Analysis of Static and Dynamic Attributes of Handwriting as Potential Indicators of Parkinson’s Disease

Abstract

Parkinson's disease is a neuro-degenerated disorder in which level of dopamine chemicals starts decreasing which in return affects the peripheral nervous system of an individual. The peripheral nervous system controls the motor and non-motor neurons and also control messages pass between every muscle from the brain. Traditional techniques detect the disease at the last stage when the patient is suffering from severe conditions. In the Literature, experts give the invasive or non-invasive technique to detect the PD at early stages. In our study, we use handwriting as a diagnostic tool, we use the PaHaW dataset to carried out research and extracting useful information like pen pressure, on-surface, and in-surface movement of stroke, etc. In which, we discuss the effectiveness of both online and offline features of handwriting in characterizing the presence and absence of PD. We exploited different pre-trained CNN models (VGG 19, AlexNet) for feature extraction of offline dataset and later on fed these extracted feature vector to SVM But in case of Online dataset, velocity and acceleration as features are extracted and later fed to various classifier such as, GRU, LSTM, RNN. The obtained finding show the effectiveness of handwriting for the diagnosis of PD at early stages.