Probabilistic mapping of the antiparkinsonian effects of pallidal deep brain stimulation

Deep brain stimulation of the internal globus pallidus effectively treats motor symptoms and drug-induced fluctuations in advanced Parkinson's disease-but is complicated by a high variability in outcomes. In a minority of patients, even stimulation-induced aggravation of akinesia has been reported. Previous works and clinical practice suppose an antiparkinsonian sweetspot in a sensorimotor segment in ventrocaudal internal globus pallidus. However, the detailed functional anatomy of the nucleus is still a matter of debate. We examined 39 patients with Parkinson's disease undergoing pallidal deep brain stimulation from two centres. Deep brain stimulation outcomes were scaled using the percentage change in the motor part of Unified Parkinson's Disease Rating Scale in MED-OFF pre- and postoperatively. The mean improvement with chronic pallidal deep brain stimulation was 16 ± 37%. Motor symptoms improved in 72% of patients. However, 10% were poor responders (<15% improvement) and 28% experienced a worsening of parkinsonism. After basic data processing, we used individual stimulation parameters to construct volumes of tissue activated and applied an established approach of voxelwise <i>t</i>-statistic to build a probabilistic map of the antiparkinsonian effect of deep brain stimulation in the internal globus pallidus. Subsequently, we sampled the volumes of tissue activated in a leave-one-out fashion and employed a multivariate linear regression model to predict an individual outcome. Using this framework, we explained 72% of the variance in motor outcomes (<i>P</i> < 0.001) by the spatial effect map. The linear model was significantly predictive in the leave-one-out cross-validation (Pearson's <i>R</i> = 0.62; <i>P</i> < 0.001). The pallidal subregion providing the best antiparkinsonian effect-as defined by the most significant voxels of the probabilistic map-was located anteriorly in the posteroventral internal globus pallidus. Furthermore, we used the probabilistic map to simulate a monopolar review in silico in each patient and calculated optimized stimulation parameters, demonstrating the potential use of our model in image-guided programming.