Category: Parkinson's Disease: Neurophysiology
Objective: We analyze and compare the spectral bands of electroencephalograms (EEGs) associated to motor activation for both patients with Parkinson’s disease (PD) and healthy patients equivalent in age. The study is carried out by combining two Deep Learning techniques: Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). This combination allows to extract invariant characteristics in the spectrum of EEGs that are distinctive of Parkinson’s disease.
Background: An extended methodology of the EEG analysis is performed through its frequency spectrum. In this study we will analyze the spectral bands by means of the spectral entropy, and then we will search for alterations in the entropy that are characteristic of PD.
Entropy is a non-linear parameter known as the disorganization tendency of a system, and it has been studied in EEG when alterations in brain function are detected in different neurological disorders PD included . Application of Deep Learning techniques  to the value of this parameter in EEGs of patients with PD, will allow to obtain characteristic patterns of this disease in the frequency spectrum when the motor zone of both hemispheres is activated.
Method: Deep Learning techniques, specifically Convolutional Neural Network (CNN) and Recurrent Neural Networks (RNN), will be used to analyze the spectral entropy of EEG in patients with PD.
Results: From the analysis of the spectrum of the EEG by means of the spectral entropy, and through Deep Learning techniques, we expect to find invariant characteristics in the spectral bands associated to the PD, so that they automatically classify the degree of progress of the disease in patients. The use of Deep Neuronal Networks allows to find such characteristics in raw EEGs, avoiding intermediary factors that produce alterations in the data.
Conclusion: The learning capacity of the Deep Neuronal Networks allows to generalize the results and to automatically apply the procedure to other EEGs that the network has not seen before. This ability to generalize makes this technique ideal for developing a marker that distinguishes characteristic properties of patients with PD, allowing to develop a methodology that efficiently assesses the degree of disease.
References:  A. Maitín, R. Perezzan, D. Herráez-Aguilar, J. Ignacio Serrano, M. Dolores Castillo, A. Arroyo, J. Andreo, J. P. Romero Muñoz (2020). Time series analysis applied to EEG shows increased global connectivity during motor activation detected in PD patients compared to controls. Preprint.  Bashivan, Pouya Rish, Irina Yeasin, M. Codella, Noel. (2015). Learning Representations from EEG with Deep Recurrent-Convolutional Neural Networks, International Conference on Learning Representation 2016. arXiv preprint 1511.06448.
To cite this abstract in AMA style:A. Maitin, A. Garcia-Tejedor, M. Castillo, I. Serrano, E. Rocon, A. Arroyo, J. Andreo, J. Romero. Application of Deep Neuronal Networks for the extraction of spectral characteristics in patients with Parkinson’s disease [abstract]. Mov Disord. 2020; 35 (suppl 1). https://www.mdsabstracts.org/abstract/application-of-deep-neuronal-networks-for-the-extraction-of-spectral-characteristics-in-patients-with-parkinsons-disease/. Accessed December 5, 2023.
« Back to MDS Virtual Congress 2020
MDS Abstracts - https://www.mdsabstracts.org/abstract/application-of-deep-neuronal-networks-for-the-extraction-of-spectral-characteristics-in-patients-with-parkinsons-disease/