Links between brain cortical regions and EEG recording sites derived from forward modelling

Abstract

Electroencephalography (EEG) provides no direct link between electrode positions and underlying signal generators. Inferences based on spatial proximity between scalp positions and cortical structures are not reliable with a higher number of electrodes. More accurate source localization is obtained by solving both the forward and the inverse problem, yet such approach is computationally demanding and may not be best suited for fast initial or exploratory analyses of EEG data. In this paper, we provide a reference table of correspondence between EEG sensors and cortical anatomical regions based on a realistic head model for the International 10-10 and EasyCap M10 electrode positioning systems. We also present a universal algorithm to compute the solution by using a forward model to determine the sensitivity for electrodes of any defined electrode positioning system and cortical anatomical parcellation. The reference tables and solutions based on applying this algorithm may be used to aid interpretation of EEG data, in particular on occasions when it is impractical to apply sophisticated and demanding source localisation methods.

Keywords: electroencephalography, source localization, electrode positions, system 10-10, forward model, brain parcellation

Additional results

The analysis outlined in the results section of the paper was also performed using the Destrieux atlas. The maximum sensitivity figures at each electrode were the same since this measure is independent on the atlas. The minimum intersection between the vertex set and the corresponding region of the greatest linkage was 8.6 % (C1; left superior parietal lobule), the maximum was 37.5 % (AF9; left orbital gyrus), 17.52 ± 6.26 % on average. When we included two regions of the strongest linkage, the mean intersection was 29.8 ± 9.2 % (min = 15.9 %, max = 59.1 %), for three areas the mean intersection was 39.7 ± 11.6 % (min = 23.2 %, max = 77.3 %). This percentage is lower than for Mindboggle likely because the number of anatomical regions (and thus location specificity) is lower for the Mindboggle than for the Destrieux atlas.

For the M10 montage, we found that each electrode yielded maximum sensitivity for at least 8 (sensor 56) and at most 528 (sensor 24) vertices, 231.56 ± 114.04 vertices on average. For the Mindboggle atlas, the minimum intersection between the vertex set and the corresponding cortical region of the greatest linkage was 9.7 % (sensor 6; precentral L) and the maximum was 52.6 % (sensor 55; lateraloccipital R), 26.62 ± 10.18 % on average. Including two areas of the strongest linkage yielded a mean of 44.46 ± 14.25 % (min = 17.8 %, max = 89.4 %), and including three regions, this value was 57.42 ± 16.01 % (min = 25.1 %, max = 100 %, more than 80 % for 6 electrodes: 2, 53, 55, 56, 57, and 60).

Finally, we performed the same analysis using the M10 montage also for the Destrieux atlas. The minimum intersection between the vertex set and the corresponding area of the strongest linkage was 13.3 % (sensor 20; right superior frontal gyrus), the maximum was 74.4 % (sensor 61; left orbital gyrus), 31.62 ± 15.18 % on average. Including two or three areas of the strongest linkage resulted in a mean intersection of 50.3 ± 18.47 % (min = 24.1 %, max = 100 %) or 62.7 ± 17.29 % (min = 33.1%, max = 100 %) respectively.