Quantitative assessment of parkinsonian tremor based on a linear acceleration extraction algorithm

Abstract Tremor detection plays a crucial role in Parkinson’s disease (PD) treatment and symptom monitoring. The current gold standard for the clinical assessment of parkinsonian tremor is the evaluation using the standard clinical rating scales, which is performed by the well-trained neurologists. However, this assessment approach relies mainly on the subjective judgment of the evaluator. This study, on the basis of the Movement Disorder Society-Sponsored Revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) criteria, proposed a custom quantitative assessment system for parkinsonian tremors. It adopted an attitude estimation-based gradient descent algorithm to separate the linear acceleration (caused by pure translational motion) from the accelerometer output, which combines gravity component. Signal features extracted from the linear accelerations and angular velocities during the tremor tasks were fitted to the clinicians’ ratings with a multiple regression model. Clinical experiments with 34 PD patients and 14 age-matched controls demonstrated that the prediction accuracy was improved by using the decomposed linear acceleration for the extraction of tremor features, which has promoted assessment accuracy compared with the relevant literature (r2 improved from 0.89 to 0.95 for rest tremor, and from 0.90 to 0.93 for postural tremor). In addition, the prediction accuracy was worse when using only the linear accelerations for regression analysis (r2 reduced from 0.95 to 0.87 for rest tremor, and from 0.93 to 0.84 for postural tremor), which means that the effect of rotational motion cannot be ignored in tremor quantification.