Automatic assessment of Parkinson's patients' dyskinesia using non-invasive machine learning methods

Paraskevi Antonia Theofilou, Georgios Tsatiris, Stefanos D. Kollias

Abstract: Parkinson's disease is a serious neurodegenerative disorder that causes loss of control of physical movements. This disease is not curable, however its development can be limited, depending on its stages, mainly with medication and specifically with the use of levodopa. However, many times, patients not only experience the symptoms of the disease, but also the disorders caused by long-term administration of levodopa. The detection of various symptoms and disorders is based on the clinical examination of patients with their evaluation according to the UPDRS and UDysRS scales. Therefore, the present work aims to develop a system for the automatic assessment of levodopa-induced dyskinesia according to the UDysRS metric and the use of non-invasive machine learning methods. Video data are used and after the proper processing of the features resulting from their frames, enter a simple and at the same time efficient CNN to be evaluated with efficiency based on the six values of UDysRS.