Objective Vision-based Assessment of Parkinsonism and Levodopa-induced Dyskinesia in Persons with Parkinsonâ s Disease

Abstract

Parkinsonâ s disease (PD) is the second most common neurodegenerative disease, with the incidence rate climbing rapidly as the global population ages. While levodopa is effective at treating PD symptoms, prolonged usage can introduce additional motor complications called levodopa-induced dyskinesias (LID). Assessment of PD/LID requires regular visits to a clinic; however, the intermittent nature of assessments can fail to capture changes in a personâ s condition. With computational power becoming increasingly more affordable, computer vision is an accessible means of performing more frequent and objective PD assessments. A study of deep learning for human pose estimation was conducted to examine the feasibility of extracting body movements from videos of PD assessments. Machine learning was applied to detect and estimate the severity of PD/LID using movement features. Results indicate computer vision is a promising candidate for objective PD assessment, thus laying the foundation for an automated system for evaluation of PD motor symptoms.