Quantifying Levodopa-Induced Dyskinesia Using Depth Camera

Maria Dyshel, David Arkadir, Hagai Bergman, Daphna Weinshall; Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops, 2015, pp. 119-126

We present a novel method to detect and assess the severity of Levodopa-Induced Dyskinesia (LID) in Parkinson's Disease (PD) patients, based on Microsoft Kinect recordings of the patients. Dyskinesia denotes involuntary movements induced by chronic treatment with levodopa in patients with PD. Detection and objective quantification of dyskinesia is essential for optimizing the medication regime and developing novel treatments for PD. We used Microsoft Kinect sensor to track limb and neck movements of a patient performing two motor tasks. Using a new motion segmentation algorithm, kinematic features were extracted from the videos and classified using Support Vector Machines (SVMs). The method was tested on 25 recordings of 9 PD patients, and achieved sensitivity of 0.82 at EER in overall dyskinesia detection. Moreover, it provided a numerical overall score for the severity of dyskinesia, which showed high correlation with the neurologist's assessment of the patient's state. The study shows that depth camera recordings can be used to monitor and grade the severity of levodopa-induced dyskinesia, and therefore can potentially provide valuable aid to clinicians and researchers.