Deep Convolutional Neural Network for Parkinson’s Disease Based Handwriting Screening
Mohamed Shaban
Abstract
In this paper, the use of a fine-tuned VGG-19 for screening Parkinson’s Disease (PD) based on a Kaggle handwriting dataset is investigated and experimented. The dataset including 102 wave and 102 spiral handwriting patterns was pre-processed where images were resized and a data augmentation based on image rotation was adopted to minimize overfitting. The Convolutional Neural Network (CNN) model was then trained on the pre-processed dataset and validated using both 4-fold and 10-fold cross validation techniques. The CNN model achieved an accuracy of 88%, 89%, and a sensitivity of 89%, 87% on the wave and spiral patterns respectively when a 10-fold cross validation was used. The proposed approach offers a promising solution for assessing and screening PD based on handwriting drawings and achieves a comparable high performance on the two different handwriting patterns as compared with the-state-of-the-art architecture that adopted a fine-tuned AlexNet.