Prediction of Movement Ratings and Deep Brain Stimulation Parameters in Idiopathic Parkinson's Disease

Abstract

Background: Deep brain stimulation (DBS) parameter fine-tuning after lead implantation is laborious work because of the almost uncountable possible combinations. Patients and practitioners often gain the perception that assistive devices could be beneficial for adjusting settings effectively.

Objective: We aimed at a proof-of-principle study to assess the benefits of noninvasive movement recordings as a means to predict best DBS settings.

Materials and methods: For this study, 32 patients with idiopathic Parkinson's disease, under chronic subthalamic nucleus stimulation with directional leads, were recorded. During monopolar review, each available contact was activated with currents between 0.5 and 5 mA, and diadochokinesia, rigidity, and tapping ability were rated clinically. Moreover, participants' movements were measured during four simple hand movement tasks while wearing a commercially available armband carrying an inertial measurement unit (IMU). We trained random forest models to learn the relations between clinical ratings, electrode settings, and movement features obtained from the IMU.

Results: Firstly, we could show that clinical mobility ratings can be predicted from IMU features with correlations of up to r = 0.68 between true and predicted values. Secondly, these features also enabled a prediction of DBS parameters, which showed correlations of up to approximately r = 0.8 with clinically optimal DBS settings and were associated with congruent volumes of tissue activated.

Conclusion: Movement recordings from customer-grade mobile IMU carrying devices are promising candidates, not only for remote symptom assessment but also for closed-loop DBS parameter adjustment, and could thus extend the list of available aids for effective programming beyond imaging techniques.

Keywords: DBS programming; Parkinson’s; deep brain stimulation; machine learning; subthalamic nucleus.