Combined action observation- and motor imagery- based brain computer interface (BCI) for stroke rehabilitation: a case report


 Introduction: Upper extremity impairment is a problem usually found in poststroke patients, and it is seldom completely improved even following conventional physical therapy. Motor imagery (MI) and action observation (AO) therapy are mental practices that may regain motor function in poststroke patients, especially when integrating them with brain-computer interface (BCI) technology. However, previous studies have always investigated the effects of an MI- or AO-based BCI for stroke rehabilitation separately. Therefore, in this study, we aimed to propose the effectiveness of a combined AO and MI (AOMI)-based BCI with functional electrical stimulation (FES) feedback to improve upper limb functions and alter brain activity patterns in chronic stroke patients.Case presentation: A 53-year-old male who was 12 years post stroke was left hemiparesis and unable to produce any wrist and finger extension.Intervention: The participant was given an AOMI-based BCI with FES feedback 3 sessions per week for 4 consecutive weeks, and he did not receive any conventional physical therapy during the intervention. The Fugl-Meyer Assessment of Upper Extremity (FMA-UE) and active range of motion (AROM) of wrist extension were used as clinical assessments, and the laterality coefficient (LC) value was applied to explore the altered brain activity patterns affected by the intervention.Outcomes: The FMA-UE score improved from 34 to 46 points, and the AROM of wrist extension was increased from 0 degrees to 20 degrees. LC values in the alpha band tended to be positive whereas LC values in the beta band seemed to be slightly negative after the intervention.Conclusion: An AOMI-based BCI with FES feedback training may be a promising strategy that could improve motor function in poststroke patients; however, its efficacy should be studied in a larger population and compared to that of other therapeutic methods.Trial registration: Thai Clinical Trial Registry: TCTR20200821002. Registered 17 August 2020, http://www.thaiclinicaltrials.org


33
Stroke is a major cause of global deaths, and most stroke survivors usually have 34 hemiparesis on one side of the body that greatly affects their activities of daily living (ADLs) (1). 35 In particular, weakness of the wrist or hand muscle is a common problem in poststroke patients 36 that vastly impacts their ADLs, such as eating, dressing, and opening a door; moreover, it is rarely 37 completely improved. Therefore, it is important to create an effective therapeutic method to 38 improve upper limb function in poststroke patients (2). 39 Currently, constraint-induced movement therapy (CIMT) is an effective therapy that can Hence, there should be some solutions to solve this problem and help these patients regain function 44 in their upper extremities (2, 3). 45 Motor imagery (MI) is a mental simulation of a movement without an actual action (4). It 46 is one of the therapeutic techniques that may be appropriate for poststroke patients who are unable 47 to move their limbs because MI can activate brain areas involved in movement execution; thus, 48 MI may be a promising therapeutic method for poststroke patients to improve their motor function, 49 especially upper extremity function (5, 6). Nevertheless, it is difficult for a therapist to determine 50 whether a patient is performing MI effectively. Thus, brain-computer interface (BCI) technology 51 also plays a key role in fixing this problem (7).

52
A BCI is a system that can monitor brain activity and translate an ongoing signal to be a 53 control signal that is used to command external devices to achieve a user's purpose or desired task. 54 Currently, noninvasive electroencephalogram (EEG)-based BCI is a popular method usually used 55 to decode a brain signal during MI and provide neurofeedback such as images, robots, tactiles and 56 functional electrical stimulations (FESs) backward to a user to inform MI performance and 57 enhance the learning process (3, 7). From EEG studies, it has been well known that executing MI 58 produces a phenomenon called event-related desynchronization (ERD). ERD is power attenuation 59 of the ongoing EEG signal in a specific frequency band, especially in the alpha or mu band (8 -60 13 Hz) and beta band (20 -24 Hz). ERD usually occurs prominently over sensorimotor areas and 61 is associated with motor cortex activation (8, 9). In BCI systems, ERD occurrence is always used 62 as a spectral feature to indicate MI and provides meaningful feedback backward to a user to 63 encourage the learning process, which is a key factor of neural plasticity ( Normally, the effectiveness of MI and AO for improving motor function in poststroke 77 patients has been studied separately; however, recent evidence from EEG, functional magnetic 78 resonance imaging (fMRI), and transcranial magnetic stimulation (TMS) studies has revealed that 79 combined AO and MI (AOMI) can provoke the activation of brain areas related to motor function 80 to a greater extent than pure MI or AO alone. AOMI imagines an action in terms of a movement 81 sensation concurrently with observing the same action displayed on the screen (20, 21). However, practice guided by synchronous AO was improved more than that in the participants who received 89 MI practice guided by asynchronous AO; however, they did not give any neurofeedback to the 90 participant while executing the cognitive task. The same result was found in Wang's, although 91 their intervention was different. In Wang's study, the participants had chronic stroke, and they also 92 provided robotic hand feedback to the participant while they were performing the cognitive task.

93
The purpose of this case report was to support the concept of using AOMI to recover upper    The participant received the AOMI-based BCI with FES feedback training for 3 days per 131 week for 4 consecutive weeks. On each training day, he had to execute the cognitive task for 6 sets 132 with resting time between sets for 3 minutes, and each set comprised 20 trials. The EEG data from 133 the first 2 sets were used to create the classification model, and FES feedback was not activated in 134 these sets. The classification model was applied in the next 4 sets to control the FES device, so 135 there were a total of 80 trials of FES feedback on each training day.

136
In each trial, the computer provided the sequences of cognitive tasks for the participant to 137 perform, which are shown in Figure 2. First, the participant started by looking at the blank screen 138 for 5 seconds. Then, a black cross appeared in the center of the screen for 3 seconds to warn the 139 participant to prepare himself for executing the coming task, and this stage was called the 140 "Preparation stage". Next, the video-guided movement that demonstrated extension of the left 141 wrist and fingers in first-person view was played on the screen for 5 seconds. At this moment, the 142 participant was asked to attentively look at the screen and simultaneously imagined as if he was 143 extending his left wrist and fingers, and this stage was called the "AOMI stage". After that, the 144 screen was blank to inform the participant to relax, and the relaxation time was random between 145 10 and 13 seconds.  Data pre-processing 174 We used a notch 50 Hz filter to remove the power-line noise and common average 175 reference (CAR) for re-reference EEG data. A bandpass filter at frequencies of 8 -30 Hz was used 176 to filter the EEG data because ERD occurred prominently in this frequency range (12). were changed from -0.03 to 0.47 and from -0.08 to -0.18, respectively, which showed in Table 1.

288
There may be many factors that explain why our intervention improved motor function in 289 the participant. First, the benefit from AOMI practice. It is well known that MI is a mental practice 290 that can access or activate the brain areas associated with motor execution, including the 291 supplementary motor area, premotor cortex, primary motor cortex, inferior/superior parietal 292 lobule, basal ganglia, and cerebellum, without physical movement (29), but its drawback is that it 293 is difficult to perform and depends on the cognitive ability of the patient. While AO is easier than 294 MI to practice, it can also provoke brain regions involved in physical movement (21). However, it 295 rarely activates the primary motor cortex (29), which is important for the recovery of motor 296 function (30). Thus, AOMI may play a crucial role in fixing these problems; moreover, previous 297 studies have shown that AOMI can activate corticomotor areas to a greater extent than MI or AO 298 alone (31-33). In this study, we provided video-guided movement, which showed movement of 299 the wrist and finger extension to the participant while he was executing MI. The video-guided 300 movement may have made him focus on kinesthetic MI more easily and required a lower cognitive 301 demand to perform the task; therefore, he could perform the cognitive task effectively 302 corresponding to the averaged classification accuracy was 83.85 percentage, which might 303 contribute to improving motor function.

304
Second, the EEG-based BCI with a neurofeedback system; although AOMI practice could 305 promote the activation of the brain areas relating to an actual movement, it still lacks feedback, 306 which is a key factor in the motor learning process (34 inspection to ensure that the participant did not use any signal artifacts from any part of the body 344 movement to be the control signal in every trial. Finally, we did not know exactly which parts of 345 the brain were damaged from the stroke because his stroke onset occurred 12 years ago, so his 346 medical information was eliminated, and he felt inconvenient for MRI examination again.

348
we would like to support the concept of using an AOMI-based BCI for stroke rehabilitation, 349 and our results have shown that it can improve upper limb function in chronic stroke patients.

350
Additionally, because of its advantages, we believe it may be a promising strategy used to improve 351 motor function in poststroke patients. Competing interests 374 The authors declare that they have no competing interests.