Dorsolateral prefrontal cortex-based control with an implanted brain-computer interface

Abstract

The objective of this study was to test the feasibility of using the dorsolateral prefrontal cortex as a signal source for brain-computer interface control in people with severe motor impairment. We implanted two individuals with locked-in syndrome with a chronic brain-computer interface designed to restore independent communication. The implanted system (Utrecht NeuroProsthesis) included electrode strips placed subdurally over the dorsolateral prefrontal cortex. In both participants, counting backwards activated the dorsolateral prefrontal cortex consistently over the course of 47 and 22 months, respectively. Moreover, both participants were able to use this signal to control a cursor in one dimension, with average accuracy scores of 78 ± 9% (standard deviation) and 71 ± 11% (chance level: 50%), respectively. Brain-computer interface control based on dorsolateral prefrontal cortex activity is feasible in people with locked-in syndrome and may become of relevance for those unable to use sensorimotor signals for control.

Trial registration: ClinicalTrials.gov NCT02224469.

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1

fMRI results and electrode strip location. T-maps from the mental calculation task (red), count task (orange), and their functional overlap (white) from both participants (left: UNP1, right: UNP4) plotted on their anatomical T1 weighted images (method from21, supplementary materials). Applied thresholds: UNP1: t > 10 for both tasks; UNP4: t > 7 and t > 3 for mental calculation and count task, respectively; voxel depth 8 mm. Strip and electrodes placed over the left-dlPFC are indicated in black on the brain surface. Electrode numbers are indicated at the bottom of the figure, e8 being the most frontal and inferior electrode in both participants. Electrodes used for control in both participants are indicated in light blue. In our visualization software, different colors are represented numerically (1–3 for mental calculation task, count task, and the overlap, respectively) and edge colors take one value lower: hence the different edge colors for different activation maps in these plots.

Figure 2

Representative examples of spectral power changes during the count task. Top graphs show a spectrogram of both conditions for a representative run from each participant (red is active, black is rest). Bottom graphs show respective signed r2 values for each 1 Hz bin, with blue circles indicating significant r2 values (α = 0.05; p values not corrected). Plotted data is from week 9 (UNP1) and week 22 (UNP4). Both runs were recorded with pair e8–e10.

Figure 3

Signed r2 values from all bipolar electrode combinations. r2 values between HFB power (65–95 Hz) and task conditions, from all bipolar pairs, with weeks since implantation on the x-axis. Different lines indicate different bipolar pairs (see legend), and asterisks indicate significant HFB activation during active blocks (α = 0.05; p values not corrected).

Figure 4

Signed r2 values from control pair e8–e10. For UNP1 (top) and UNP4 (bottom), the signed r2 values between HFB power (65–95 Hz) and count task conditions from all runs with pair e8–e10 are plotted, with weeks since implantation on the x-axis. Y-axis runs from 0 to 1, because no r2 value from pair e8–e10 fell below 0. Black dots indicate significant HFB activation during active blocks (α = 0.05; p values not corrected). Because task duration was varied and p values are influenced by sample size (i.e. the number of blocks), the statistical significance of specific r2 values may vary (e.g. UNP1′s 0.44 in week 23 is not significant, even though UNP4′s 0.39 in week 21 is). Runs from UNP4 recorded during high fatigue are indicated by squares and were removed from final linear regression. Values from electrode pair e8–e10 from Fig. 3 are included in this figure.

Figure 5

Target task performance. Performance on the target task for both participants, with weeks since implantation on the x-axis, and score (0–100%) on the y axis. Chance level is 50%. Fill indicates strategy (black means with cues; white without). Because runs with the same score from one session are plotted on the same location, the number of dots in these figures does not match number of reported runs.