Brain connectivity profiles seeding from deep brain stimulation (DBS) electrodes have emerged as informative tools to estimate outcome variability across DBS patients. Given the limitations of acquiring and processing patient-specific diffusion-weighted imaging data, a number of studies have employed normative atlases of the human connectome. To date, it remains unclear whether patient-specific connectivity information would strengthen the accuracy of such analyses. Here, we compared similarities and differences between patient-specific, disease-matched and normative structural connectivity data and their ability to predict clinical improvement.

Data from 33 patients suffering from Parkinson’s Disease who underwent surgery at three different centers were retrospectively collected. Stimulation-dependent connectivity profiles seeding from active contacts were estimated using three modalities, namely patient-specific diffusion-MRI data, age- and disease-matched or normative group connectome data (acquired in healthy young subjects). Based on these profiles, models of optimal connectivity were calculated and used to estimate clinical improvement in out of sample data.

All three modalities resulted in highly similar optimal connectivity profiles that could largely reproduce findings from prior research based on this present novel multi-center cohort. In a data-driven approach that estimated optimal whole-brain connectivity profiles, out-of-sample predictions of clinical improvements were calculated. Using either patient-specific connectivity (R = 0.43 at p = 0.001), an age- and disease-matched group connectome (R = 0.25, p = 0.048) and a normative connectome based on healthy/young subjects (R = 0.31 at p = 0.028), significant predictions could be made.

Our results of patient-specific connectivity and normative connectomes lead to similar main conclusions about which brain areas are associated with clinical improvement. Still, although results were not significantly different, they hint at the fact that patient-specific connectivity may bear the potential of explaining slightly more variance than group connectomes. Furthermore, use of normative connectomes involves datasets with high signal-to-noise acquired on specialized MRI hardware, while clinical datasets as the ones used here may not exactly match their quality. Our findings support the role of DBS electrode connectivity profiles as a promising method to investigate DBS effects and to potentially guide DBS programming.