Textural feature of EEG signals as a new biomarker of reward processing in Parkinson’s disease detection

Background: Parkinson’s disease (PD) detection holds great potential for providing effective treatments, slowing the disease process, and improving the quality of patient’s life, but the development of a clinical accurate, generalized, robust and cost-effective method is a challenge. Method: In this paper, a novel PD detection method based on textural features of clinical electroencephalogram (EEG) signals has been proposed. In contrast to most existing methods, which do not consider reward positivity (RP)-relevant features for automatic PD detection, this method has focused on providing a novel EEG marker of RP using an enhanced time-frequency representation, texture descriptors based on Gray Level Co-occurrence Matrix, local binary pattern, and sparse coding classifier. Results: The proposed method has been evaluated using EEG signals recorded during a reinforcement-learning task from 28 patients with PD and 28 sex- and age matched healthy controls. Results have demonstrated that the proposed architecture reaches a high detection with an average accuracy rate of 100%, presenting better performance and outperforming previous techniques. Conclusions: it can provide a new solution to detect RP changes in PD and can offer obvious stability advantages on several clinical and technical variables (medication states, type of textural descriptors, reduced channels), suggesting a generalizable detection system.