Probabilistic maps for deep brain stimulation – Impact of methodological differences

Group analysis of patients with deep brain stimulation (DBS) has the potential to help understand and optimize the treatment of patients with movement disorders. Probabilistic stimulation maps (PSM) are commonly used to analyze the correlation between tissue stimulation and symptomatic effect but are applied with different methodological variations. To compute a group-specific MRI template and PSMs for investigating the impact of PSM model parameters. Improvement and occurrence of dizziness in 68 essential tremor patients implanted in caudal zona incerta were analyzed. The input data includes the best parameters for each electrode contact (screening), and the clinically used settings. Patient-specific electric field simulations (n = 488) were computed for all DBS settings. The electric fields were transformed to a group-specific MRI template for analysis and visualization. The different comparisons were based on PSMs representing occurrence (N-map), mean improvement (M-map), weighted mean improvement (wM-map), and voxel-wise t-statistics (p-map). These maps were used to investigate the impact from input data (clinical/screening settings), clustering methods, sampling resolution, and weighting function. Screening or clinical settings showed the largest impacts on the PSMs. The average differences of wM-maps were 12.4 and 18.2% points for the left and right sides respectively. Extracting clusters based on wM-map or p-map showed notable variation in volumes, while positioning was similar. The impact on the PSMs was small from weighting functions, except for a clear shift in the positioning of the wM-map clusters. The distribution of the input data and the clustering method are most important to consider when creating PSMs for studying the relationship between anatomy and DBS outcome. [Display omitted] •A high-resolution patient-specific template of the deep brain anatomy was created.•Probabilistic stimulation maps (PSM) were developed for improvement and side effect.•Distribution of input data is an important factor when creating PSMs.•Response clusters are highly dependent on the analysis method.•Weighting functions and high resolution has less importance in resulting PSMs.