|
Size: 4209
Comment:
|
Size: 4209
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 23: | Line 23: |
| file= 'VPibv_10_11_02/CenterSpellerMVEP_VPibv'; [cnt, mrk, mnt]= eegfile_loadMatlab(file); |
|
| Line 31: | Line 34: |
file= 'VPibv_10_11_02/CenterSpellerMVEP_VPibv'; [cnt, mrk, mnt]= eegfile_loadMatlab(file); |
Documentation: Some Practical Examples of how to use the BBCI Toolbox
IN CONSTRUCTION
Table of Contents
ERP Analysis: #ERP-Aanylsis
Spectral Analyses: <to come>
<more to come>
ERP Analysis
1 file= 'VPibv_10_11_02/CenterSpellerMVEP_VPibv';
2 [cnt, mrk, mnt]= eegfile_loadMatlab(file);
3
4 % Define some settings
5 disp_ival= [-200 1000];
6 ref_ival= [-200 0];
7 crit_maxmin= 70;
8 crit_ival= [100 800];
9 crit_clab= {'F9,z,10','AF3,4'};
10 clab= {'Cz','PO7'};
11 colOrder= [1 0 1; 0.4 0.4 0.4];
12
13 % Apply highpass filter to reduce drifts
14 b= procutil_firlsFilter(0.5, cnt.fs);
15 cnt= proc_filtfilt(cnt, b);
16
17 % Artifact rejection based on variance criterion
18 mrk= reject_varEventsAndChannels(cnt, mrk, disp_ival, 'verbose', 1);
19
20 % Segmentation
21 epo= cntToEpo(cnt, mrk, disp_ival);
22
23 % Artifact rejection based on maxmin difference criterion on frontal chans
24 epo= proc_rejectArtifactsMaxMin(epo, crit_maxmin, ...
25 'clab',crit_clab, 'ival',crit_ival, 'verbose',1);
26
27 % Baseline subtraction, and calculation of a measure of discriminability
28 epo= proc_baseline(epo, ref_ival);
29 epo_r= proc_r_square_signed(epo);
30
31 % Select some discriminative intervals, with constraints to find N2, P2, P3 like components.
32 fig_set(1);
33 constraint= ...
34 {{-1, [100 300], {'I#','O#','PO7,8','P9,10'}, [50 300]}, ...
35 {1, [200 350], {'P3-4','CP3-4','C3-4'}, [200 400]}, ...
36 {1, [400 500], {'P3-4','CP3-4','C3-4'}, [350 600]}};
37 [ival_scalps, nfo]= ...
38 select_time_intervals(epo_r, 'visualize', 1, 'visu_scalps', 1, ...
39 'title', untex(file), ...
40 'clab',{'not','E*'}, ...
41 'constraint', constraint);
42 printFigure('r_matrix', [18 13], opt_fig);
43 ival_scalps= visutil_correctIvalsForDisplay(ival_scalps, 'fs',epo.fs);
44
45 fig_set(3)
46 H= grid_plot(epo, mnt, defopt_erps, 'colorOrder',colOrder);
47 grid_addBars(epo_r, 'h_scale',H.scale);
48 printFigure(['erp'], [19 12], opt_fig);
49
50 fig_set(2);
51 H= scalpEvolutionPlusChannel(epo, mnt, clab, ival_scalps, defopt_scalp_erp2, ...
52 'colorOrder',colOrder);
53 grid_addBars(epo_r);
54 printFigure(['erp_topo'], [20 4+5*size(epo.y,1)], opt_fig);
55
56 fig_set(4, 'shrink',[1 2/3]);
57 scalpEvolutionPlusChannel(epo_r, mnt, clab, ival_scalps, defopt_scalp_r2);
58 printFigure(['erp_topo_r'], [20 9], opt_fig);
Even in this very basic ERP analysis, there are several steps, in which the choice of processing can have quite a big impact on the results, but nevertheless, the correct choice is not clear to us (maybe we will find out a recommendable choice at some point).
- Highpass filter (lines 14-15). Highpass filtering may be beneficial to reduce the impact of drifts.
- But the choice of the filter has quite some impact on the ERPs, in particular, the later ERPs.
Alternatives are proc_subtractMovingAverage(cnt, 1500, 'centered', 'sinus'), other highpass filters, or no highpass filtering at all. Also bandpass filtering is an alternative (e.g. [0.5 30]), but typically the lowpass filtering is not required since high frequency are dampened also by the normal ERP averaging across trials.
- But the choice of the filter has quite some impact on the ERPs, in particular, the later ERPs.
- Baseline correction (line 28). For the ERP themselves the choice of baseline correction does not matter.
- They stay the same. But for the measure for discriminability (e.g. r^2-values) it may have a big impact. In this example, baseline correction is performed on a trialwise basis. This way it is done for online experiments. For the ERP analysis this may be have the disturbing consequency, that the r^2-values in/near the baseline interval can become spuriously high, since the trial-to-trial variance is artificially reduced. Alternatives are to subtract the across-trials
average of the baseline (option 'trialwise',0 in proc_baseline) or to subtract the classwise average of the baseline (option 'classwise',1 in proc_baseline).
- They stay the same. But for the measure for discriminability (e.g. r^2-values) it may have a big impact. In this example, baseline correction is performed on a trialwise basis. This way it is done for online experiments. For the ERP analysis this may be have the disturbing consequency, that the r^2-values in/near the baseline interval can become spuriously high, since the trial-to-trial variance is artificially reduced. Alternatives are to subtract the across-trials
- Measure for discriminability (line 29). Choices are, e.g., signed r^2, t-values, p-values, AUC-score.