Early Diagnosis of Parkinson’s Disease Using Hand Drawings Images

Abstract

This work presented a brief of the most recent technologies that tried to early diagnose Parkinson’s Disease (PD). Early diagnosis of PD can help to detect its symptoms and evolution which increases the chances of treatment because there’s no treatment for PD. So, we compiled several papers published in some well-established databases, all of which have almost the same diagnosis phases: preprocessing and classification and a few papers applied augmentation before classification. Each one has been mentioned to identify its objective, methodology, and results. Our brief showed that most papers used different approaches that use machine learning mechanisms to perform the automatic PD diagnosis. The main contribution of this paper is to propose an analyzed survey for the most recent proposed computer-assisted diagnosis systems mainly from 2018 to 2021, and how they can affect while handling the PD identification issue, and there are two experiments done using HandPD and NewHandPD datasets: the first experiment done using hybrid classification approach based on both SVM and PCA, while the second done using our modified CNN architecture, the modification done by changing of the number of filters: using 30 convolutional layers instead of 10, 50 pooling layers instead of 20 and 250 neurons at dense layer instead of 100. HandPD gave us the best accuracy using our modified CNN by 100.00% which is much better compared with “hybrid classification approach” mentioned above & “CNN model classification based”.