EEG denoising with a recurrent quantum neural network for a brain-computer interface

Abstract

Brain-computer interface (BCI) technology is a means of communication that allows individuals with severe movement disability to communicate with external assistive devices using the electroencephalogram (EEG) or other brain signals. This paper presents an alternative neural information processing architecture using Schrödinger wave equation for enhancement of the raw EEG signal. The raw EEG signal obtained from the motor imagery (MI) of a BCI user is intrinsically embedded with non-Gaussian noise while the actual signal is still a mystery. The proposed recurrent quantum neural network (RQNN) is designed to filter such non-Gaussian noise using an unsupervised learning scheme without making any assumption about the signal type. The proposed learning architecture has been modified to do away with the Hebbian learning associated with the existing RQNN architecture [1] as this learning scheme was found to be unstable for complex signals such as EEG. Besides this the soliton behaviour of the non-linear Schrödinger wave equation was not properly preserved in the existing scheme. The unsupervised learning algorithm proposed in this paper is able to efficiently capture the statistical behaviour of the input signal while making the algorithm robust to parametric sensitivity. This denoised EEG signal is then fed as an input to the feature extractor to obtain the Hjorth features. These features are then used to train a Linear Discriminant Analysis (LDA) classifier. It is shown that the accuracy of the classifier output over the training and the evaluation datasets using the filtered EEG is much higher compared to that using the raw EEG signal. The improvement in classification accuracy computed over nine subjects is found to be statistically significant.

High performance EEG signal classification using classifiability and the Twin SVM

“…We compare the results with those presented in Table 3 of [63], and observe that on the average, the FBCSP approach obtains a Ä value of 0.493, whereas our approach attains a Ä value of 0.526, which is a 6.6% improvement on the average across sessions and subjects. Also, when compared with the techniques that cite best case Ä values, our approach outperforms BS [35] for 6 out of 9 subjects, and RQNN [38] and DDFBS [36] for 7 out of 9 subjects across sessions. In terms of average accuracy across subjects and sessions, we obtain an improvement of 21.9%, 13.16% and 43.22% over BS, RQNN and DDFBS respectively.…”

“…A bi-spectrum (BS) approach for feature extraction has been presented by Shahid et al [35], whereas Suk and Lee propose a Data-Driven Frequency Band Selection (DDFBS) approach [36]. A review of EEG processing paradigms has been presented by Majumdar in [37], whereas the use of a modern classification technique, the Recurrent Quantum Neural Network (RQNN) has been presented by Gandhi et al [38]. We compare our work with these approaches in Sections 4.2 and 4.3.…”

“…The key novelty of our approach is in the use of a method which is computationally simple and cost-effective, as well as provides significant improvement in terms of performance, specifically the classification accuracy. Approaches proposed in literature [33][34][35][36]38] use feature selection methods that involve using the classifier to determine the robustness of the feature set, which is computationally intensive. Our approach entails selecting the band based on classifiability of the corresponding feature set.…”

“…We choose to compare our results with those in [39] as they provide results using similar metrics on the same datasets. Table 2, as well as with other approaches based on BS [35], RQNN [38] and DDFBS [36]. The paper on FBCSP compares the average Ä values across subjects and sessions, whereas the papers on BS, RQNN and DDFBS compare the best case Ä values.…”

“…Table 2 Results for BCI Competition IV, Dataset 2b comparing our approach against state-of-art approaches, viz. FBCSP [33], Bi-spectrum [35], RQNN [38], DDFBS [36]. The datasets have 60-80 trials and results have been obtained using ten fold cross validation, as indicated in the column "Our Approach".…”