Early Identification of Parkinson's Disease from Hand-drawn Images using Histogram of Oriented Gradients and Machine Learning Techniques

Abstract

Parkinson's disease is one of the supreme neurodegenerative problems of the human's vital nervous organism. It is a matter of sorrow that no specific clinical tests were introduced to detect Parkinson's disease correctly. As Parkinson's disease is non-communicable, early-stage detection of Parkinson's can prevent further damages in humans suffering from it. However, it has been observed that PD's presence in a human is related to its hand-writing as well as hand-drawn subjects. From that perspective, several techniques have been proposed by researchers to detect Parkinson's disease from hand-drawn images of suspected people. But, the previous methods have their constraints. In this investigation, an approach to predict Parkinson's disease from hand-drawn wave and spiral images using computer vision and machine learning techniques has been recommended. Decision Tree, Gradient Boosting, K-Nearest Neighbor, Random Forest, and some other classification algorithms with the HOG feature descriptor algorithm was applied. The proposed strategy with Gradient Boosting and K-Nearest Neighbors accomplished better execution in accuracy, sensitivity, and specificity as well as in system design flexibility. Gradient Boosting algorithm got 86.67%, 93.33%, and 80.33% for accuracy, sensitivity, specificity and KNN got 89.33%, and 91.67% for accuracy, and sensitivity respectively.