A Quantitative Assessment of Bradykinesia Using Inertial Measurement Unit

Abstract

Parkinson’s disease (PD) is a neurodegenerative disorder known to affect movement. Approximately seven million people around the world suffer from PD [1]. Tremor in one hand characterizes the onset of PD. Population suffering with PD shows symptoms of slowed movement. Consequently, PD patients struggle to complete a simple task like picking up a book. This slowness of movement is called bradykinesia. Bradykinesia measurement is vital for monitoring the progression of PD. The current method of assessing bradykinesia requires patients to perform certain motor tasks in clinical settings. A Unified Parkinson Disease Rating Scale (UPDRS) score is assigned to each task based on the observation by a physician. However, PD patients do not always show natural symptoms during a clinical visit. Also, subjective bias occurs during such assessment of bradykinesia [2]. To overcome these limitations, several attempts have been made to quantify bradykinesia using wearable sensors [3]. Accelerometer, gyroscope or a combination of both have been employed for acquisition of movement data to evaluate bradykinesia [3]. Time domain parameters derived from sensor signals for characterizing bradykinesia which includes speed, amplitude, hesitations, and halt have been evaluated in previous studies. However, the effect of frequency domain parameters and non-linear features extracted from sensor signals for evaluating the severity of bradykinesia is unknown. Whether or not it leads to an improvement in the assessment of bradykinesia needs to be investigated. It is known that the patients suffering from severe bradykinesia have their movement signal distorted due to unpredictable movement or hesitation. Nonlinear features can characterize the degree of complexity and provide further relevant insights regarding the severity of bradykinesia. In this study, we investigated the efficacy of various frequency-domain and nonlinear parameters to quantify bradykinesia. The objective was to develop a predictive model based on a combination of sophisticated linear (frequency) and non-linear features derived from the sensor signal which has not been previously explored in the literature.