Modeling and Classification of Voluntary and Imagery Movements through fNIR & EEG Signals for Brain-Computer Interface

Abstract

Neural activation measurement regarding voluntary and imagery movement is a crucial argument for brain-computer interface. Generally, movement-related hemodynamics is measured from the central lobe of the brain which may be erroneous for the paralyzed people due to their inactiveness of this brain area. To overcome this limitation, this thesis work proposes an approach to measure the movement-related hemodynamics from the prefrontal cortex. In the proposed research work, the changes of the oxidized hemoglobin (HbO2) and deoxidized hemoglobin (dHb) concentration regarding different voluntary and imagery movement stimuli are captured by functional near-infrared spectroscopy (fNIR) from several subjects. With necessary preprocessing, the fNIR signals are statistically analyzed by ANOVA and effect size method to localize the most significantly activated regions of the prefrontal cortex regarding the voluntary and imagery stimuli. The experimental results show that voluntary and imagery movements have a strong correlation with the prefrontal cortex. The temporal pattern of HbO2 and dHb signals regarding the most activated regions are modeled by polynomial regression. Consequently, the model activation patterns are used to classify the voluntary and imagery tasks based on the maximum similarity approach. In addition, conventional classification methods are used to classify the signals. In this work, we consider two, four, and six class fNIR data of movement-related tasks for classification. The classification accuracies of the proposed method are convincing and found almost similar to the conventional procedure. The outcomes of this proposed work suggest that the prefrontal hemodynamics can be used for the modeling and classification of the voluntary and imagery movement-related tasks which will be helpful for the brain-computer interface applications. The combination of fNIR and electroencephalography (EEG) signals has become the best choice of accurate brain-computer interface (BCI) because of their finer spatiotemporal resolution. The purpose of this work is to develop an effective BCI model to classify the brain signals (fNIR and EEG) regarding the voluntary and imagery movements. For achieving the high classification accuracy from the developed BCI system, Convolutional neural network (CNN) has been used to extract the features automatically from the multiple channel fNIR and EEG signals instead of the manual feature selection. In this work, eight different movement-related stimuli (four voluntary and four imagery movements of hands and feet) have been considered. The multiple channel fNIR and EEG signals are used to prepare functional neuroimages to train and test the performance of the proposed BCI system. In addition, the proposed procedure is applied to prepare neuroimages from the individual modality (fNIR and EEG) to train and test the performance of the CNN based BCI system. The results reveal that the combined-modality approach of fNIR and EEG provides improved classification accuracy than the individual one. From the results, we found that the proposed CNN-based BCI system of bimodal (fNIR+EEG) approach outperforms the unimodal (only fNIR) methods in terms of the classification accuracy. Therefore, the outcomes of the proposed research work will be very helpful to implement the finer BCI system, in future.