Decoding four hand gestures with a single bipolar pair of electrocorticography electrodes

Abstract

Objective.Electrocorticography (ECoG) based brain-computer interfaces (BCIs) can be used to restore communication in individuals with locked-in syndrome. In motor-based BCIs, the number of degrees-of-freedom, and thus the speed of the BCI, directly depends on the number of classes that can be discriminated from the neural activity in the sensorimotor cortex. When considering minimally invasive BCI implants, the size of the subdural ECoG implant must be minimized without compromising the number of degrees-of-freedom.Approach.Here we investigated if four hand gestures could be decoded using a single ECoG strip of four consecutive electrodes spaced 1 cm apart and compared the performance between a unipolar and a bipolar montage. For that we collected data of seven individuals with intractable epilepsy implanted with ECoG grids, covering the hand region of the sensorimotor cortex. Based on the implanted grids, we generated virtual ECoG strips and compared the decoding accuracy between (a) a single unipolar electrode (Unipolar Electrode), (b) a combination of four unipolar electrodes (Unipolar Strip), (c) a single bipolar pair (Bipolar Pair) and (d) a combination of six bipolar pairs (Bipolar Strip).Main results.We show that four hand gestures can be equally well decoded using 'Unipolar Strips' (mean 67.4 ± 11.7%), 'Bipolar Strips' (mean 66.6 ± 12.1%) and 'Bipolar Pairs' (mean 67.6 ± 9.4%), while 'Unipolar Electrodes' (61.6 ± 5.9%) performed significantly worse compared to 'Unipolar Strips' and 'Bipolar Pairs'.Significance.We conclude that a single bipolar pair is a potential candidate for minimally invasive motor-based BCIs and encourage the use of ECoG as a robust and reliable BCI platform for multi-class movement decoding.

Keywords: bipolar; brain–computer interface; electrocorticography; minimally invasive; sign language; unipolar.

Figure 1 |

Electrode selection. Position of implanted electrodes of all individual participants, overlayed on the participant’s anatomical T1-weighted scan and the fMRI activity pattern of the course finger-tapping movement (blue areas; t > 5–16). White circles indicate electrodes of interest included in one or more of the virtual electrode strips, whereas black circles indicate other implanted electrodes that were included in the common-average re-reference associated with the ‘Unipolar’ selection schemes. Of note, P1 and P3-P7 also had interhemispheric electrodes (included in CAR) that are not visualized here.

Figure 2 |. Task description and channel selection scheme.

A) Example of the strip definition and channel selection method. B) A representative data glove trace of an active trial of the gesture ‘D’ and subsequent rest trial. C) The mean power amplitude (60–130 Hz; arbitrary units, a.u.) over trials per gesture of one representative channel of P4, centered on movement onset (t = 0s). The standard deviation across trials is represented by the shaded region with colors corresponding to the respective gesture (G1 = ‘D’; G2 = ‘F’; G3 = ‘V’; G4 = ‘Y’). D) Channel selection schemes. ‘Unipolar electrode’ consists of a single electrode within a strip. ‘Unipolar strip’ consists of four electrodes within a strip. ‘Bipolar Pair’ consists of a single bipolar pair within a strip. ‘Bipolar strip’ consists of the combination of six bipolar pairs within a strip. Example of a virtual strip is delineated by the black line. The white lines indicate the bipolar referencing, that is the subtraction between electrodes. Large black circles correspond to all the electrodes included in the four schemes. UP – unipolar; BP- Bipolar.

Figure 3 |

Classification scores. A) Histograms of the classification scores for P1. The frequency plots display the distribution of classification scores of all channels of each scheme. The highest scores (outlined in black) were selected for subsequent analyses. The dashed and solid lines indicate the individual’s chance level and significance (p = 0.05) level, respectively. B) The highest classification score per method for the first run of each participant. The dashed line indicates the overall chance level. Individual’s chance and significance levels can be found in Supplementary Table 2. Significant pairwise comparisons at * p< 0.05, ** p< 0.01. C) Confusion-matrices averaged across participant’s highest classification scores. A dark blue diagonal from the upper left to the lower right indicates perfect classification. UP - Unipolar; BP - Bipolar.

Figure 4 |

Optimal localization and orientation of ‘Bipolar Strips’ and ‘Bipolar Pairs’. A) Localization of ‘Bipolar Pairs’ (blue lines) and ‘Bipolar Strips’ (dark gray lines) with maximum classification scores (100^th percentile) on the individual participant’s brain, per hemisphere. The mean (± standard deviation) of these maximum classification scores across participants, per hemisphere, is given below the brains for both ‘Bipolar Pairs’ and ‘Bipolar Strips’ schemes. B) The 100^th percentile ‘Bipolar Pairs’ and ‘Bipolar Strips’ displayed on a MNI standard brain in order to compare the optimal spatial orientation across participants. Each colored line represents a participant. BP - bipolar; LH - left hemisphere; RH - right hemisphere.