FV 11. Prediction of mobility ratings and DBS electrode parameters in idiopathic Parkinson's disease

Abstract

Introduction. Idiopathic Parkinson Syndrome (iPS) is a chronic neurodegenerative disorder with rising prevalence. Initially satisfactory drug treatment often gets insufficient later on giving rise to hypokinetic motoric states (OFF) several times a day. Deep Brain stimulation (DBS) is a treatment alternative especially in later stages of the disease. Yet, DBS settings require adjustment by specialized physicians which is time consuming and can be exhausting for patients. Mobility measurements from mobile devices with an inertial measurement unit (IMU) could enable more efficient detection of DBS parameters resulting in optimal symptom control. Method. We measured 23 patients suffering from iPS with chronic DBS lead placement in the subthalamicus nucleus. During consequent monopolar testing, patients were rated at every setting for tremor, bradykinesia and rigidity and performed four hand motor tasks: i) tapping, ii) diadochokinesia, iii) posture holding and iv) rest. Linear acceleration and angular velocity of the forearm were measured with a commercially available armband carrying an IMU. Finally DBS settings were set to optimal values by an experienced physician. We used random forest models to predict 1. clinical mobility ratings and 2. The importance of each electrode in the optimal DBS setting. Models were trained on 80% of the data while 20% served for validation. Results. We used correlations between true and predicted values for model assessment. IMU measurements from the diadochokinesia task were associated with the best prediction of clinical mobility ratings with a maximum correlation of .678 ( p < .001, r 2 = .46) for the prediction of clinical diadochokinesia ratings. Correlations between true and predicted stimulation strength of electrodes in the optimal settings were lower but still substantial. Here IMU data from the tapping task had the best predictive value ( r = .57, p < .001, r 2 = .32). Discussion. We present an automated approach to mobility rating and optimal DBS parameter estimation in patients suffering from iPS. We show that clinical mobility ratings can be predicted with substantial validity using inexpensive IMUs. We also show that one can estimate the role of a particular electrode in an optimal DBS setting according to a specialized physicians adjustment. Our results may pave the way for implementation of machine learning algorithms enabling self-adjusting DBS devices in the future. Fig. 1. Correlations between true and predicted ratings of diadochokinesia, tapping and rigor. Colors show from which task IMU data was used for prediction (dd: diadochokinesia, hold: holding, rest: resting tap: tapping). Errorbars are 95% confidence intervals. Fig. 2. Correlations between true and predicted optimal current (mA) of a particular electrode contact. Groups show from which task IMU data was used for prediction (dd: diadochokinesia, hold: holding, rest: resting tap: tapping). Errorbars are 95% confidence intervals.