Abstract:
Major contributors to impairment are neurological conditions such epilepsy, Alzheimer’s
disease, multiple sclerosis, Parkinson’s disease, and others. These neurodegenerative
conditions are life-altering, and those afflicted with them must fight a new battle every day.
This manuscript presents the implementation of a Thought-to-Text Conversion BCI system.
The system employs a joint CNN-LSTM deep neural network to learn high-level combined
spatial and temporal features from MI-EEG signals. To eliminate various artifacts from EEG
signals, pre-processing is employed, such as utilizing a band-pass filter. Although an Auto encoder layer is included, it has been found to decrease efficiency. The XGBoost Classifier is
utilized to classify EEG signals into five EEG imagery tasks. This Deep Learning model is then
pickled using Python and deployed to the server. The server prompts the development of a GUI
that maps the input EEG signals to corresponding alphabets. The Neural Network is trained on
multiple open-sourced EEG datasets, utilizing various optimization algorithms to achieve high
system accuracy and desirable efficiency for real-time applications. The textual output obtained
can be utilized to aid in the rehabilitation of paralysis and stroke patients. The proposed hybrid
model achieved an accuracy of 96.89% on testing data. While taking approximately 6 seconds
per letter to be typed on the GUI screen.
