Anatomic Landmarks

The method proposed so far is expected to provide a broad range of channels over the cortex and thus cover most regions that can be reached with fNIRS. However, as these channels are formed by sources and detectors placed on 10–10 international system, users interested in running EEG-fNIRS multi-modal measures would not be capable of using the toolbox, as EEG placement by standard is based on 10–20 and 10–10 international systems^{12 (link)}.
To extend the possible optodes positions to 10–5 international system positions regarding layouts available for EEG caps, we have considered as a reference a cap with 130 positions in total. With this new design, one has the possibility to use either 32 positions (Fig. 6A) or 64 positions for EEG electrodes (Fig. 6B), while the locations for fNIRS optodes do not overlap with the 10–10 system.Figure 6

(A) Expansion of the method described for 10–10 international system to (B) the 10–5 system to allow for multi-modal measurements with EEG, either 32 or 64 electrodes. EEG and fNIRS positions are based on a layout accommodating 130 positions in total. EEG 1–32 electrodes positions are depicted in green, while the complimentary 33–64 are in yellow; fNIRS sources positions are in red and the detectors are in blue.

We have visually assigned sources and detectors to 10–5 system positions with the goal to maximize the number of possible channels considering adjacent optodes. This resulted on the layout illustrated in Fig. 6, which presents 28 sources positions and 28 detectors positions over the scalp. From these positions, we considered 89 possible channels in total.
Once the positions have been assigned and their coordinates have been retrieved for both head atlases (Methods section), we proceeded with the methods described in the Methods section. Therefore, we ran the photon transport simulations and computed the normalized sensitivity, ROIs specificity and channels coordinates, and obtained the anatomical landmarks results of each parcellation atlas.
The derived results were also stored in Matlab files included in the fOLD toolbox.

The first level comparisons of encoding (scenes) vs. delay (scrambled scenes) between high and for low WM load were output as individual subject maps in Talairach space with 2 mm isotropic resolution and thresholded using a false discovery rate of q = 0.01. They were then imported into BESA Research 7.0 as functional activation weight maps for constrained dipole source analysis (Scherg, 1990 ).
For each participant, the scalp positions of the electrodes used in the simultaneous EEG-fMRI scanning sessions were estimated initially using an approximation of locations from a standard montage template (BESA-MRI-Standard-Electrodes) and then adjusted manually based on visual inspection of the indentation-artifacts caused by electrode on the scalp, which appeared as dips on the scalp surface reconstructions. An example of electrode locations for a single subject is shown in Supplementary Figure 2. Each participant’s T1-weighted anatomical MRI was segmented manually in BESA MRI v2.0 to create a 4-layer Finite Element Model (FEM) realistic head model to be used in the source analysis. Based on individual electrode coordinates, segmentation with anatomical landmarks transformed to Talairach Space, and fMRI statistical maps imported for each condition, BESA calculated the best fitting ellipsoid of each participant (Scherg, 1992 (link)). The fMRI-informed regional EEG source estimation with anatomical constraints approach has been documented to be a better modeling than seeding dipoles based solely on anatomical locations (Phillips et al., 2002 (link); Ahlfors and Simpson, 2004 (link); Ou et al., 2010 (link)).
Seed-based dipole fitting was based on a priori hypotheses to explain ERP changes as a function of task period and WM load. For encoding, two equivalent dipoles were fitted onto bilateral parahippocampal cortex (PHC) for each participant at low and high load WM conditions. For delay, two equivalent dipoles were fitted onto bilateral thalamus. For each participant, a time window was chosen from onset to the peak of the first Global Field Potential (GFP) peak, which is a measure for spatial standard deviation as a function of time (Strik and Lehmann, 1993 (link)). An example of a single participant’s GFP waveform is shown in Supplementary Figure 3 and an example analysis window used in the source analysis is shown in Supplementary Figure 4. During seeding of dipole locations, weighting with fMRI activation maps was initially turned off to avoid potential bias in determining the initial seed location. The dipoles were then fit onto the respective sources weighted by the fMRI statistical map activation using the RAP-MUSIC algorithm as implemented in BESA source space that estimates the dipole locations using the weighted fMRI images (Grech et al., 2008 (link)). The dipole positions were constrained to stay within the target regions, but their orientations were kept free before the fit. All the dipoles fell within the appropriate brain regions (PHC and thalamus) after the fit. An example fit with fMRI weighting for thalamus is shown in Supplementary Figure 5. The dipole positions were expressed as Talairach coordinates in units of millimeters (mm) and averaged across all subjects. The source waveforms for each participant and condition were exported and then imported for group source statistical analyses in BESA Statistics v2.0.