Wavelet Analysis on Electroencephalographic Time Series to Identify Key Patterns Corresponding to Arm Movements for Brain-Computer Interface

摘要

Electroencephalography (EEG) is the recording of electric potentials produced by neuronal activity within the cerebral cortex by placing electrodes on specific locations of the scalp. In our research, subjects performed real and imaginary movements of their right arm as EEG signals were recorded with a 32-electrode cap at a sampling rate of 1 kHz. Our experiment consisted of twelve 30-second tests, each composed of 5 repetitions of two opposite movements. Indications on timing, speed, and type of movement were provided by a computer program during the experiment. The 12 tests included two base lines with open and closed eyes, four tests covering all degrees of freedom of the shoulder and elbow, and one for hand movement, as well as imagination of all five movements. We used the Continuous Wavelet Transform, using the complex Morlet mother wavelet, to identify how characteristic frequencies change over time. We then developed a method for generating binary classifier functions capable of distinguishing between two sets of movements based on the wavelet coefficients at a variety of frequencies over all 32 electrodes. We then used the binary classifiers to generate a master classifier function which determines the most probable movement at any given time. Our results showed a 56.4% ± 7.9% success rate for determining the correct movement out of 23 different possibilities. Furthermore, the classifier functions themselves provide information regarding the key frequencies and regions that distinguish one arm movement from another.