The merit of group-level analyses in TMS targeting: group-based task-driven TMS coil positioning can outperform individual approaches - evidence from resting-state-, diffusion-, and functional-MR

Individualized TMS coil positioning enhances the precision of targeting a specific brain region. The gold standard method for this is individual task based fMRI guided TMS. Alternatively, individual resting state fMRI (rs-fMRI) or the combination of rs-fMRI and Diffusion Weighted Imaging (DWI) can be used to guide TMS coil positioning. Another option is to use group MRI maps instead of individual maps. A last possibility is the atlas-based approach, which utilizes group data from previous studies, transferred to atlas space, e.g. Talairach coordinates (TAL). We examined all these positioning approaches while localizing the core nodes of the Dorsal Attention Network (DAN) as well as Ventral Attention Network (VAN). We collected task-based fMRI data from a simple oculomotor task, rs-fMRI data with eyes open, and structural data with DWI (n = 20). The task fMRI data were analyzed with a GLM, the rs-fMRI with ICA, and the DWI data were processed with the TractSeg algorithm. Data from rs-fMRI and DWI were combined to constrain end-point localization. Each approach yielded individual region of interest (ROIs) and group ROIs. First, individual TMS target points were computed by determining the center of gravity within each resulting ROI. Success rates for all TMS target localization approaches and regions were obtained. There was variability per approach and area, with task-based fMRI showing a slight advantage for most regions. Using individual task fMRI coordinates as the reference, localization accuracy was measured as the Euclidean distances from this gold standard. Group-level TMS target points were projected into individual space, after which the localization accuracy was similarly ascertained. Task-based group data (leave-one-out) yielded the closest approximation to individual task-based targets (mean difference = 7.57 mm), while individual resting-state data deviated by >10 mm. Atlas-based approaches performed worst, with deviations >15 mm. These findings highlight spatial discrepancies between TMS target localization approaches and suggest that individual data do not consistently outperform group data.