Fig. 2. Accuracy of different models compared to actual measurements.

Accuracy is quantified as the Relative Difference Measure (RDM); Lower values indicate greater model accuracy. (a) Plot of RDM between intracranial potentials predicted by a head model and measured potentials averaged across 61 stimulation sites in 4 subjects. Isotropic FEM models differ in their numbers of tissue types (3, 4 and 15 for models ISO_I, II and III; in columns 1–3 from left). In some anisotropic models the anisotropy is assumed to be uniform (middle); these differ in the conductance increase parallel to the fiber direction (parallel/transverse ratios of 2, 5, 7, and 10 for models ANISO_WM_II_a, b, c, and d; in columns 4–7). In other anisotropic models, the degree of anisotropy is estimated from DTI data, and applied to the white matter only (ANI-SO_WM_I; column 8), or to both white and gray matter (ANISO_IC; column 9). Model types are color-coded as indicated below the bar-chart. (b) Error (color-coded RDM) for 7 different model types (columns) and 23 different stimulation sites (rows) in BI18 (averaged across all recording electrodes). Stimulation site is given as the electrode name followed by the contact numbers for monopolar source and sink. (C) RDM between model and experiment when averaged over the 23 stimulation sites in BI18. In both group and individual subject experiments, the highest accuracy was obtained with the ANI-SO_IC model, which estimated anisotropy for both gray and white matter from individual subject DTI. Individual data for subjects BI14, BI15 and BI17 are shown in Supplementary Figure 1