BCI Data Repository Generation and Processing

Abstract

The scope of this project encompasses the identification and resolution of the technical issues within the EEG equipment available at NCRA, the design of protocols for data acquisition across various brain regions, the EEG data acquisition and repository generation and the analysis of the acquired EEG data through preprocessing and machine learning algorithms. Initially, the training for use of EEG equipment was acquired and multiple sessions were conducted in CMH to learn the EEG data acquisition technique. Meanwhile, efforts were made to contact the equipment OEM to provide after-sales service so that issues can be resolved. The software was updated and features including the selection of up to 128 channels for data acquisition, option to check for the connectivity status for EEG cap and the option to check the connectivity status of each channel respectively were introduced. All issues within EEG software were resolved. The EEG caps were verified via shortcircuit test and there were no issues found within the connectivity of the electrodes. The ground zero test was conducted for verification of the master/amplifier box. The results showed that amplifier IC was malfunctioning and the data for channels Fp1, Fp2, C3 and C4 was erroneous. These channels also had an impact on the data of other channels as the data for other channels was accurate to some extent but erroneous too. These results were verified by the OEM R&D. Meanwhile, the data was collected according to the desired protocol within optimum environmental conditions and testing parameters and an EEG data repository was generated. After the data acquisition, the data was preprocessed and analyzed. For preprocessing, the data was filtered using bandpass filter of 0.5-30 Hz and notch filter of 50 Hz. After filtering the EEG data, the Fast Fourier Transform (FFT) was done to analyze the frequency domain response of the EEG data and statistical analysis was done to analyze the time domain response of the EEG data. After these steps, the filtered and preprocessed EEG data was analyzed using machine learning algorithms including Support Vector Machine (SVM), Linear Discriminant Analysis (LDA) and Random Forest (RF) algorithms and the results were found for unlabeled data. From this analysis, it was concluded that the channels Fp1, Fp2, C3 and C4 showed a different and erroneous behavior. However, the response for other channels was accurate to some extent but still erroneous due to the interference and impact of these channels. From all these observations, we can conclude that the data can be acquired using this equipment, but it would be erroneous for other channels due to the impact of data of these channels. It is proposed that in order to get accurate data and continue research using the equipment available in NCRA, the amplifier IC within amplifier box needs to be either repaired or replaced.