Deep Learning Approach: Integrated Contextual Gate Networks for fNIRS-BCI

Brain Computer Interface (BCI) systems involve five stages: brain signal acquisition, signal preprocessing to remove artifacts, feature extraction, signal classification, and command generation for BCI application. Deep learning (DL) algorithms presents an important role in improving accuracy of functional near-infrared spectroscopy-BCI (fNIRS-BCI) systems. Like conventional machine learning (ML) classifiers, DL algorithms eliminate the necessity for manual feature extraction, as DL algorithms automatically recognize features and hidden patterns in the dataset for classification. In this study, we acquired a two-class motor activity dataset (Hand closing and opening) from twenty healthy participants. After preprocessing we applied z-score normalization on dataset to standardize the dataset and then DL integrated contextual gate network (ICGN) algorithm is used to enhance the classification accuracy of motor activity. The DL-ICGN network is designed to extract features from the preprocessed data to generate patterns based on the previously hidden cells information within the neural network, consequently classifying data based on these features and generated patterns. The DL-ICGN classification performance is compared with that of long short-term memory (LSTM) and bidirectional long short-term memory (Bi-LSTM) networks. The DL-ICGN algorithm attained a classification accuracy of 97.21 ± 0.73, significantly (p<0.025) outperforming the 89.72±2.85<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> and 94.02 ±1.73% accuracies of LSTM and Bi-LSTM, respectively. The results show that z-score normalization along with DL-ICGN classification algorithm can be successfully used for classifying both two-class and three-class classification problems in fNIRS-BCI systems applications.