A Hybrid Signal Processing and Deep Learning Framework for Parkinson's Disease Classification and Determination of ON and OFF Medication Condition.
A Hybrid Signal Processing and Deep Learning Framework for Parkinson's Disease Classification and Determination of ON and OFF Medication Condition.
Where did the research take place?
The study site has not been established. Author addresses may differ from where the research occurred.
Doha, QA · Author affiliation
Department of Electrical Engineering, Qatar University, Doha, Qatar.Location evidence
A plain-language reading has not been prepared for this paper yet.
Original abstract
INTRODUCTION/OBJECTIVE: Parkinson's disease (PD) is a neurodegenerative disease that can cause motor or non-motor symptoms as a result of damage to nerve cells in the brain. Recently, studies on the diagnosis and prediction of this type of disease continue with the application of machine learning (ML) and deep learning (DL) methods to electroencephalogram (EEG) signals. This study proposes a novel diagnostic framework for classifying PD across both ON and OFF medication states, leveraging EEG-based signal analysis and DL architectures. METHODS: In the proposed approach, EEG signals were obtained from 15 patients with PD (ON and OFF medication), with a mean age of 62.60, and 16 healthy controls (HC), with a mean age of 63,50. Data augmentation was achieved by decomposing EEG signals into subbands using Variational Mode Decomposition (VMD) as a signal processing method. Spectrogram images were obtained by applying the Wavelet Coherence (WC) method to these subbands. The obtained images were classified using a 2D convolutional neural network (2D-CNN). RESULTS: Following the classification process, accuracy, sensitivity, specificity, and F1-Score values were obtained and interpreted. In this study, which examined subband performance, the best results were achieved with 99.80% accuracy, 99.79% sensitivity, 99.80% specificity, and 99.79% F1-Score. DISCUSSION: The findings of this study confirm that the proposed approach outperforms similar approaches in the literature. Given the clinical need for reliable, non-invasive methods for diagnosing Parkinson's disease, the success of the developed model represents a significant contribution to the existing literature. CONCLUSION: This study has contributed a high-performance, innovative approach to the existing literature on Parkinson's disease diagnosis. The successful results obtained clearly demonstrate that the model developed can serve as a robust and reliable framework not only for the diagnosis of Parkinson's disease, but also for similar biomedical signal analyses and the investigation of various neurological disorders.