Mind reading! Decoding imagined speech from brain signals

Abstract

Background: Brain traumas, mental disorders, and vocal abuse can result in permanent or temporary speech impairment, significantly impacting the quality of life and may result in social isolation. Brain-Computer Interface (BCI) can provide support using brain signals in these scenarios. EEG signal-based BCI has attained significant attention in the last two decades because clinical research has yielded detailed knowledge of EEG signals, inexpensive EEG devices have emerged and new applications in both medical and social fields are developing. Objective: This project aims to identify the best brain rhythm and associated set of features that can be used to train machine learning-based models for efficiently predicting imagined words from the brain signals. Methods: The dataset used in this project is obtained from an online 2020 International BCI competition. The EEG signals are filtered using the bandpass filtering technique in the different cut-off frequency ranges to identify the best brain rhythm and features associated with imagined speech. Then channel wrapping, and channel ranking techniques are applied to find the best combinations. Next kernel-based principal component analysis technique is used to reduce high dimensional features space to low dimensional feature space. Finally, various machine learning models are trained and tested using the extracted features to evaluate the efficacy of the models. Result: Among the available rhythms in the dataset, the Gamma rhythm provided the best results for predicting the imagined speech from EEG brain signals. Among multiple machine learning models we tested, kNN was the best model to predict imagined words from EEG brain signals. kNN model had produced an average accuracy of 73% in a 10-fold cross-validation scheme that was significantly higher than the average accuracies reported in the existing literature.

Date of Award	2022
Original language	American English
Awarding Institution	HBKU College of Science and Engineering