Table 2.
Narrative summary of procedures and related technical details utilized in the experimental condition of the included studies
| Study | Experimental condition | Channels | Update rate | Pre-processing | Frequency range (Hz) | Real-time processing |
|---|---|---|---|---|---|---|
| Bhagat 2020 [65] | Participants instructed to first think about movement and then move. A center-out reaching task was performed using a robotic manipulandum with a graphical interface that presented targets requiring either elbow flexion (up arrow) or extension (down arrow). MRCP classifier detection corroborated by EMG triggered assistance; otherwise robot resisted motor attempts. | SMC 15 electrodes | – | – | MRCP (0.1–1) | SVM Classifier |
| Biasiucci 2018 [22] | Classifier built to differentiate active wrist and finger extension from rest. After a start cue was given, FES assistance was triggered. The threshold was optimized to avoid false positives so FES was never activated unless movement occurred. | SMC 16 electrodes | – | Laplacian | SMR (10–12) (18–24) | Gaussian Classifier |
| Chen 2020 [23] | Participants’ hands were inserted into a robotic force feedback device. A computer screen cued them when to move or to rest. BCI-detected activation indicating wrist extension attempt triggered the device to assist movement. | 31 electrodes | – | CSP | Mu & Beta (8–30) | LDA Classifier |
| Chowdhury 2018 [66] | Participants first practiced opening and closing of thumb, index and middle fingers in assist-as-needed robotic device (physical practice). During the NF condition (mental practice), participants were instructed to as slowly as possible extend these fingers when cued from a computer screen showing a virtual hand grasping a remote. At 1.5 s after the cue and for the next 8500 ms intervals, each successful BCI detection triggered visual and robotic feedback of finger opening one step (up to 8 total steps). | SMC 12 electrodes | – | CSP | SMR (8–12 16–24) | SVM Classifier |
| Chowdhury 2020 [67] | Protocol similar to Chowdhury et al. 2018, see above, but with EMG electrodes placed on right and left forearm and collected synchronously with EEG to improve detection of finger opening. | 10 EEG-EMG pairs SMC (12) + finger extension muscles (4) | – | CBPT | EEG: Mu (8–12) EMG: 30–50 | SVM Classifier |
| Cisotto 2014 [59] | Participants performed a center-out reaching task using a robotic manipulandum with a graphical interface depicting targets in 4 directions. Targets to hit were shown in random order, and participants were cued to move to it within 500-740 ms. If successful, the target exploded; if too slow, it turned blue and if too fast, it turned red. ERD detection triggered robotic force assistance proportional to the ERD amplitude. | 3 electrode-frequency pairs | 8 ms | BCI2000 | 10–20 | BCI2000 Linear Classifier |
| Daly 2009 [72] | Participant without index finger extension was instructed to attempt or alternatively imagine movement or relaxation. A red rectangle in the up position on the computer screen cued movement and one in the down position cued relaxation. Accurate detection of real or imagined movement triggered FES to assist finger extension. If ERD power threshold achieved during the task, rectangle turned green. | CP 3 | – | BCI2000 | 5–30 | BCI2000 Linear Classifier |
| Ibáñez 2017 [68] | Participants instructed to reach for a glass target in front of them at 75% of their maximum reach. BCI detection of movement intention triggered a multimodal FES activation reach sequence to assist them. Gyroscopes indicated whether movement occurred to evaluate accuracy of the BCI detection. | 10 best electrode-frequency pair + a virtual channel average of C1,C2, and CZ | 100 ms | Laplacian + ERD threshold identification | MRCP (0-1) & 6–35 | (Naïve Bayes &Match filter) Logistic Regression Classifier |
| Jang 2016 [63] | Participants performed 4 shoulder movements demonstrated on a computer screen. A concentration index (CI), defined as EEG theta to beta power ratio was used as the detection threshold to trigger FES to two muscles that reduced shoulder subluxation during motor tasks. | FP1 | - | CI Threshold identification | Beta (12–15) Theta (4–7) | – |
| Jovanovic 2020 [73] | Participants performed reaching, with or without hand opening when cued by a therapist. ERD detection of movement triggered FES to muscles that varied by participant and task. If detection did not occur, therapist could trigger FES. | C2 | - | ERD/ERS measurement | Mu (8–12) | ERD Detection |
| Marquez-Chin 2016 [74] | Participant performed 5 reaching tasks (to mouth, opposite shoulder, knee on same side, in front and to the side), held the position, and returned to start. ERD detection of movement triggered a multichannel FES neuroprosthesis, if not, therapist could activate FES. Therapist also demonstrated and assisted movement as needed. | FCz | 125 ms | ERD threshold identification | Beta (18–28) | ERD Detection |
| McCrimmon 2014 [70] | Participants attempted ankle dorsiflexion. BCI detection of movement intention triggered FES assistance. If FES not activated, they were to continue movement. If FES activated in error, they were told to not move. | 1 subject-specific channel (CZ, C5 or CPz) | 250 ms | CPCA | Mu & Beta (8–30) | LDA or AIDA |
| Mrachacz-Kersting 2016 [61] 2019 [26] | Participants instructed to dorsiflex as fast as possible once the cursor on the screen moved upward, hold for 2 s then relax. Timing of MRCP peak negativity during dorsiflexion attempt triggered FES assistance. | 10 FP1, F3, F4, Fz, Pz, P3, P4, C3, C4, Cz | – | Laplacian | MRCP (0.5–10) | – |
| Mukaino 2014 [75] | Participants attempted to extend their fingers at maximum effort for 3s as demonstrated visually on computer screen. If ERD threshold reached for 1 s during the attempt FES assistance was triggered. | 2 C3 and C4 | 30 ms | Laplacian | SMR (8–12 18–26) | LDA Classifier |
| Norman 2018 [69] | This study had 3 phases. Phase 1: participants attempted 1 of 4 cued commands to open only index, middle, both or neither fingers, with movement activating a finger extension robotic device. Phase 2: they practiced modulating SMR rhythms with success shown by increasing object brightness on a visual display. Phase 3: SMR modulation success as indicated by increasing object brightness, cued participants to move which in turn activated the robotic device. | 1–3 electrode-frequency pairs | 50 ms | BCI2000 | Beta (12–24) | BCI2000 Linear Classifier |
| Ono 2013 [76] | Participants performed block trials of 5 hand openings for 3 s at maximal effort and 5 rests. If EEG classifier detected the hand opening attempt within 1s of cue to move, FES assistance was triggered. | 2 C3 and C4 | 30 ms | – | Mu & Beta | LDA classifier |
| Osuagwu 2016 [64] | A buffer was established that set the number of consecutive successful BCI detections of either right- or left-hand movement required to trigger the FES. The buffer caused a needle on a gauge on a computer screen to point to 0 when the preset number was achieved and then triggered a multichannel FES to assist repetitive hand opening and closing . If the preset number for detection was not reached, no assistance occurred. Buffer size could be adjusted for difficulty. | Central SMC (3 bipolar electrodes, CP3- CF3, CPz–CFz, CP4–CF4) | – | – | Mu & Beta (7–30) | LDA Classifier |
| Ramos-Murguialday 2013 [4] | Participants performed either hand opening and closing, or moved the entire limb forward and backwards. If ERD was sustained below a threshold for 0.2 s during an attempt, a robotic exoskeleton was triggered to provide assistive force. | Ipsilesional motor cortex | - | ERD threshold identification | Mu (8–13) | ERD Detection |
| Remsik 2019 [41] | EEG classification of attempted grasping activated horizontal movement of a cursor on a screen, triggered FES assistance to facilitate finger opening or closing depending on participant choice and delivered contingent vibrotactile tongue stimulation. | C3–C4 | - | BCI2000 | Mu (8–12) | BCI2000 Linear Classifier |
| Silvoni 2013 [60] | Paradigm similar to Cisotto et al, 2016. | Right affected arm C3, CP1, P3, CP54 Left affected arm C4, CP2, P4, CP6 | 16 ms | BCI2000 | Right affected arm 14–17 Hz Left affected arm 11–14 Hz | BCI2000 Linear Classifier |
| Takahashi 2012 [77] | Participant viewed a red square on a screen to cue ankle dorsiflexion. If ERD detection threshold reached, FES was triggered with stepwise increases or decreases every 500ms, depending on successful detection as shown by changing colors on 8 bars on the screen representing 8 possible steps. | Bipolar FCz-CPz | 500 ms | ERD threshold identification | Beta 24–26 | ERD Detection |
| Vourvopoulos 2019 [71] | Participants wore a head mounted VR display while instructed to attempt wrist and elbow extension. If EEG activation during attempt exceeded baseline, this triggered a virtual arm on a screen to move towards the target (visual FB). | 2 C3, C4 | 500 ms | Laplacian | Mu and Beta 8–24 | – |
MRCP Movement-related cortical potential, SVM Support Vector Machine, FES Functional Electrical Stimulation, SMR Sensorimotor rhythm, CSP Common Spatial Patterns, LDA Linear Discriminant Analysis, SMC Sensorimotor cortex, EMG Electromyography, EEG Electroencephalography, ERD Event-related Desynchronization, AIDA Approximate Information Discriminant Analysis, CBPT Correlation of band-limited power time-courses