To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Early Parkinson Detection Using Fully-Connected Deep Neural Network based on Vocal Features |
|
Author |
Ahlem Kehili, dabbabi Karim, Cherif Adnen |
| Citation |
Vol. 20 No. 6 pp. 132-143 |
|
Abstract |
Parkinson’s disease (PD) is considered to be a growing neurodegenerative disease characterized by many motors and non-motor specifications. During the early stages of this disease, vocal impairments are usually the disorders that can appear for patients with PD. Thus, diagnosis systems based on vocal perturbations have become at the head of recent studies for the detection of PD. In our study, a PD detection system based on vocal features was proposed using Full-Connected Deep Neural Network (FC-DNN) as a classifier, and Jitter and shimmer variants plus mean fundamental frequency (meanF0), harmonic to-noise ratio (HNR) and duration as vocal features. The experimental tests were performed on Spanish dataset and the results have shown the superiority of FC-DNN in terms of the evaluated performances (Accuracy=100%, precision=99%, Recall=99, F-measure=99.1, and Matthew Correlation (MCC=0.95) with comparison to other tested classifiers of machine learning and to classic approaches. |
|
Keywords |
Deep learning algorithm, machine learning algorithm, early Parkinson detection, vocal features, PD classification, FC-DNN |
|
URL |
