Electrolyte bridges in dense array EEG data
Calculations of cross correlations between electrode signals on the 128 channel EGI nets revealed some interesting patterns of very high values (>0.995) showing strong links between groups of electrodes located in steep areas of the head. Perhaps the flow of electrolyte was causing bridging between electrodes. Furthermore, the number of these strong correlations decreased for recordings made later in a session after some tens of minutes - enough time for electrolyte bridges to have dried out. The following plots have blue lines connecting electrodes showing a correlation greater than 0.995 over a one minute period at the beginning of an EEG recording session (left) and approximately four minutes later (right).
The following plots start about 8 minutes into the same session as above and show correlations greater than 0.995 over a one minute period at the beginning (left) and end (right) of these approximately nine minute runs.
The strong correlations appear concentrated in the steeper parts of the head and there's an obvious decrease in the number of strongly correlated electrodes over the nearly 30 minutes of recording. Note that some connections break and re-connect such as electrodes 72/76 between the first and third plots. The third plot displays a rather extended connection pattern between electrodes 73/77, 72/77, 72/76 and 75/76 which mostly disappears by the next plot (except for 73/77). These data were collected after about eight minutes of eyes open/closed baselines.
Tenke and Kayser (Clinical Neurophysiology, in press) suggest using the intrinsic Hjorth transform to detect electrolyte bridges in dense electrode arrays. This method consists of plotting the difference between an electrode's values and a weighted average of its nearest neighbors. The following excerpt from their paper describes the method in detail:
In practice they use the single nearest neighbor given by Dij in equation 4. Thus, Wij becomes 1 for this single nearest neighbor and H in equation 1 becomes the difference between an electrode's values and the values at the electrode at the smallest "electrical distance". So H(t) should have very small magnitude for all t if there exists an electrolyte bridge between two electrodes. Tenke and Kayser suggest that perhaps the "electrical distance" alone can be used to determine the presence of electrolyte bridging.
An electrolyte bridging experiment was run using the EGI recording system. Lots of electrolyte was used in an attempt to create a bridge between electrodes 6/13 with excess creating a possible bridge to electrode 5. Also electrodes 54/62 were positioned so that their sponges were touching in order to make a definite bridge between them. The Hjorth waveforms and "Electrical distances" (equations 1 and 4, respectively) were computed with the following results:
Hjorth waveforms for 10 seconds of data which was first 20Hz low pass filtered and then detrended.
20Hz low pass filtered and detrended data
The "Electrical distances" between electrodes were ranked from smallest to largest:
Rank Channel
Neighbor Distance
1
54
62
0.225
2
62
54
0.225
3
61
62
1.057
4
64
65
1.539
5
65
64
1.539
6
50
51
3.500
7
51
50
3.500
8
98
99
3.544
9
99
98
3.544
10
43
48
4.508
11
48
43
4.508
. . . .
124
33
22 70633.813
125
74
92 78644.133
126
83
82 85752.221
127
59
44 88607.213
128
75
9 94537.954
The Hjorth waveform and "Electrical distance" clearly locate the physical bridge between electrodes 54/62 whereas the attempt to create an electrolyte bridge between electrodes 6 and 13 appears to have failed.
This method was next applied to the DAL014 data set described above with the following results:
Hjorth waveforms for 10 seconds of data which was first 20Hz low pass filtered and then detrended DAL014EC1_NO60.DAT
20Hz lowpassed data for first 10 seconds of data of DAL014EC1_NO60.DAT
Rank Channel
Neighbor Distance
1
36
41
0.131
2
41
36
0.131
3
28
34
0.224
4
34
28
0.224
5
40
28
0.228
6
46
41
0.257
7
109
110
0.296
8
110
109
0.296
9
39
34
0.305
10
111
109
0.327
11
49
34
0.331
12
45
39
0.362
13
52
59
0.373
14
59
52
0.373
15
79
86
0.422
16
86
79
0.422
17
118
110
0.438
18
117
110
0.469
19
124
117
0.634
20
3
10
0.657
21
10
3
0.657
22
43
52
0.661
23
93
86
0.708
24
65
70
0.720
25
70
65
0.720
26
123
124
0.796
27
23
27
0.807
28
27
23
0.807
29
35
40
0.844
30
29
35
0.902
31
25
28
1.015
If we try cutoff thresholds for "Electrical distance" of slightly larger than the 0.225 found from the bridging experiment say at 0.250 and 1.00 then the following plots show the potential bridging and can be compared to the first two correlation maps above:
The bridging patterns are similar but less extensive than those indicated by the large correlations shown above even for the relatively large threshold of 1.00 microV^2. However, these electrical distances are much larger than the 0.140 (0.064 for detrended data) found by Tenke and Kayser for ERP data.
Following are results from DAL014R_ERP which is an average of the first second after the feedback event (n=27).
Hjorth waveforms:
Original data:
Rank Channel Neighbor Distance
1 36 41 0.027
2 41 36 0.027
3 110 117 0.029
4 117 110 0.029
5 123 124 0.029
6 124 123 0.029
7 103 104 0.031
8 104 103 0.031
9 92 93 0.032
10 93 92 0.032
11 109 110 0.032
12 118 117 0.032
13 46 41 0.033
14 34 39 0.034
15 39 34 0.034
16 42 47 0.034
17 47 42 0.034
18 72 76 0.036
19 76 72 0.036
20 40 34 0.037
21 52 59 0.038
22 59 52 0.038
23 50 57 0.042
24 57 50 0.042
25 73 77 0.042
26 77 73 0.042
27 58 64 0.043
28 64 58 0.043
29 119 118 0.043
30 69 64 0.043
31 35 40 0.044
32 116 109 0.046
33 28 35 0.046
34 97 98 0.050
35 98 97 0.050
Although many of the electrical distances are relatively small the Hjorth
waveform plots don't show any electrodes which are distinctly flatter than the
rest (see for example Tenke and Kayser Fig. 2B).
For comparison similar plots were produced for some 22 channel EEG where electrolyte bridging is much less likely to be a problem:
Hjorth waveforms for 10 seconds of data which was first 20Hz low pass filtered and then detrended I001RSF.DAT
20Hz lowpassed data for first 10 seconds of data of I001RSF.DAT
Rank Channel
Neighbor Distance
1
5
19
20.765
2
19
5
20.765
3
2
21
20.780
4
21
2
20.780
5
17
22
23.796
6
22
17
23.796
7
4
21
24.199
8
10
17
26.853
9
14
22
29.403
10
3
21
30.556
11
16
17
31.906
12
15
22
33.357
13
1
18
37.870
14
18
1
37.870
15
9
12
39.442
16
12
9
39.442
17
20
14
53.879
18
13
12
69.714
19
6
19
72.552
20
7
6
206.618
21
8
11
357.446
22
11
8
357.446
These results show no evidence of electrolyte bridging.
Conclusions:
Very high correlations between neighboring electrodes which are located in steep areas of the head and the diminishing of the correlations with time are suggestive of electrolyte bridging.
The single electrolyte bridge experiment failed to produce bridging where excess electrolyte was used. Electrode sponges in physical contact did produce bridging effects.
"Electrical distance" between electrodes gave results similar to correlations for a rather large threshold of 1.00 microV^2 but for a more realistic threshold of 0.25 microV^2 indicated relatively few bridges (5 electrodes versus 36).