EEG-based motor imagery classification using neuro-fuzzy prediction and wavelet fractal features

• Published in: Journal of neuroscience methods Vol. 189; no. 2; pp. 295 - 302
• Main Author: Hsu, Wei-Yen
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 15.06.2010
• Subjects: Adaptive neuro-fuzzy inference system (ANFIS); Algorithms; Area Under Curve; Brain - physiology; Brain–computer interface (BCI); Databases as Topic; Discriminant Analysis; Electroencephalography (EEG); Electroencephalography - methods; Fractals; Functional Laterality; Fuzzy Logic; Humans; Imagination - physiology; Linear Models; Modified fractal dimension; Motor Activity - physiology; Motor imagery (MI); Neural Networks (Computer); ROC Curve; Signal Processing, Computer-Assisted; Time Factors; Time-series prediction; User-Computer Interface; Time-series prediction; Motor imagery (MI); Electroencephalography (EEG); Modified fractal dimension; Adaptive neuro-fuzzy inference system (ANFIS); Brain–computer interface (BCI)
• ISSN: 0165-0270; 1872-678X; 1872-678X
• DOI: 10.1016/j.jneumeth.2010.03.030

In this paper, a feature extraction method through the time-series prediction based on the adaptive neuro-fuzzy inference system (ANFIS) is proposed for brain–computer interface (BCI) applications. The ANFIS time-series prediction together with multiresolution fractal feature vectors (MFFVs) is applied for feature extraction in motor imagery (MI) classification. The features are extracted from the electroencephalography (EEG) signals recorded from subjects performing left and right MI. Two ANFISs are trained to perform time-series predictions for respective left and right MI data. Features obtained from the difference of MFFVs between the predicted and actual signals are then calculated through a window of EEG signals. Finally, a simple linear classifier, namely linear discriminant analysis (LDA), is used for classification. The proposed method is estimated with classification accuracy and the area under the receiver operating characteristics curve (AUC) on six subjects from two data sets. I also assess the performance of proposed method by comparing it with well-known linear adaptive autoregressive (AAR) model, AAR time-series prediction, and neural network (NN) time-series prediction. The results indicate that ANFIS time-series prediction together with MFFV features is a promising method in MI classification.