A comparative study on classification of magnetoencephalography signals using probabilistic neural network and multilaye
- PDF / 882,541 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 64 Downloads / 207 Views
(0123456789().,-volV)(0123456789(). ,- volV)
METHODOLOGIES AND APPLICATION
A comparative study on classification of magnetoencephalography signals using probabilistic neural network and multilayer neural network Onursal Cetin1
•
Feyzullah Temurtas1
Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Visual decoding is a critical way to understand the face perception mechanism of the brain in the neuroscience field. Magnetoencephalography (MEG) is a completely noninvasive measurement technique that provides information about brain cortical functioning using the magnetic field of neuronal electrical activity. These neuromagnetic signals measured by MEG may emerge as a result of a visual stimulus during the brain decoding process for the face perception mechanism. In this research study, two classes of visual stimuli (face/scrambled face), including 9414 trials in total, were used. In order to obtain meaningful data from noisy MEG recordings, feature extraction approaches and classification systems are required. For the purpose of feature extraction, the Riemannian approach, characterized by its competitive nature, has been used. A probabilistic neural network and a multilayer neural network structures were proposed to classify magnetoencephalography signals. The obtained results were presented comparatively with the results of prior studies using the same dataset. The classification accuracies of 82.36% and 77.78% were achieved for the probabilistic neural network and multilayer neural network, respectively. Moreover, the probabilistic neural network classifier could be expected to be an alternative method to other competing methods. Because PNN does not use the back-propagation algorithm, so there is no need to train the network with the whole dataset. Thus, the classification process is performed faster. Keywords Magnetoencephalography Riemannian approach Probabilistic neural network Multilayer neural network Classification
1 Introduction Brain activities produce an electrical potential in the scalp, as well as creating a magnetic field on the head. While the electrical potential is recorded with electroencephalography (EEG), the magnetic field is recorded using a relatively new technique, magnetoencephalography (MEG) (Cohen and Cuffin 1983). Even though produced through the same brain activity, the MEG map shows a different spatial pattern than does an EEG map. Those pattern variations may provide new information for brain decoding (Cohen and Cuffin 1983). Brain decoding, which is of great interest by Communicated by V. Loia. & Onursal Cetin [email protected] 1
Electronics and Communication Engineering Department, Bandırma Onyedi Eylu¨l University, Balıkesir, Turkey
scientific communities, is a critical data analysis approach that investigates the relationship between stimuli and brain activity (Olivetti et al. 2014). Researchers have been trying to predict the class of visual stimuli presented to a subject by decoding mental activity (Caliskan et al. 2017). Magnetoencephalo
Data Loading...
