Essentials and Analytic Methods of EEG/MEG signals 徐徐徐 徐徐徐徐徐徐徐徐徐徐徐
Essentials and Analytic Methodsof EEG/MEG signals
徐峻賢中央研究院語言學研究所
by Steven Luckhttp://erpinfo.org/the-erp-bootcamp
references of this presentation
Demo
• EEG– https://www.youtube.com/watch?v=bdsfydAjeUQ
• MEG– https
://www.youtube.com/watch?v=kgInT8hbDuQ&index=9&list=PL16Q0tGI4WvIes7ezj0VoF90tTBG1F1o7
From EEG to ERP
(Lee et al., 2007)
Main Issues
• What is EEG/ MEG?• How to analyze EEG/MEG data?• Advanced approaches
Electroencephalography (EEG)
• ElectroEncephaloGraphy: a recording (graphy) of electrical signal (electro) from the brain (encephalo).
• Action potentials generally make little or no contribution to scalp EEG
• EEG/ERPs reflect mainly the summed PSPs – (EPSPs and IPSPs at that moment) of large populations of pyramidal neurons
• for an ERP, be active in a consistent temporal relationship with the stimulus
(EPSPs and IPSPs)
http://www.ctf.com/Pages/page33.html
Magnetoencephalography (MEG)
EEGMEG
10-20 system
Figure adopted from Malmivuo & Plonsey, 1995
High-Density EEG Recording
Figure adopted from Malmivuo & Plonsey, 1995
Advantages of using ERPs• Lots of data– channels x times x trials• e.g., 62 x 1000 x 1000 (one participants)
– channels x times x trials x frequency• e.g., 157 x 1000 x 1000 x 100 (one participants)
• Freedom from Extraneous Task Demands• Modality Neutral
Analytic stepsContinuous waveforms with event
marks
Epochs aligned to the time-locking
events
Averaged waveforms for analysis
•Epoching•Visual Inspection
•Baseline•Artifact
• Rejection• Correction
•Filtering•Averaging
• Clean EEG data
• Noisy EEG data– line noise, EOG, EMG, etc.
Analytic stepsContinuous waveforms with event
marks
Epochs aligned to the time-locking
events
Averaged waveforms for analysis
•Epoching•Visual Inspection
•Baseline•Artifact
• Rejection• Correction
•Filtering•Averaging
Baseline correction
• Use pre-stimulus interval as baseline– e.g., 100-200 ms before stimulus onset– The amplitude in this period is unaffected by the stimulus.
• CAUTION: any noise in the baseline will add noise to your measures.
Filter
Avoiding artifacts from participants and the experimental procedure• Lexical decision
– with children…..hum…?• go/ no-go semantic judgment
• naming task– simultaneously recording EEG??– homophone judgment
Length of epoch
Silent naming
Homophone judgment
From Epoch to Grand Average
Averaging
• EEG/MEG activity = event-related activity
+random
noise (mean = 0)
Exclusion Criteria
• Behavioral Exclusion Criteria– Error rate
• EEG/MEG Exclusion Criteria– The number of “clean” epochs– Is the “abnormal” pattern informative?
Analytic stepsContinuous waveforms with event
marks
Epochs aligned to the time-locking
events
Averaged waveforms for analysis
•Epoching•Visual Inspection
•Baseline•Artifact
• Rejection• Correction
•Filtering•Averaging
ERP waveforms
http://erpinfo.org/the-erp-bootcamp
MEG response to onsets of single words (visual)
M100
MEG response to onsets of single words (visual)
M170
MEG response to onsets of single words (visual)
M250
MEG response to onsets of single words (visual)
M350
Measuring Amplitudes
Mean amplitudecalculate the mean amplitude in a defined time-window
Area amplitudemean amplitude × number of time point
(Hsu et al. 2014)
CAUTION• Components might overlap
Lateralized Readiness PotentialColes, 1989, Psychophysiology
Examples of components overlapping: semantic judgment taskInstruction: press the left key if the noun is an animal name
press the right key if the noun is not an animal name
P200
N400
P3 + LRP
To avoid components overlapping: go/no-go taskInstruction: press a key if the noun is an animal name
do not press any key if the noun is not an animal name
• Sometimes adopting the factorial design might not allow to have enough trials for estimating ERPs.
• e.g., psycholinguistic studies
frequency 詞頻 : e.g., 村 vs. 皴regularity ( 發音 ) 規則性 : e.g., 楓 vs. 埋orthography-to-phonology consistency 表音一致性 : e.g., 搖 vs. 梳imageability, concreteness 文字指稱的概念特性 : e.g., 蜂 vs. 風grammatical class 語法類型 : e.g., 跑 vs. 紙semantic ambiguity (e.g.: bank, 黃牛 )…etc.
Try your best to control these factors:
Single-trial regression analysis with MEG data
• Random Variable– Subjects, Items
• Fixed Variables– trial numbers (the rank of trials in the list)– number of strokes– phonetic combinability– semantic combinability– frequency– noun-to-verb ratio– semantic ambiguity
physical level
lexical level
orthographic level
semantic level
(Hsu, Lee and Marantz, 2011)
The contributions of bilateral occipital-temporal regions in the reading of Chinese words
• The semantic combinability effects in RH M170 reflects the decomposition of characters.
• Effect of visual complexity in LH M170 suggests that LH fusiform gyrus is a general mechanism for visual word recognition.
(Hsu, Lee and Marantz, 2011)
• Why EEG/MEG is important for studying human mind?
– temporal dynamics of cognitive functions• source analysis
– brain mechanisms• time-frequency analysis, etc.
Source Analysis (1)
Multilayer model:
skull and scalp taken into account,
conductivities needed
Homogeneous model:
skull taken as an insulator,
result independent of conductivity
Head model for EEG Head model for MEG
Source Analysis (2)
• Dale et al. (2000)– L2 minimum norm solution– Dynamic statistical parametric mapping (dSPM)
Time-Frequency Analysis
(Talon-Baudry & Bertrand, 1999, TICS)
Time-frequency activity of auditory-evoke responses of MEG
Two mechanisms of of acoustic change detection (Hsu et al. 2014):1: memory updating (theta bands and T1/ T3 contrasts)2: functional inhibition (alpha bands and T2/ T3 contrasts)
Multiscale entropy (MSE) analysis of EEG signals
Complex EEG
Regular EEG
(Yang, et al, 2014)
A Final Remark
Data are cheap (well… it depends),Facts are expensive,Insight is priceless.
People are desperate to be inspired!