Decoding of Ankle Flexion and Extension from Cortical Current Sources Estimated from Non-invasive Brain Activity Recording Methods

Abstract

The classification of ankle movements from non-invasive brain recordings can be applied to a brain-computer interface (BCI) to control exoskeletons, prosthesis, and functional electrical stimulators for the benefit of patients with walking impairments. In this research, ankle flexion and extension tasks at two force levels in both legs, were classified from cortical current sources estimated by a hierarchical variational Bayesian method, using electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) recordings. The hierarchical prior for the current source estimation from EEG was obtained from activated brain areas and their intensities from an fMRI group (second-level) analysis. The fMRI group analysis was performed on regions of interest defined over the primary motor cortex, the supplementary motor area, and the somatosensory area, which are well-known to contribute to movement control. A sparse logistic regression method was applied for a nine-class classification (eight active tasks and a resting control task) obtaining a mean accuracy of 65.64% for time series of current sources, estimated from the EEG and the fMRI signals using a variational Bayesian method, and a mean accuracy of 22.19% for the classification of the pre-processed of EEG sensor signals, with a chance level of 11.11%. The higher classification accuracy of current sources, when compared to EEG classification accuracy, was attributed to the high number of sources and the different signal patterns obtained in the same vertex for different motor tasks. Since the inverse filter estimation for current sources can be done offline with the present method, the present method is applicable to real-time BCIs. Finally, due to the highly enhanced spatial distribution of current sources over the brain cortex, this method has the potential to identify activation patterns to design BCIs for the control of an affected limb in patients with stroke, or BCIs from motor imagery in patients with spinal cord injury.

Keywords: brain computer interface; electroencephalography; functional magnetic resonance imaging; walking.

Figure 1

Images with motor task instructions used during the fMRI and the EEG experiments. Blink and fixation crosses were only used for the EEG experiment.

Figure 2

(A) Experimental paradigm for the fMRI experiment. The fMRI experiment consisted 7 runs. In each run, 9 blocks were included (8 active tasks and a control task), and each block consisted of one experimental task and a rest interval, both repeated 6 times. (B) Experimental setting for the fMRI experiment. EMG electrodes were attached to the participants to confirm task execution and the feet of the participant were fixed to a custom made-platform to reduce head movements inside the scanner. (C) Experimental paradigm for the EEG experiment. The EEG experiment consisted of three modules: “Flexion vs. Extension,” “Right vs. Left,” and “High Force vs. Low Force.” In each module, 2 active tasks were repeated 10 times (10 trials) and a still task was repeated 5 times (5 trials). The 2 active tasks were selected based on the module (i.e., HLE and HLF for the “Flexion vs. Extension” module). After 25 trials (10 trials for each of the 2 active tasks and 5 trials for the still task), the active tasks were changed until completing all the experimental tasks. (D) Experimental setting for the EEG experiment. A cap with 32 EEG electrodes and 8 EMG electrodes were attached to the participant. A custom-made platform was also used in this experiment to attach the participant's feet during the EEG experiment in order to reduce movement artifacts and allow for isometric contractions.

Figure 3

Classification accuracies across participants for current sources estimated using priors from and fMRI group analysis (Group-Con), pre-processed EEG signals (EEG), and a random label classification for current sources. ^***p < 0.001.

Figure 4

(A) Normalized weights obtained for each task in EEG classification. Bars located on the green area correspond to the sensors located in the left hemisphere, bars in the white area correspond to the midline, and electrodes in the pink area correspond to the right hemisphere. (B) Location of 32 EEG electrodes over the scalp using the 10–20 extended system.

Figure 5

Activation patterns for a representative participant. Colored areas show Brodmann areas 1, 2, 3, 4, 6 OP3, OP4, and the IPC. Lower left panel shows the merged activation patterns for all left and right leg tasks (all right leg tasks: HRE + HRF + LRE + LRF; all left leg tasks: HLE + HLF + LLE + LLF), and lower right panel shows the activation patterns for all extension tasks, all flexion tasks, all high force tasks and all low force tasks (all flexion tasks: HRF + HLF + LRF + LLF; all extension tasks: HRE + HLE + LRE + LLE; all high force tasks: HRF + HLF + HRE + HLE; all high force tasks: LRF + LLF + LRE + LLE). Time series represent the temporal patterns for the current source vertex in bold black circles. These signals correspond to a vertex selected by the classifier as relevant for more than one task. The vertex located in MNI [12, −36, 68], was selected by the classifier for both HLE and HLF task, and the vertex located in MNI [8, −36, 68] was selected for both HLF and LLF tasks.