Automated assessment of symptom severity changes during deep brain stimulation (DBS) therapy for Parkinson's disease

Abstract

Deep brain stimulation (DBS) is currently being used as a treatment for symptoms of Parkinson's disease (PD). Tracking symptom severity progression and deciding the optimal stimulation parameters for people with PD is extremely difficult. This study presents a sensor system that can quantify the three cardinal motor symptoms of PD - rigidity, bradykinesia and tremor. The first phase of this study assesses whether data recorded from the system during physical examinations can be used to correlate to clinician's severity score using supervised machine learning (ML) models. The second phase concludes whether the sensor system can distinguish differences before and after DBS optimisation by a clinician when Unified Parkinson's Disease Rating Scale (UPDRS) scores did not change. An average accuracy of 90.9 % was achieved by the best ML models in the first phase, when correlating sensor data to clinician's scores. Adding on to this, in the second phase of the study, the sensor system was able to pick up discernible differences before and after DBS optimisation sessions in instances where UPDRS scores did not change.