Quantifying motor function using eeg signals
A motor control score is automatically generated for a subject based on electroencephalography (EEG) data represented as a plurality of EEG waveforms obtained from a plurality of EEG electrodes. Each waveform is separated into at least one window, containing a portion of the EEG waveform representative of neural activity corresponding to a movement performed by the subject. At least one or more frequency components and/or signal components are determined from the EEG data in the window, and a motor control score is determined based on these components. The frequency components may correspond to event-related desynchronization (ERD) response frequency band and event-related synchronization (ERS) response frequency band. In other embodiments, the frequency components correspond to a phase response and the signal components correspond to a signal power spectrum density of the window, which are provided to a primary neural network model to determine the motor control score.
1. A computer-implemented method to automatically generate a motor control score for a subject based on electroencephalography (EEG) data represented as a plurality of EEG waveforms obtained from a plurality of EEG electrodes, the method comprising:
(a) for each EEG waveform of the plurality of EEG waveforms, separating the EEG waveform into at least one window, wherein each window contains a portion of the EEG waveform representative of neural activity corresponding to a movement performed by the subject;
(b) determining, from the EEG data in the at least one window, at least one or more frequency components; and
(c) calculating the motor control score based on the at least one or more frequency components, wherein
the one or more frequency components correspond to at least one phase response of the at least one window;
at least one or more signal components are determined from the EEG data in the at least one window, the one or more signal components corresponding to at least one signal power spectrum density of the at least one window; and
the motor control score is calculated using a primary neural network, the primary neural network being provided with the EEG data in the at least one window as input comprising the at least one phase response and the at least one signal power spectrum density.
2. The computer-implemented method of claim 1 , wherein the plurality of EEG waveforms are filtered using a 1 Hz to 45 Hz bandpass finite impulse response filter prior to the step of separating the EEG waveforms into the at least one window.
3. The computer-implemented method of claim 1 , wherein the phase response of the EEG data in the at least one window comprises elements within at least one of an alpha band and a beta band.
4. The computer-implemented method of claim 1 , wherein fast Fourier transform is performed on the at least one window to obtain the at least one phase response and the at least one signal power spectrum density.
5. The computer-implemented method of claim 1 , wherein the primary neural network is configured using regression with a training data set having training phase responses and training power spectrum densities in a plurality of training EEG waveforms, and a plurality of manually-assessed clinical motor function scores associated with the plurality of training EEG waveforms.
6. The computer-implemented method of claim 1 , wherein the at least one phase response and the at least one signal power spectrum density is calculated from a corresponding frequency domain representation of the at least one window.
7. The computer-implemented method of claim 1 , wherein the primary neural network model is configured using a configuration framework comprising at least two two-dimensional convolutional neural network layers, at least two max pooling layers, and at least two fully connected layers.
8. The computer-implemented method of claim 1 , wherein the primary neural network model is configured using a configuration framework comprising a two-dimensional convolutional neural network layer and a fully connected layer.