Table 1.
Demographic information of patients (contact and voltage) included in the machine learning analysis.
| Patient cohort | Age (mean ± SD) | Gender | Duration of disease (mean ± SD) | UPDRS III (OFF) (mean ± SD) | Levodopa equivalence (mean ± SD) | Days from surgery (mean ± SD) | DBS target | Location of STN electrode (mean MNI coordinates) | Location of GPi electrode (mean MNI coordinates) |
|---|---|---|---|---|---|---|---|---|---|
| Train data—contact (n = 20) | 63 ± 5.7 | 12M 8F | 12.4 ± 4.8 | 34.2 ± 10.4 | 935.6 ± 532.8 | 427.5 ± 488.5 | 18 STN 2 GPi | −13.3, −12.7, −3.6 | −23.2, −5.0, −0.09 |
| Train data—voltage (n = 19) | 61.8 ± 8.4 | 8M 11F | 13.7 ± 5.1 | 38.9 ± 11.3 | 1485.2 ± 662 | 830.8 ± 623.1 | 17 STN 2 GPi | −13.5, −12.6, −3.5 | −20.4, −7.3, −1.5 |
| Test data—pre-programmed (n = 9) | 64.8 ± 9.4 | 8M 1F | 12.4 ± 4 | 28.4 ± 5.3 | 1818.4 ± 1189.1 | 712.1 ± 406.8 | 9 STN | −11.8, −12.7, −5.1 | None |
| Test data— early (n = 9) | 58.5 ± 11.8 | 8M 1F | 9.8 ± 5.4 | 35.8 ± 12.9 | 1450.4 ± 802 | 92.8 ± 94.3 | 8 STN 1 GPi | −12.5, −13.4, −4.4 | −21.2, −7.8, −2.3 |
| Frequency data (n = 10) | 63.2 ± 8.1 | 5M 5F | 13.9 ± 5.8 | 39.4 ± 26.5 | 836.3 ± 407.4 | 1069.36 ± 673.9 | 10 STN | −13.1, −12.2, −3.4 | None |
Patient cohort was divided into three datasets: train, unseen test datasets 1 and 2. Train and unseen dataset 1 include patients a priori clinically optimized. Unseen dataset 2 comprise stimulation naïve patients. Of note, one patient in the unseen dataset 2 (early) underwent the MRI 11.3 months from his surgery, since he had dysphonia after surgery causing a long delay before his programming could be started. Levodopa equivalence was ascertained via the methods of Tomlinson et al. 62. MNI coordinates of active contacts were obtained using Lead-DBS (Lead-DBS v2.0; https://www.lead-dbs.org/) To obtain mean MNI coordinates of active contacts in the frequency data, right-sided electroides were flipped to the left. As the frequency data were acquired with a different paradigm than the contact and voltage data (bilateral, rather than unilateral stimulation), it was not incorporated into the machine learning model.
UPDRS-III Unified Parkinson’s Disease Rating Scale Part III, DBS deep brain stimulation, MNI Montreal Neurological Institute, PD Parkinson’s disease, SD standard deviation, STN subthalamic nucleus, GPi internal globus pallidus.