Automated Detection of Parkinson’s Disease through Kinematic and Geometric Analysis of Handwritten Strokes using Deep Learning
TL;DR
A novel Deep Learning model based on DenseNet121 for automated detection of PD using spiral and meander handwriting images is introduced and is able to achieve 92.8% accuracy, which implies a reliable classification of PD and healthy samples.
Abstract
Parkinson's Disease (PD) is a progressive brain disorder that affects human motor functions. Early Parkinson's Disease (PD) detection plays a major role in facilitating clinical treatment. Handwriting analysis has proven to be an efficient way to detect motor problems; nonetheless, traditional methods have shown poor feature representation capability and classification robustness. In this study, a novel Deep Learning (DL) model based on DenseNet121 for automated detection of PD using spiral and meander handwriting images is introduced. Image processing, hierarchical features extraction, Global Average Pooling technique to produce 2,880-dimensional feature representation, Spatial Dropout technique, and binary classification are included in the proposed methodology. Handwriting benchmark datasets such as PaHaW, HandPD, and NewHandPD are used to test the performance of the proposed framework. The experimental findings show that the presented method is able to achieve 92.8% accuracy, which implies a reliable classification of PD and healthy samples.