Š”์ง€ Decoding ํ•  ์ˆ˜ ์žˆ๋‹ค.
  • Spatial distribution of ERP contains information about orientation !
  • orientation์ด ๊ธฐ์–ต๋˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค€๋‹ค.

 

 

์ •๋ง Orientation ๋ณ„๋กœ Decoding์ด ๋˜๋Š” ๊ฑธ๊นŒ?

 

Confusion Matrix

  • stimulus, classification, probability๊ฐ€ ์กด์žฌํ•œ๋‹ค.
  • ๋Œ€๋ถ€๋ถ„์˜ orientation๋“ค์ด ๋‹ค Decoding ๋จ์„ ๋ณด์—ฌ์ค€๋‹ค.
  • Decoding was above chance for most of the orientations !

์ถœ์ฒ˜ : https://www.researchgate.net/figure/Confusion-matrix-for-the-temporal-orientation-classification_fig2_325447081

 

demonstration ์ด ์™„๋ฃŒ ๋˜์—ˆ๋‹ค.

 

์ด๋ฅผ ํ†ตํ•ด ์šฐ๋ฆฌ๊ฐ€ ๋ญ˜ ํ•  ์ˆ˜ ์žˆ์„๊นŒ?

 

 

Experiment 2.

 

  • ์ด๋ฒˆ์—๋Š” ๊ธฐ์–ตํ•ด์•ผํ•˜๋Š” ์ž๊ทน์ด ๋‚˜์˜ค๋Š” ์œ„์น˜์™€ ๋ฐ˜์‘ํ•ด์•ผํ•˜๋Š” ๊ฒƒ์ด ๋‚˜์˜ค๋Š” ์œ„์น˜๊ฐ€ random์œผ๋กœ ์„ค์ •ํ•˜์˜€๋‹ค.
  • Attention์„ ๋ณด๊ธฐ ์œ„ํ•จ์ด๋‹ค.
  • Location of sample teardrop and test teardrop varied randomly every trial.
  • Attending to this location is not a useful strategy.
  • Orientation value์™€ Location value๋ฅผ ์™„์ „ํžˆ randomํ•˜๊ฒŒ independentํ•˜๊ฒŒ ์‹ค์‹œํ–ˆ๋‹ค.
  • ์ด์ „ trial์— ๋ณธ ์ž๊ทน์„ ํ˜„์žฌ trial์— decodingํ•  ์ˆ˜ ์žˆ๋‚˜.
  • Information of the previous-trial orientation mush be present during the current trial.
  • The two orientations were independent.
  • But the report of the current trial was biased by the orientation from the previous trial.
  • Decoding ๋œ๋‹ค๊ณ  ํ•˜๋ฉด, ๊ณผ๊ฑฐ ์‹œํ–‰์ด ํ˜„์žฌ์—๋„ ๋‚จ์•„ ์•˜๋‹ค.
  • ์ด์ „ ์‹œํ–‰์— ๋ดค๋˜ ๊ฒƒ์ด ํ˜„์žฌ ์‹œํ–‰์—์„œ๋„ ์˜ํ–ฅ์„ ๋ผ์น˜๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ๋‹ค.
  • Memory reactivation์„ Decoding์„ ํ†ตํ•ด ์•Œ ์ˆ˜ ์žˆ๋‹ค.

 

  • motion direction ๊ณผ ๊ฐ™์€ ๊ฒƒ์—์„œ๋„ ๊ฐ€๋Šฅ. ( Continuous direction estimation task )
    • ์ •ํ•ด์ง„ ์‹œ๊ฐ„์— motion direction ์ •๋ณด๋ฅผ ์ถ•์ ํ•˜์—ฌ ๋ฐฉํ–ฅ์„ ๋งž์ถ”์–ด์•ผ ํ•จ.
  • ERP-based stimulus decoding:
  • Spatial pattern of sustained ERP activity reflects both direction information and shift of attention.

 

  • Decoding faces from the scalp distribution of ERP.
  • ERP Decoding์—์„œ Indentity or facial expression ๋‘˜ ๋‹ค decoding ๊ฐ€๋Šฅํ•œ๊ฐ€.
  • We can decode ID independently of expression and expression independently of ID.

 

  • Decoding Positive vs. Negative images
  • ๊ธ์ •, ๋ถ€์ • ์‚ฌ์ง„์— ๋Œ€ํ•ด ์–ด๋– ํ•œ ERP ์ฐจ์ด๋ฅผ ๋ณด์ผ ๊ฒƒ์ธ๊ฐ€.
  • ๊ฐ’ ์ž์ฒด๋Š” ์ฐจ์ด๋ฅผ ๋ณด์ด์ง€ ์•Š์ง€๋งŒ, Decoding accuracy๋Š” ๊ทน๋ช…ํ•œ ์ฐจ์ด๋ฅผ ๋ณด์ธ๋‹ค.
  • Barely any difference between positive and negative images.

 

  • Why is decoding more sensitive than univariate ERP analysis?
  • univariate ์ˆ˜์ค€์—์„œ ํŒ๋‹จํ•˜๊ธฐ ์–ด๋ ค์šด ์ˆ˜์ค€์˜ ์ฐจ์ด๋ฅผ multivariate analysis์—์„œ ๋” ์ž˜ ๊ตฌ๋ถ„ํ•  ์ˆ˜ ์žˆ๋‹ค.

 

Limitaions in the ERP decoding analysis

  • Decoding is done from averaged ERPs
  • ํ‰๊ท ์„ ๋‚ด์„œ ํŒ๋‹จํ•˜๋ฏ€๋กœ ๊ณตํ•™์ ์œผ๋กœ ํ•œ๊ณ„๊ฐ€ ์กด์žฌํ•จ.
    • Single-trial decdoing is not reliable.
    • However, depending on the signal, reliable decoding may be possible.
  • Decoding accuracy is far from 100%.
  • Decoding accuracy๊ฐ€ ๋‚ฎ์€ ๊ฒƒ์€ real world์—์„œ๋Š” ํ•œ๊ณ„์ด์ง€๋งŒ scientificํ•œ ๊ฒƒ์—์„œ๋Š” ์ด์šฉ๋  ์ˆ˜ ์žˆ๋‹ค.
    • Suitable for answering scientific questions.
    • Not suitable for engineering applications.
  • Usually, we do not see reliable correlation between decoding accuracy and behavioral performance.
  • decoding์ด ์ข‹์œผ๋ฉด ํ–‰๋™ ๋Šฅ๋ ฅ์ด ์ข‹์„ ๊ฒƒ์ธ๊ฐ€?
    • Individual differences in decoding accuracy are probably driven mainly by cortical geometry and signal-to-noise ratio.
    • ์‚ฌ๋žŒ๋งˆ๋‹ค ๋‹ค๋ฅธ ๊ฒƒ ๋•Œ๋ฌธ์— decoding์— ์ฐจ์ด๊ฐ€ ์žˆ์„ ๋ฟ, ํ–‰๋™ ๋Šฅ๋ ฅ๊ณผ๋Š” ๊ด€๋ จ ์—†๋‹ค.
  • We do not know which neural process is being used in the decoding.
  • ์–ด๋–ค Neural process๊ฐ€ decoding์— ์ฐจ์ด๋ฅผ ๋ณด์ด๋Š”์ง€ ์•Œ ์ˆ˜๊ฐ€ ์—†๋‹ค.
  • ๋‘ ๊ฐœ ์‚ฌ์ด์˜ ์ฐจ์ด๋งŒ ์žˆ์œผ๋ฉด classify๊ฐ€ ๊ฐ€๋Šฅํ•˜๊ณ  ์ด๋ฅผ ํ†ตํ•ด above chance decoding์ด ๊ฐ€๋Šฅํ•˜๋‹ค.
    • ๊ทธ๋Ÿฌ๋‚˜ ์–ด๋–ค ์‹ ํ˜ธ๊ฐ€ ๊ทธ๊ฒƒ์„ ๋งŒ๋“ค์–ด ๋‚ด๋Š”์ง€๋Š” ์•Œ ์ˆ˜๊ฐ€ ์—†๋‹ค.

 

 

ERP Analysis์™€ ERP Decoding ๋‘ ๊ฐœ๋Š” ์„œ๋กœ์˜ ์ฒ˜๋ฆฌ์— ๊ด€ํ•ด ์˜์—ญ์ด ๋‹ค๋ฅผ ๋ฟ, ๋‘˜ ๋‹ค ์˜๋ฏธ ์žˆ๋Š” ๋ถ„์„ ๋ฐฉ๋ฒ•์ด๋‹ค.

 

 

How is decoding different from the standard approach ?

  • We can separate signals associated with the representational contents from signals associated with contents-free support process.
  • ERP๋Š” ํ”„๋กœ์„ธ์Šค๋ฅผ ๋ณด๋Š” ๊ฒƒ์œผ๋กœ, ํ”„๋กœ์„ธ์Šค๊ฐ€ ERP Amplitude๊ฐ€ ์–ด๋–ค ๊ฒƒ์ธ์ง€ ์•Œ ์ˆ˜ ์—†๋‹ค.
  • Decoding๋Š” ์ด์™€ ๋‹ค๋ฅด๊ฒŒ content์— ๋Œ€ํ•œ, ๋ญ๊ฐ€ ์ง€๊ฐ๋˜๋Š”์ง€๋ฅผ ์•Œ ์ˆ˜ ์žˆ๋‹ค.

 

  • Decoding answers the quesion "Does t he neural signal contain information about the stimulus?" with minimal assumptions about how the information is encoded in the neural signal
    • We do not know which neural process is being used in the decoding
  • ์ด ERP pattern์ด ์–ด๋–ค ์ •๋ณด๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ๋Š๋ƒ. ์ฐจ์ด๊ฐ€ ์กด์žฌํ•˜๊ธฐ๋งŒ ํ•œ๋‹ค๋ฉด ๋ถ„๋ฆฌํ•˜์—ฌ ๋ถ„์„ ๊ฐ€๋Šฅํ•˜๋‹ค. Assumption์ด ์—†๋Š” ๊ฒƒ์ด ์žฅ์ .
  • ๊ทธ๋Ÿฌ๋‚˜ ์–ด๋–ค specificํ•œ neural process๊ฐ€ decoding above chance๋ฅผ ๋งŒ๋“ค์–ด๋‚ด๋Š”์ง€๋Š” ์•Œ ์ˆ˜ ์—†๋‹ค.
  • Decoding is done separately for each subject, whereas the standard approach looks for consistency across subjects.
    • Greater statistical power in basic science research.
  • Decoding์€ subject ๋งˆ๋‹ค ๋”ฐ๋กœ ์ด๋ฃจ์–ด์ง„๋‹ค.
  • Individual difference๊ฐ€ ์˜ํ–ฅ์ด ์—†์ง€๋งŒ, ERP์—์„œ๋Š” ์ค‘์š”ํ•˜๋‹ค.
  • Decoding์€ ๊ฐ๊ฐ ๋ถ„์„์ด ๋˜๊ธฐ ๋•Œ๋ฌธ์— individual difference๊ฐ€ ์žˆ๋‹ค๊ณ  ํ•˜๋”๋ผ๋„ ํšจ๊ณผ๊ฐ€ ์ ๊ฒŒ ๋œ๋‹ค.
  • ์ด ์ฐจ์ด๋ฅผ ๋ณด๊ธฐ ์œ„ํ•ด Effect Size๋ฅผ ๋ณธ๋‹ค.

 

 

๋งŒ์•ฝ Decoding์ด above chance๊ฐ€ ์•„๋‹ˆ๋ผ๋ฉด ?

  • If decoding accuracy of one condition/group is worse than another, this may be because the data are noisier in that group/condition.
  • Decoding accuracy depends on Signal-to-Noise ratio of th data
  • Decoding ๊ฒฐ๊ณผ ์กฐ๊ฑด A ๋ณด๋‹ค B์—์„œ ํŒจํ„ด์ด ๋” ๊ฐ•ํ–ˆ์„ ๊ฒƒ์ด๋‹ค ๋ผ๋Š” ๊ฒƒ์€, ๊ฒฐ๊ณผ๋ฅผ ๋‹ค๋ฅด๊ฒŒ ํ‘œํ˜„ํ•œ ๊ฒƒ ๋ฟ.
  • ์™œ ๋†’์€์ง€๋Š” ์•Œ๊ธฐ ํž˜๋“ค๋‹ค. Decoding์€ ์–ด๋–ค ์ฐจ์ด๋“  ๋‹ค ๋ฐ˜์˜ํ•˜๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค.
  • Decoding accuracy๋Š” Signal-to-Noise์˜ ๋น„์œจ์— ์˜ํ•ด ๊ฒฐ์ •์ด ๋˜๊ธฐ ๋•Œ๋ฌธ์—, 
    • Signal is the distance between the mean values between conditions in the multidimensional space.
    • Noise is the spread of the data within each condition.
    • Lower decoding accuracy for one condition/group may be driven by the same level of single but higher level of noise.

 

Interpretation issue

  • Through decoding analysis, we are looking at a correlation between brain activity and the stimulus category.
    • Above-chance decoding means that 'any' differences in the stimulus category is correlated with the neural activity.
    • ์–ด๋–ค ๋‹จ์ˆœํ•œ ์ฐจ์ด๋„ decoding ๊ฒฐ๊ณผ์— ์˜ํ–ฅ์„ ์ค€๋‹ค.
    • ๊ทธ๋Ÿฌ๋‹ˆ ์ฃผ์˜๊ฐ€ ํ•„์š”ํ•˜๋‹ค.
  • When possible, decoding-irrelevant feature dimension of the stimulus should be controlled.

 

Understand complex human cognition

  • ERP๋กœ๋ถ€ํ„ฐ decodingํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ.
  • What can we decode from ERPs. 
    • Location and Orientation
    • The orientation from the previous trial
    • Face identities and expressions
    • motion directions
    • positive vs negative valence of images
    • letters
  • standard ERP์™€ Decoding ๋Š” ์ฐจ์ด์ ์ด ์žˆ๋‹ค.
  • How is decoding better/different from the standard ERPs analysis?
    • Decoding is more sensitive than ERP analysis.
    • Decoding is done separately for each subject, whereas the standard approach looks for consistency across subjects.
    • Grater effect size.
  • Limitations in the decoding analysis.
    • 'any' differences between conditions potential produce above chance decoding.
    • Decoding irrelevant feature dimensions should be controlled/counter-balanced.

 

 

 

'Biology > EEG' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๋‹ค๋ฅธ ๊ธ€

[ERP Decoding] EEG recording & Preprocessing  (0) 2023.02.12
[ERP Decoding] ERP Decoding procedure  (0) 2023.02.11
[ERP Decoding] EEG - Electronencephalography  (0) 2023.02.07

BELATED ARTICLES

more