Parkinson's disease effective biomarkers based on Hjorth features improved by machine learning

Parkinson's disease (PD) is the second most common neurodegenerative condition in the world and is caused by reduced levels of dopamine in the central nervous system. The diagnosis of PD is a difficult and time-consuming task, and there is no definitive protocol for achieving it. Therefore, several studies have been performed in order to find reliable PD biomarkers. The analysis of characteristics of electroencephalogram (EEG) signals is one of the techniques that have been used in the search for biomarkers. EEG signals capture the activity of neurons through electrodes placed on the scalp and with the advancement of Artificial Intelligence (AI) techniques, their characteristics have started to be used in machine learning (ML) algorithms for the automatic diagnosis of brain diseases, suggesting that EEG signals are promising biomarkers that could be used for automatic identification of individuals with PD. Thus, this work evaluates the performance of Hjorth features obtained from electroencephalographic signals, as biomarkers for Parkinson's disease. Using the database available at the public repository called The Patient Repository for EEG Data + Computational Tools (PRED + CT), we analyzed EEG data from PD individuals periodically exposed to auditory stimuli. The analysis of the proposed biomarkers showed differences between healthy and PD patients in parietal, frontal, central, and occipital lobes. For classification the Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest algorithms were used followed by a 5-fold cross-validation methodology. The proposed model achieved an accuracy of 89.56% when differentiating patients with PD and healthy individuals with an SVM classifier. The results suggest that the Hjorth features extracted from EEG signals could be used as PD biomarkers.