Reconstructing three-dimensional hand movements from noninvasive electroencephalographic signals

Abstract

It is generally thought that the signal-to-noise ratio, the bandwidth, and the information content of neural data acquired via noninvasive scalp electroencephalography (EEG) are insufficient to extract detailed information about natural, multijoint movements of the upper limb. Here, we challenge this assumption by continuously decoding three-dimensional (3D) hand velocity from neural data acquired from the scalp with 55-channel EEG during a 3D center-out reaching task. To preserve ecological validity, five subjects self-initiated reaches and self-selected targets. Eye movements were controlled so they would not confound the interpretation of the results. With only 34 sensors, the correlation between measured and reconstructed velocity profiles compared reasonably well to that reported by studies that decoded hand kinematics from neural activity acquired intracranially. We subsequently examined the individual contributions of EEG sensors to decoding to find substantial involvement of scalp areas over the sensorimotor cortex contralateral to the reaching hand. Using standardized low-resolution brain electromagnetic tomography (sLORETA), we identified distributed current density sources related to hand velocity in the contralateral precentral gyrus, postcentral gyrus, and inferior parietal lobule. Furthermore, we discovered that movement variability negatively correlated with decoding accuracy, a finding to consider during the development of brain-computer interface systems. Overall, the ability to continuously decode 3D hand velocity from EEG during natural, center-out reaching holds promise for the furtherance of noninvasive neuromotor prostheses for movement-impaired individuals.

Figure 1.

Experimental setup and finger paths. The reaching apparatus is shown in the middle along with the Cartesian coordinate system we used. The distance from the center position to each of the targets was ∼22 cm. Mean finger paths for center-to-target (black) and target-to-center (gray) movements exhibited movement variability among subjects.

Figure 2.

EEG decoding accuracy of hand velocity. A, The mean (black) ± SEM (gray) of the r values across subjects (n = 5) versus the number of sensors exhibited a peak at 34 sensors. B, With 34 sensors, we computed the mean ± SEM of the r values across cross-validation folds (n = 8) for each subject for x (black), y (gray), and z (white) velocities. C, Reconstructed (black) and measured (gray) velocity profiles demonstrated similarities. Exemplar velocity profiles from the subjects with the best (subject 1, top row) and worst (subject 5, bottom row) decoding accuracies are shown.

Figure 3.

Scalp and current sources that encoded hand velocity. A, Mean (n = 5) scalp maps of the best 34 sensors revealed a network of frontal, central, and parietal involvement along with a large individual contribution from sensor CP3. Light and dark colors represent high and low contributors, respectively. Each scalp map with its percentage contribution is displayed above its associated 10 ms time lag, revealing the 16.0% maximal contribution of EEG data at 60 ms in the past. B, We overlaid localized sources (yellow) from 60 ms in the past onto MRI structural images to reveal the involvement of the precentral gyrus (x = −30, y = −30, z = 52), postcentral gyrus (x = −35, y = −30, z = 47), and IPL (x = −35, y = −36, z = 42).

Figure 4.

Relationship between movement variability and decoding accuracy. A, The CVs for MT (black) and ML (white) ranged across subjects. B, The kurtosis of the velocity profiles also varied across subjects. C, All movement variability measures demonstrated high negative correlations with the decoding accuracy shown in Figure 2B. Rectangles demarcate the confidence intervals for the bootstrapped r values, with each rectangle possessing a horizontal line at the median. The confidence intervals are 70, 90, and 70%, respectively, for MT, ML, and kurtosis.