Estimating Bradykinesia in Parkinson's Disease with a Minimum Number of Wearable Sensors

Monitoring motor function of patients with Parkinson's disease (PD) over long periods of time is essential in order to improve symptom management and avoid complications. Wearable technologies can be useful in this context as long as they do not unnecessarily increase patient and caregiver burden. The goal of the current study was to identify whether using more wearable sensors improved the estimation of whole-body bradykinesia scores. Ten patients diagnosed with idiopathic PD were recruited to take part in this study. Data was collected over 3 separate occasions using clinical evaluations and three-axis acceleration. In order to estimate the clinical scores associated with bradykinesia of the upper-and lower-limbs, a machine learning algorithm using a leave-one-subject-out cross-validation paradigm was implemented. Using two sensors per limb did not improve estimation error within the upper-or lower-limbs. Results demonstrate that the use of multiple sensors on a single limb does not significantly improve the estimation of clinical scores related to bradykinesia. However, in order to obtain whole-body limb-specific bradykinesia scores, a minimum of one wearable sensor per limb is required.