TY - JOUR
T1 - Automatic electroencephalographic information classifier based on recurrent neural networks
AU - Alfaro-Ponce, Mariel
AU - Argüelles, Amadeo
AU - Chairez, Isaac
AU - Pérez, Arizbeth
N1 - Publisher Copyright:
© 2018, Springer-Verlag GmbH Germany, part of Springer Nature.
PY - 2019/9/1
Y1 - 2019/9/1
N2 - The aim of this study was to design an automatic classifier for electroencephalographic information (EEGI) registered in evoked potentials experiments. The classifier used a parallel associative memory based on recurrent neural networks (RNNs). Each RNN was trained to classify signals belonging to an individual class. A recurrent method based on the application of Lyapunov controlled functions served to design the training procedure of each RNN in the classifier. A parallel structure of RNN with fixed weights (obtained after training process) performed the validation stage. This structure formed a classifier assemble. The selected class assigned to a new segment of EEGI signal is estimated by the minimum value of the least mean square error among the RNNs forming the assemble. The generalization-regularization and a k-fold cross validation (k= 5) were the validation methods evaluating the classifier efficiency. The confusion matrix method justified the application of the classification method introduced in this study. The EEGI obtained from two different annotated databases served to test the classifier based on RNNs. The first database contained signals divided in five different classes and collected from patients suffering from epilepsy. The second database has 90 signals divided in three classes that corresponded to EEGI signals corresponding to 3 different visual evoked potentials. The pattern classifier achieved a maximum correct classification percentage of 97.2% using the information of both databases. This value prevailed over results reported in similar studies using the first database. In comparison with other pattern recognition algorithms, the proposed RNNs based classifier attained similar or even better correct classification results.
AB - The aim of this study was to design an automatic classifier for electroencephalographic information (EEGI) registered in evoked potentials experiments. The classifier used a parallel associative memory based on recurrent neural networks (RNNs). Each RNN was trained to classify signals belonging to an individual class. A recurrent method based on the application of Lyapunov controlled functions served to design the training procedure of each RNN in the classifier. A parallel structure of RNN with fixed weights (obtained after training process) performed the validation stage. This structure formed a classifier assemble. The selected class assigned to a new segment of EEGI signal is estimated by the minimum value of the least mean square error among the RNNs forming the assemble. The generalization-regularization and a k-fold cross validation (k= 5) were the validation methods evaluating the classifier efficiency. The confusion matrix method justified the application of the classification method introduced in this study. The EEGI obtained from two different annotated databases served to test the classifier based on RNNs. The first database contained signals divided in five different classes and collected from patients suffering from epilepsy. The second database has 90 signals divided in three classes that corresponded to EEGI signals corresponding to 3 different visual evoked potentials. The pattern classifier achieved a maximum correct classification percentage of 97.2% using the information of both databases. This value prevailed over results reported in similar studies using the first database. In comparison with other pattern recognition algorithms, the proposed RNNs based classifier attained similar or even better correct classification results.
KW - Biosignal automatic classifier
KW - Classifier assemble
KW - Electroencephalographic information
KW - Parallel associative method
KW - Recurrent neural networks
UR - http://www.scopus.com/inward/record.url?scp=85070714280&partnerID=8YFLogxK
U2 - 10.1007/s13042-018-0867-9
DO - 10.1007/s13042-018-0867-9
M3 - Artículo
SN - 1868-8071
VL - 10
SP - 2283
EP - 2295
JO - International Journal of Machine Learning and Cybernetics
JF - International Journal of Machine Learning and Cybernetics
IS - 9
ER -