A Neural Network-Based Optimal Spatial Filter Design Method for Motor Imagery Classification

Abstract

In this study, a novel spatial filter design method is introduced. Spatial filtering is an important processing step for feature extraction in motor imagery-based brain-computer interfaces. This paper introduces a new motor imagery signal classification method combined with spatial filter optimization. We simultaneously train the spatial filter and the classifier using a neural network approach. The proposed spatial filter network (SFN) is composed of two layers: a spatial filtering layer and a classifier layer. These two layers are linked to each other with non-linear mapping functions. The proposed method addresses two shortcomings of the common spatial patterns (CSP) algorithm. First, CSP aims to maximize the between-classes variance while ignoring the minimization of within-classes variances. Consequently, the features obtained using the CSP method may have large within-classes variances. Second, the maximizing optimization function of CSP increases the classification accuracy indirectly because an independent classifier is used after the CSP method. With SFN, we aimed to maximize the between-classes variance while minimizing within-classes variances and simultaneously optimizing the spatial filter and the classifier. To classify motor imagery EEG signals, we modified the well-known feed-forward structure and derived forward and backward equations that correspond to the proposed structure. We tested our algorithm on simple toy data. Then, we compared the SFN with conventional CSP and its multi-class version, called one-versus-rest CSP, on two data sets from BCI competition III. The evaluation results demonstrate that SFN is a good alternative for classifying motor imagery EEG signals with increased classification accuracy.

CSP-TSM: Optimizing the performance of Riemannian tangent space mapping using common spatial pattern for MI-BCI

“…Research is also being carried out to improve the CSP algorithm. In [65], a neural network-based optimal spatial filter design method has been proposed. The spatial filter and the classifier are trained simultaneously using a neural network named spatial filter network (SFN).…”

Virtual and Actual Humanoid Robot Control with Four-Class Motor-Imagery-Based Optical Brain-Computer Interface

“…Support vector machines with nonlinear kernels may be able to achieve higher classification accuracy than LDA classifiers. The more powerful classification abilities of neural networks may also prove beneficial for improving BCI performance, as has been explored recently with EEG-based BCIs [ 66 – 69 ].…”

Data-Driven Transducer Design and Identification for Internally-Paced Motor Brain Computer Interfaces: A Review

“…End-to-end learning exhibits several advantages; for example, it can be implemented with minimal preprocessing procedures [e.g., centering (Wang Z. et al, 2013 ; Nurse et al, 2015b ; Schirrmeister et al, 2017 ), scaling (Wang Z. et al, 2013 ; Schirrmeister et al, 2017 ), outlier removal (Nurse et al, 2016 ), or band pass filtering (Yuksel and Olmez, 2015 ; Sturm et al, 2016 ; Tang et al, 2017 )]. It additionally holds the promise of highly accurate decoding because of the joint optimization of feature extraction and decoding.…”

“…It additionally holds the promise of highly accurate decoding because of the joint optimization of feature extraction and decoding. While statistically significant performance improvements have been reported when comparing end-to-end models with approaches combining CSP-based feature extraction with generic classifiers (Yuksel and Olmez, 2015 ; Lu et al, 2017 ; Tang et al, 2017 ), end-to-end models have not yet clearly outperformed state of the art methods (Nurse et al, 2015b , 2016 ; Schirrmeister et al, 2017 ). Some of the difficulties that may impair the efficiency of end-to-end approaches include difficulties to fit end-to-end models, such as to gather enough data, and/or properly regularize the models so as to avoid overfitting.…”