Non-motor and motor symptoms that are linked with Parkinson's disease are often clinically assessed by neurologists using the Unified Parkinson's Disease Rating Scale (UPDRS). UPDRS scores are described as qualitative and are dependent on neurologist's experience. Consequently, clinical scores may differ among neurologists. We develop an application for measuring bradykinesia in the UPDRS finger tapping task, with which patients are recorded with a depth camera and by analyzing videos, given a more objective rating. In the first stage, we detect touches and thumb's and pointer's fingertips. Following, we calculate distances between the fingertips. From distances we then extract finger tapping features. We record a group of people with Parkinson's disease and a control group. Furthermore, we define a model that best separates instances with different UPDRS scores. Considering the small number of training data, the model successfully separates the instances, however, we need to obtain more data for classification.
|
