Table 3.
Important active hybrid brain–computer interface studies with applications to increased accuracy and number of commands for brain–computer interface studies (BCI) (from 2010 to 2016).
| Reference | Brain area | Activity | Modality | Application | Analysis type | Classifier | Commands | Accuracy | Window size |
|---|---|---|---|---|---|---|---|---|---|
| Li et al. (2010) | Whole brain | Motor imagery (MI) and P300 | Electroencephalography (EEG) + electrooculography (EOG) | Cursor control in 2D | Online | Support vector machine (SVM) | 4 | 92.8% | 0–600 ms after button flashes on the screen for 8 s |
| Allison et al. (2010) | Motor and occipital regions | MI and steady-state visual evoked potential (SSVEP) | EEG | Option selection from the screen | Offline | Linear discriminant analysis (LDA) | 4 | 74.8% for MI, 76.9% for SSVEP, and 81% for hybrid | 3–5 s window |
| Zhang et al. (2010) | Motor, parietal, and occipital regions | Mental task | EEG + EOG + electromyography (EMG) | Application to devices control | Offline | Fisher discriminant analysis combined with Mahalanobis distance | 4 | 75.3% average for two-class and 54.1% for four-class | 0–1 s |
| Su et al. (2011) | Whole brain | MI and P300 | EEG | Virtual environment control | Online | SVM and fisher LDA | 5 | 84.5% for MI and 81.7% for P300 | 0–2 s for MI and 0.7 s for P300 |
| Leeb et al. (2011) | Motor cortex | Motor execution | EEG + EMG | Application to patient motor training | Online | Bayesian | 2 | 87% for individual and 91% for hybrid case | 0.5 s for EEG and 0.3 s for EMG |
| Long et al. (2012a) | Frontal, central, parietal, and occipital regions | P300 and MI | EEG | Direction and speed control for wheelchair | Online | LDA | 5 | 75.4% for hybrid task | 1 s |
| Yong et al. (2012) | Motor cortex | Hand and eye movement | EEG + EOG (eye tracker) | Artifact removal for choice selection | Online | SW-LDA | 2 | True positive rate increases from 44.7 to 73.1% (in 1 s) | 1 s |
| Fazli et al. (2012) | Frontal, motor, and parietal cortex | MI and Motor execution | EEG + functional near infrared spectroscopy (fNIRS) | Application to control | Offline | LDA | 2 | 93.2% (motor execution) and 83.2% (MI) | 0.75 s for EEG, 6 s prior to stimulus onset and up to 15 s after stimulus onset using 1 s sliding window for fNIRS |
| Choi and Jo (2013) | Whole brain | SSVEP, MI, and P300 | EEG | Humanoid robot navigation and recognition | Real time | CCA | 6 | 84.6% for P300 and 84.04% for SSVEP | 2 s |
| Cao et al. (2014) | Frontal, central, parietal and occipital cortex | SSVEP and MI | EEG | Brain-actuated switch for wheelchair control | Online | SVM | 8 | 90.6% | – |
| Wang et al. (2014) | Whole brain | MI, P300 and eye blinking | EEG + EOG | Asynchronous wheelchair control | Online | SVM | 7 | 91, 93, 89, and 92% for forward, backward, stop with special threshold, and stop with optimal threshold, respectively | 4 s |
| Khan et al. (2014) | Prefrontal and motor cortex | Mental arithmetic, mental counting and motor execution | EEG + fNIRS | Application to wheelchair control | Online | LDA | 4 | 94.7% for left and right movement commands (EEG) and 80.2 and 83.6% for forward and backward using fNIRS | 0–10 s for fNIRS and 0–1 s for EEG |
| Kim et al. (2014) | Complete brain | Eye movement | EEG + Eye tracker | Quadcopter control | Real time | SVM | 8 | 91.67% | 5 s |
| Jiang et al. (2014) | Motor cortex | MI and eye movement | EEG + EOG | Application to BCI control | Online | LDA | 4 | 90.4% for MI, 91.1% for relax, 96.4% for gaze left, and 97.3% for gaze right | 3 s |
| Kaiser et al. (2014) | Motor cortex | MI | EEG + fNIRS | Application to brain monitoring | Online | LDA | 1 | 3.6% increase in accuracy by hybrid modality | 3–7 s |
| Lorenz et al. (2014) | Whole brain | ERP and MI | EEG | BCI driven neuro-prosthesis | Online | LDA | 6 | Maximum selection accuracy of 98.46% and maximum confirmation accuracy of 96.26% | 1 s |
| Blokland et al. (2014) | Motor cortex | MI and motor execution | EEG + fNIRS | Application to tetraplegia patients | Offline | – | 2 | 87% for motor attempt and 79% for MI in tetraplegia patients | 3–15 s for fNIRS and 0–15 s for EEG |
| Bai et al. (2015) | Whole brain | MI and P300 | EEG | Opening, closing, selection of files on explorer | Online | SVM | 9 (can achieve 50) | >90% | 4 s window for MI and 600 m for P300 |
| Hortal et al. (2015) | Motor and parietal cortex | Mental imagination | EEG + EOG | Robotic arm control for pick and place task | Real time | SVM | 6 | Task 1: 71.13% and Task 2: 61.51% | 0.5 s to synchronize output to BMI |
| Hong et al. (2015) | Prefrontal and motor cortex | Mental arithmetic and MI | fNIRS | Applications to three choice selection | Offline | LDA | 3 | 75.6% | 2–7 s |
| Naseer and Hong (2015a) | Prefrontal and motor cortex | Mental arithmetic, mental counting and MI | fNIRS | Decoding answers to four-choice questions | Offline | LDA | 4 | RMI, LMI, MA, and MC were correctly classified as 72.9, 64.2, 65.1, and 71.0%, respectively | 2–7 s |
| Yin et al. (2015c) | Motor cortex | MI task | EEG + fNIRS | Increase in accuracy for BCI | Online | ELM | 2 | 88% | 0.5 s for EEG and 0–12 s for fNIRS |
| Koo et al. (2015) | Motor cortex | Self-paced MI | EEG + fNIRS | Application to device control | Online | SVM | 2 | 88% average accuracy | 10 s for fNIRS and three 5 s time windows with step size of 2.5 s for EEG |
| Buccino et al. (2016) | Motor cortex | Arm and hand movement | EEG + fNIRS | Hand movement discrimination | Online | LDA | 2 commands simultaneously | 94.2% (for rest-task classification) | 0~6 s hybrid |
| Shishkin et al. (2016) | Whole brain | Eye gaze | EEG + EOG | Game control | Offline | LDA | – | 90% | 0.3 s for EEG and 0.2–0.5 s for EOG |
| Khan and Hong (2017) | Frontal | Mental task and eye movement | NIRS + EEG | Applications to quadcopter control | Online | LDA | 8 | 76.5% for NIRS and 86% for EEG | 1 s for EEG and 2 s for NIRS |