Fig. 3: Probabilistic anatomical mapping and clinical risk stratification analyses for stimulation-induced bradykinesia following GPi-DBS for dystonia.

a Probabilistic “sour spot” map illustrating GPi regions significantly associated with elevated risk (red), based on retrospective cohort voxel-wise analyses. b Combined probabilistic map integrating the bradykinesia “sour spot” (red) and dystonia symptom control “sweet spot” (green), highlighting critical areas of overlap and divergence for optimal clinical DBS programming. c A linear model based on probabilistic map features (derived using a leave-one-out principle in map generation) demonstrated a strong fit in explaining bradykinesia severity, indicating robust internal consistency of the spatial associations (R² = 0.51, p < 0.00001). d Patient-level leave-one-out cross-validation demonstrates clinically relevant predictive power for bradykinesia severity (correlation between predicted and actual severities R² = 0.16, p = 0.0013). e Box plots displaying significant differences among stratified patient risk groups (low, medium, high), particularly prominent between low vs. high (p = 0.0019) and medium vs. high-risk patients (p = 0.0067). f Computational simulations show that machine-assisted reprogramming of DBS settings, guided by the probabilistic risk stratification maps, could theoretically reduce stimulation-induced bradykinesia risk while preserving dystonia motor symptom control (p < 0.001).