Table 1.
State-of-the art summary of hand-motor decoding experiments involving high-gamma based invasive BCIs.
| Reference | No. of subjects | Electrode spacinga | Frequency bands (Hz) | Feature reductionb | Classifierc | Trial alignment | Trial length (s)d | Protocole | Classesf | Mean accuracy (%) |
|---|---|---|---|---|---|---|---|---|---|---|
| Shenoy et al., 2007 | 6 | Macro | 11 − 40 71 − 100 101 − 150 |
None | LPM | None | Not reported | Finger | 5 | 77.0 |
| Kubánek et al., 2009 | 5 | Macro | 8 − 12 18 − 24 75 − 115 125 − 159 159 − 175 |
None | LMD | Data glove | [−1.0, +1.0] | Finger | 5 | 80.3 |
| Onaran et al., 2011 | 3 | Macro | 65 − 200 |
CSP | SVM | Data glove | [−0.75,+1.0] | Finger | 5 | 86.3 |
| Yanagisawa et al., 2011 | 1 | Macro | 1 − 8 25 − 40 80 − 150 |
None | SVM | None | n/a (online) | Move RPS |
1 + 1 3 |
79.6 68.3 |
| Pistohl et al., 2012 | 3 | Macro | 2 − 6 14 − 46 54 − 114 |
None | rLDA | Data glove | [−1.0,+0.5] | Grasp | 2 | 87.8 |
| Chestek et al., 2013 | 3 | Mixed | 66 − 114 | None | NB | Data glove | [−0.5, +1.5] | Gesture Finger |
4 + 1 4 + 1 |
77.7 79.3 |
| Kapeller et al., 2014 | 2 | Macro | 60 − 90 110 − 140 160 − 190 |
FS CSP |
LDA | None | [−0.5,+1.5] | RPS | 3 | 83.8 95.8 |
| Xie et al., 2015 | 4 | Macro | Auto | FS | LDA | None | Various | Gesture | 2 3 |
95.5 92.7 |
| Bleichner et al., 2016 | 2 | Micro | 70 − 125 | FS | PM | Data glove | [−1.0,+2.0] | Gesture | 4 | 85.5 |
| Hotson et al., 2016 | 1 | Micro | 72 − 110 | FS | LDA | Data glove | [−0.4, +1.0] | Finger | 5 | 96.5 |
| Branco et al., 2017 | 5 | Micro | 70 − 125 | None | PM | High-gamma | [−1.0,+2.6] | Gesture | 4 | 85.0 |
| Jiang et al., 2017 | 2 | Micro | 60 − 200 | CSP | LDA | Not reported | [−0.15, +0.35] | Gesture | 2 | 100.0 |
| Li et al., 2017 | 3 | Macro | 4 − 12 70 − 135 |
FS | SVM | None | [±0.0, +0.9] | RPS | 3 | ≈80 |
| Pan et al., 2018 | 5 | Micro | 4 − 12 12 − 40 40 − 70 70 − 135 135 − 200 |
FS | RNN | Data glove | [±0.0, +0.5] [±0.0, +1.2] |
RPS | 3 | ≈80 ≈90 |
a
Macro, standard ECoG grid; Micro, high-density ECoG grid; Mixed, standard and high-density ECoG grids.
b
CSP, common spatial patterns; FS, algorithm-based or manual channel/feature selection.
c
LPM, linear programming machine; LMD, linear multivariate decoder; SVM, support vector machine; (r)LDA, (regularized) LDA; NB, naive Bayes; PM, pattern matching; RNN, recurrent neural network.
d
Specified relative to cue, movement onset, or high-gamma onset (depending on trial alignment).
e
Finger, finger movement or tapping; Move, movement vs. rest; RPS, rock-paper-scissors; Gesture, arbitrary hand gestures.
f
Inclusion of a resting-state class denoted by “+1”.