Detection and assessment of the severity of Levodopa-induced dyskinesia in patients with Parkinson's disease by neural networks

Abstract

Levodopa-induced dyskinesias (LID) in Parkinson's disease (PD) have remained a clinical challenge. We evaluated the feasibility of neural networks to detect LID and to quantify their severity in 16 patients with PD at rest and during various activities of daily living. The movements of the patients were measured using four pairs of accelerometers mounted on the wrist, upper arm, trunk, and leg on the most affected side. Using parameters obtained from the accelerometer signals, neural networks were trained to detect and to classify LID corresponding to the modified Abnormal Involuntary Movement Scale. Important parameters for classification appeared to be the mean segment velocity and the cross-correlation between accelerometers on the arm, trunk, and leg. Neural networks were able to distinguish voluntary movements from LID and to assess the severity of LID in various activities. Based on the results in this study, we conclude that neural networks are a valid and reliable method to detect and to assess the severity of LID corresponding to the modified Abnormal Involuntary Movement Scale.