DATA! NOW WHAT? Preparing your ERP data for analysis

Transcription

1 DATA! NOW WHAT? Preparing your ERP data for analysis Dennis L. Molfese, Ph.D. Caitlin M. Hudac, B.A. Developmental Brain Lab University of Nebraska-Lincoln 1

2 Agenda Pre-processing Preparing for analysis steps 2

3 EEG to ERP 3

4 EEG to ERP 4

5 EEG to ERP 5

6 Pre-processing in NetStation 6

7 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 7

8 Filtering Filtering reduces total frequencies by specific amounts 8

9 Filtering 60 cycle : Without proper grounding, impedance between electrode and amplifier can become very large. Causes: skin-to-skin contact, high-frequency interference from nearby electrical devices. Adjust using a notch filter 9

10 FILTERS CHANGE YOUR DATA! 10

11 FILTERS CHANGE YOUR DATA! 11

12 FILTERS CHANGE YOUR DATA! 12

13 FILTERS CHANGE YOUR DATA! 13

14 Filtering Keep in mind: Filters are additive -- the data changes each time you filter. Use same filter settings (and amplifiers) for one dataset. Try different filters out. 14

15 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 15

16 Segmentation Raw EEG signal stamped with stimulus event markers. Segments are the length we want to choose for our ERP. Can be different for each kind of stimulus. 16

17 Segmentation Event markers: Codes are linked to objects in EPrime. 17

18 Segmentation 18

19 Segmentation Offsets: Actual display of object will depend on the connection between PC (EPrime) and Mac (NetStation) computers. PC sends stimulus Stimulus Offset Offset Time Stimulus presented NetStation marks EEG ba 6 da 4 ga 4 bu 5 du 7 gu 6 19

20 Segmentation 20

21 Segmentation Visual Inspection Eye channels could be poor due to no contact with skin or blinks/eye movements. If poor across high number of trials, electrode must be marked BAD for entire recording. 21

22 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 22

23 Artifact Strategy First: Artifact Detection Inspect Eye Channels Check impedance levels from records Have Artifact Detection check for you. Second: Artifact Rejection Replace bad channels through spherical interpolation 23

24 Artifact Detection What electrodes have bad or noisy signal? Bad channels Eye blinks Eye movements Define artifacts by threshold. Ex: Max-Min = 200µV is bad channel over segment. Define artifacts by slope of line -- how quickly does the signal change? Define artifacts by modeling examples. 24

25 Artifact Detection Bad Channel Scan - shorted electrodes - bad scalp connection µv over segment Eye Blinks & Movement -electrode pairs -differ by net type 25

26 Artifact Detection Eye Movements: 125 & 128 Eye Blinks: 8-126,

27 Artifact Detection Be careful: The algorithms can be too simple. If eye channels detect an eye movement or blink, NetStation may mark the segment bad. If an eye channel is bad, may mark all segments bad. 27

28 Artifact Detection 28

29 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 29

30 Artifact Regression Spherical interpolation 30

31 Artifact Regression NetStation uses an automatic ICA Automatically removes eye movement and blink artifacts using blind component separation. Prior to artifact regression with ocular artifacts without ocular artifacts After artifact regression (Joyce, Gorodnitsky, & Kutas, 2004) 31

32 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 32

33 Baseline Correction Important to avoid artifacts in baseline period! Adjust pre-stimulus value (e.g ms) by averaging across pre-stimulus points during baseline. Subtract average value from post-stimulus period. 33

34 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 34

35 Bad Channel Correction No user settings. Instead, we interpolate missing channels using weighted average. 35

36 ERP Pre-processing Fix high/low frequencies Need to chop up EEG into ERP Fix blinks, muscle movements, etc. ERP signal level may vary for each segment Channels with poor signal (e.g impedance, bad electrode) Re-reference; combine trials Filter Segmentation Artifact detection Artifact rejection Baseline correction Bad channel correction Averaging 36

37 Average Reference Why re-reference? Data originally recorded to represent the difference between an electrode and reference (Cz). Define a new zero point. Using an average reference reduces the bias of a particular reference. 37

38 Average Reference 38

39 Averaging trials Averaging is a noise reduction technique. Combine trials to form a single averaged waveform. What trials to average? same condition (happy faces) all stimulus of one type (happy, angry, and fearful faces) 39

40 Averaging Important Strategies to keep in mind: Equal Number of Averages / Condition Equal Number of Averages / Subject Weighted Averaging Random Selection of Trials Early vs. Late in Experiment 40

41 Automating analyses Build scripts for each study. Easy -- just drag and drop! 41

42 Controversy of Order Does the order of operation change the data? Baseline Correction Re-referencing the data Averaging segments 42

43 Controversy of Order 1. Average 2. Average reference 3. Baseline correction 1. Average reference Average 2. Baseline correction 1. Baseline correction Average 2. Average reference 43

44 Pre-processed data: Now what? Perform NetStation analyses Peak amplitude /latency Export data to work in statistical package of choice (e.g. SPSS, SAS, MATLAB). Rotate the data? Combine regional electrodes? 44

45 Exporting data Use NetStation exporter (AVG --> RAW) then convert using Text Exporter program (RAW --> TXT) Use File Export tool in Netstation 45

46 Exporting data BE CAREFUL ABOUT FILE NAMING! One text file per person per condition 00001_6x25.ave_ba.txt 00001_6x25.ave_da.txt 00001_6x25.ave_ga.txt 00001_6x25.ave_bu.txt 00001_6x25.ave_du.txt 00001_6x25.ave_gu.txt 00001_6x25.ave.RAW 00002_6x25.ave_ba.txt 00002_6x25.ave_da.txt 00002_6x25.ave_ga.txt 00002_6x25.ave_bu.txt 00002_6x25.ave_du.txt 00002_6x25.ave_gu.txt 00002_6x25.ave.RAW 46

47 MCAT Program 1. Rotate data matrix for input to stat programs. 2. Concatenate subject data text files. 3. Combine electrode sites into regional electrodes. 47

48 MCAT Program 1. Rotate data matrix for input to stat programs. 2. Concatenate subject data text files. 3. Combine electrode sites into regional electrodes. BEFORE AFTER Channel 1 Data Pt 1 Channel 2 Data Pt 1 Channel 3 Data Pt 1 Channel 1 Data Pt 1 Channel 2 Data Pt 1 Channel 3 Data Pt 1 Channel 1 Data Pt 2 Channel 2 Data Pt 2 Channel 3 Data Pt 2 Channel 1 Data Pt 2 Channel 2 Data Pt 2 Channel 3 Data Pt 2 Channel 1 Data Pt 3 Channel 2 Data Pt 3 Channel 3 Data Pt 3 Channel 1 Data Pt 3 Channel 2 Data Pt 3 Channel 3 Data Pt 3 Suitable for Spatial PCA Suitable for Temporal PCA 48

49 MCAT Program 1. Rotate data matrix for input to stat programs. 2. Concatenate subject data text files. 3. Combine electrode sites into regional electrodes _6x25.ave_ba.txt 00001_6x25.ave_da.txt 00001_6x25.ave_ga.txt 00001_6x25.ave_bu.txt 00001_6x25.ave_du.txt 00001_6x25.ave_ba.txt _6x25.ave_da.txt _6x25.ave_ga.txt _6x25.ave_bu.txt _6x25.ave_du.txt _6x25.ave_gu.txt 00001_6x25.ave_gu.txt 00001_6x25.ave.RAW 00001_6x25.ave.RAW 49

50 MCAT Program 1. Rotate data matrix for input to stat programs. 2. Concatenate HydroCel subject Geodesic data text Sensor files. Net 256 Channel Map Version Combine electrode sites into regional electrodes Fp1 Fp F7 21 F Fz F3 F REF T3 C3 T4 217 CZ C COM P3 P Pz 190 LM RM T T O1 137 O2 123 Oz S/N For questions or additional assistance please refer to the EGI Sensor Net Technical Manual or contact us at: 256 Hydrocel Net Electrical Geodesics, Inc Millrace Drive, Suite 307 Eugene, Oregon Phone: (541) Fax:(541) support@egi.com or info@egi.com 50

51 MCAT Program 1. Rotate data matrix for input to stat programs. 2. Concatenate subject data text files. 3. Combine electrode sites into regional electrodes Fp1 Fp F F Fz F F REF T3 C3 T4 217 CZ C COM P3 P Pz 190 LM RM T T O1 137 O2 123 Oz S/N Hydrocel Net 51

52 MCAT Program For Temporal PCA analyses, drag n drop text files.! 52

53 MCAT Program Subject Time points for ba! 53

54 Summary Pre-processing involves making informed decisions about your data! Review the literature for similar paradigms to decide how to start. Make a plan -- be prepared and organized. It s okay to revise your plan! 54

55 QUESTIONS??? 55

56 MCAT 56

57 MCAT to SPSS 57