TABLE 1.
Artificial Intelligence Algorithms applied for detection and classification of P300 or SSVEP signals in BCI Applications for motor rehabilitation.
| First author, year | BCI signal | Application/actuator | Subjects |
# Electrodes | Visual stimulation pattern | Feature extraction method | Classifier | Performance |
Validation Method | ||
| Impaired | Healthy | Accuracy (%) | ITR (bpm) | ||||||||
| Stan et al., 2015 | P300 | Hand orthosis | None | 9 | 8 | Flashes: 75 ms Flash-time: 100 ms | NS | LDA | 100 | NS | NS |
| Kwak et al., 2017 | SSVEP | Lower limb exoskeleton | None | 7 | 8 | 5 LEDs flashing at 9, 11, 13, 15, 17 Hz with 50% DC | NR | CNN | Static: 99.28, Ambulatory: 94.93 | NS | 10-fold CV |
| Kwak et al., 2015 | SSVEP | Lower limb exoskeleton | None | 11 | 8 | 5 LEDs: 9, 11, 13, 15, 17 Hz with 50% DC | CCA | k-nearest neighbors | 91.3 | 32.9 | 5-fold CV |
| Delijorge et al., 2020 | P300 | Robotic hand orthosis | 8 ALS | 18 | 8 | 2–30 random flashes | CCA | RLDA | Offline: 78.7 (target), 85.7 (non-target). Online: 89.83 | 18.13 | 5-fold CV |
| Zhao et al., 2016 | SSVEP | FES, upper limb rehabilitation | None | 5 | 14 | Squares flashing at 12, 15, 20 Hz | Power spectrum | LDA | Offline: 79.37–85.13 Online: 54.32–87.5 | Offline: 27.54 | 10-fold CV |
| Tidoni et al., 2017 | SSVEP | VR, Propioceptive Stimulation | 3 SCI | 18 | 8 | 3 × 3 grid. flash-time: 133.33 ms dark-time: 83.34 ms | NS | LDA | 83.33 | 1.55 | NS |
| Yao et al., 2011 | SSVEP | FES, upper limb rehabilitation | None | 4 | 8 | White blocks of lights flickering at 6.82, 7.5, 8.33, 9.37, and 12.5 Hz | 5 flickering frequencies and their harmonic components | LDA | Online: 82.22 | Ns | Ns |
| Brunner et al., 2011 | Hybrid: SSVEP + P300 | Moving both hand or both feet | None | 12 | SSVEP: 2. MI: 3. | LEDs flickering at 8 Hz (top) and LED at 13 Hz (bottom) | logarithmic band power: SSVEP and ERD | LDA | ERD: 79.9 SSVEP: 98.1 Hybrid: 96.5 | ERD: 3.2. SSVEP (6.1) hybrid (6.3) | CV |
| Edlinger et al., 2011 | Hybrid: SSVEP + P300 | VR, control of virtual smart home environment | None | 3 | SSVEP: 8. parietal/occipital. P300: 8 frontal, central occipital, parietal | P300: rectangular matrix with characters or icons, flashed in a random order SSVEP: flickering lights (LEDs) or flickering symbols (5 -25 Hz) | SSVEP: minimum energy (ME) algorithm, P300: NA | P300: LDA, SSVEP: LDA | P300: 100 | NS | NS |
| Su et al., 2011 | Hybrid: P300 + MI | VR | None | 4 | P300: 14. MI: 22. | NS | P300: piecewise cubic spline interpolation+ Butterworth filter + average. MI: multiple band-pass filters | P300: SVM, MI: FLDA | Offline (MI): 92.5–100 | NS | NS |
| Sakurada et al., 2013 | SSVEP + P300 | Upper limb rehabilitation,. Occupational therapy | 3 (upper cervical SCI) | 12 | SSVEP: 3 | SSVEP: 3 LEDs flickering at 8 Hz (green and blue). P300: Flash matrix | power spectrum (FFT) + CCA | SVM | Healthy: 88.46. Patients: 81.19 | NS | NS |
| Choi et al., 2016 | SSVEP + MI | FES, hand-wrist rehabilitation. SSVEP to stop FES | None | 4 | MI: 3 central. SSVEP: 2 occipital. | SSVEP: LED flickering at 9 Hz | MI: ERD/ERS, SSVEP: averaged Pearson’s correlation (r-value) | MI: FLDA SSVEP: CCA | MI: 90.485 | NS | 10-fold CV |
| Yao et al., 2012 | SSVEP | FES, knee rehabilitation (movement training system) | None | 2 | 8 | a red horizontal bar, flickering light at 6.82, 8.33 and 12.5 Hz | Power spectrum | LDA | Online: 80.36–96.4 | NS | 10-fold CV |
| Duvinage et al., 2012 | P300 | Lower limb rehabilitation. Foot lifting orthosis | None | 5 | 32 | NS | xDAWN + two epochs average | LDA | 94.30 | NS | NS |
| Ortner et al., 2011 | SSVEP | Hand Orthosis | None | 7 | 1: O1 | 2 LEDS, flickering at 8 and 13 Hz | PSD | HSD | 78 | NS | NS |
| Rohani et al., 2014 | P300 | VR | None | 5 | 4 | NS | NS | SVM | NS | NS | NS |
| Son et al., 2020 | SSVEP | FES, upper limb rehabilitation | None | 11 | 19 | flickering action video at 15 Hz | STFT, Power average | CSP (discriminating 2 class) | 93.51 | NS | 10-fold CV |
| Chen et al., 2019 | High-frequency SSVEP | Robotic arm | None | 10 | 9: parietal or occipital | Flicker: 30, 31, 32, and 33 Hz | Spectral amplitude | FBCCA | Online: 97.75 | Online: 17 | NS |
| Li et al., 2018 | SSVEP | Hand prosthesis | None | 6 | 2: occipital | Scene graph paradigm -drinking & eating-, (8, 9.24, 10.9, and 12 Hz) | Time-frequency spectra, STFT | CCA | 94.58 | 19.55 | NS |
| Horki et al., 2011 | SSVEP + MI | Prosthesis: artificial upper limb, elbow control | None | 12 | 26: occipital and central | 2 bars of red LEDs, flickering at 8 and 13 Hz | Sequential floating forward selection | CCA | Offline: 91 | NS | 10-fold CV |
| Koo et al., 2015 | SSVEP | VR | None | 3 | 8: central, parietal and occipital | Flickering lights at 5.5, 6.7, 7.5, and 8.6 Hz | NS | CCA for SSVEP detection | 100 | 24.58 | NS |
| Chu et al., 2018 | SSVEP | Robotic rehabilitation system | None | 6 | 14: frontal, parietal, occipital | Three squares flashing at 12, 15, 20 Hz | Power spectrum | LDA (voting) | 82.30 | 27.40 | NS |
| Gui et al., 2015 | SSVEP | Lower limb rehabilitation system (hip and knee) | None | 6 | 4: occipital and parietal | Flickering at 6.82, 7.5, 8.33, and 12.5 Hz | Spectral amplitude | LDA | 92.40 | NS | NS |
| Huang et al., 2019 | P300 | VR | None | 6 | 32 | 3D stereo visual stimuli | NS | BLDA | 96 | 42.51 | 10-fold CV |
| Yao et al., 2019 | SSVEP | VR | None | 10 | 9: parietal and occipital | 2 stimulus presentation methods. 3D stimulus at 9, 10, 11, 12, 45 Hz | NS | FBCCA | Static mode: 92 | Static mode: 22.49 | Leave one-out CV |
| Touyama and Sakuda, 2017 | Collaborative SSVEP | VR | None | 8 | 2: parieto-occipital | two virtual cubes flickering at 6 and 8 Hz | Spectral amplitude | FLDA | 95.2 | NS | NS |
| Bhattacharyya et al., 2014 | P300 | Robot arm control for prosthetics application | None | 5 | 1: Pz | Oddball-like paradigm | (Temporal) Average of 4 epochs | SVM (linear kernel) | Offline: 95.2. Online: 81.5 | Online: 23.83 | NS |
| Chen et al., 2018 | SSVEP | Robotic arm control | None | 12 | 10: P3, Pz, P4, PO3, PO4, T5, T6, O1, Oz, O2 | 15 targets (8–15 Hz in 0.5 Hz steps) | FBCCA for EEG decomposition | Ensemble Classifier | Robotic movement task: 92.78 | 49.25 | NS |
| Casey et al., 2019 | P300 | Robotic arm control | None | 4 | 6: Pz, P3, P4, PO3, PO4, and Oz | P300 speller programmed to control a robotic arm | Minimum and maximum amplitudes in the frequency domain (6 features per electrode) | 2 classifiers: SVM (RBF kernel), and Random Forest | 38.023 | NS | NS |
| Achanccaray et al., 2019 | P300 | Robotic arm Control | None | 8 | 16 | Two images flashing randomly: a wheelchair and a robotic arm | CSP | BLDA | Training: 91.6. Test: 82.6. | NS | NS |
| Ding-Guo and Ying, 2012 | SSVEP | FES, lower limb | None | 6 | NS | NS | Frequency-domain | LDA | 85 | NS | NS |
| Huang et al., 2013 | P300 | Elbow rehabilitation robot | None | NS | NS | Panel with 25 commands | NS | SVM | Online: 90.82 | NS | NS |
| Okahara et al., 2018 | SSVEP | Neuro-prosthesis | 3-ALS | NS | 1: Oz | 4 × 4 LED flicker at 32–54 Hz | PSD | Classification Threshold | Online 83.3 | NS | NS |
| Xu et al., 2021 | SSVEP | Upper Limb Exoskeleton | None | 5 | 6: O1, O2, Oz, P3, Pz, P4 | 4 Flickering squares at 8.57, 10, 12, 15 Hz | Frequency domain | CCA | Offline: 86.1 | NS | NS |
BLDA, Bayesian linear discriminant analysis; CCA, canonical correlation analysis; CSP, common spatial patterns; DC, duty cycle; FLDA, Fisher’s Linear discriminant analysis; LDA, linear discriminant analysis; NS, non-specified; SVM, support vector machine; VR, virtual reality; CV, cross validation; FES, Functional Electrical Stimulation; MI, motor imagery; SSVEP, steady state visually evoked potentials; SCI, spinal cord injury; HSD, harmonic sum decision; STFT, short-time Fourier Transform.