Digitized spiral drawing classification for Parkinson's disease diagnosis

Abstract

Parkinson's disease (PD) is the most common neurodegenerative disease affecting significantly motor functions of elderly persons. The diagnosis and monitoring of PD is costly and inconvenient process even today, in under developing parts of the world. The observable symptoms of PD at early stage include disorders in handwriting and repetitive tasks of spiral drawing. With advancement of IT it is easier to collect spiral drawing samples using digitized tablet. We proposed detailed analysis of Static and dynamic spirals drawn by PD patients. For this, in-air and on-surface kinematic variables are taken out from data files generated for 25 patients and 15 healthy controls, using mathematical models. Results demonstrated nearly 91% classification accuracy to separate PD patients from healthy controls by applying feature engineering and four machine learning (ML) classifiers Logistic Regression, C-Support Vector Classification (SVC), K- nearest neighbor(KNN) classifier and ensemble model Random Forest Classifier(RFC). This paper confirms that digitized spiral drawings have major impact on classification of PD patients and healthy controls and hence can support future differential diagnosis of PD.