In this study we investigated whether people with SCI and with CNP can practice NFB on their own or with the help of their caregivers at home. Although all participants suffered from long standing CNP their level and completeness of injury varied, which enabled us to investigate the needs of people with different levels of disability. Data had been collected in a home environment, experiencing environmental noise, while using a consumer grade EEG.
The challenge in developing home based BCI was to create a system that was easy to use, reliable and accessible to people with disabilities and their caregivers. Notwithstanding was a requirement to create an inexpensive system by using consumer grade EEG and a freeware software platform, in order to provide at the same time BCI to a relatively large number of end users within a limited budget. We used C++ rather than Matlab which is cheaper and adequate for tablet computers. In the future C++ can be used with mobile phone applications.
Effectiveness
Five out of twenty initially recruited participants were either not able to control NFB or found the protocol too demanding (difficult to use or time consuming). However, all fifteen interested participants learned how to use the system within four sessions. There are several factors which contributed to successful learning. We created custom made user manuals and instructed participants to take photos or videos of themselves as a reminder of the setup procedure. Other research groups have also reported recording EEG from the central cortex using Emotiv, however this was performed by researchers rather than by participants [
63]. An additional facilitating factor was that the consumer EEG device used has been designed for non-professionals and has additional sources of information on the Internet. The donning time in this study was up to 15 minutes, comparable to results of studies managed by caregivers only [
31]. This time would probably be reduced if the headset had been originally designed to cover the central cortex. During NFB with the Epoch, we recorded EEG from one electrode location only to minimise the setup time, though we instructed participants to check impedance levels of all electrodes.
Participants reported low levels of stress when using BCI for the first time at home on their own. We did not use NASA task load index [
67] because asking participants to fill this questionnaire regularly at home would likely result in a low level of compliance. The answer that we collected was based on recollection, during participants’ check-up visits to the hospital.
Although previous studies showed that Epoch had reasonable performances [
63,
68], they were performed under laboratory conditions. Our results showed that in the home environment, only 10–20% of recorded data was very noisy, which is an extremely encouraging result for future real world BCI applications. On the other hand, the physical design of the headset which was not originally created to record EEG from the central cortex was the major problem, in particular for those with smaller heads or thick hair. To the best of our knowledge, at the moment there is no other consumer grade multichannel EEG device that is designed to record brain activity with non-gel electrodes from the sensory-motor cortex, costing under £1000. Inexpensive technical solutions for EEG recordings (e.g OpenBCI) could be used with custom made headsets but do not provide easy solutions for non-professional users. As noted by Miralles et al. [
30] the price of BCI is currently the biggest restricting factor for large scale usability studies in the home environment.
Most participants preferred training with bars rather than with cars. One reason might be that the car game was not as entertaining as games available with commercial NFB devices but the other reason could be that the GUI with bars provided clearer response-reinforcement association which could be overshadowed by more complex games [
16]. In this study a threshold for NFB was fixed throughout the training, based on the baseline measurement for that day. NFB practitioners occasionally use a “moving” threshold based on the most recent performance in order to provide a reward, irrespective of patient performance. This may however lead to training in an undesired direction [
16] and would make later quantitative analysis difficult.
Efficacy
Efficiency was measured by the ability to control NFB and to achieve reduction in pain. The principle of NFB is operant conditioning, a learning strategy that increases a preferred behaviour and decreases the undesired behaviour by providing a reward or punishment [
69]. Results from the literature show that similar to the general BCI illiteracy problem, some people cannot learn to use NFB [
70]. Prior to this study, we tested the NFB protocol with Emotiv in one 30 min session on eighteen able bodied people [
45]. Fourteen participants increased the alpha power by more than 10%, eight decreased theta and seven decreased higher beta by more than 10%. In the current study, only two participants were not able to control NFB.
An important measure of successful NFB is selectivity, i.e. the ability to regulate only selected frequency bands rather than to increase or decrease the whole frequency spectrum [
60]. Twelve out of fifteen participants selectively modulated frequency bands as required by NFB while three increased or decreased the whole frequency spectrum. Similar to able-bodied participants, participants with CNP were most successful upregulating the alpha band with twelve out of fifteen participants upregulating it by more than 10%, seven downregulated theta and ten downregulated their beta band power by more than 10%.
An important observation is that most participants actually upregulated their individual alpha range (6–8 Hz) which on average was lower than in able bodied people, while some participants created two ‘peaks‘, one around their individual alpha and the other in the 9-12 Hz range. A reduced dominant alpha frequency was reported in several previous studies [
8‐
11] and has been attributed to the thalamo-cortical dysrhythmia [
68]. In our recent study [
12] we found that the reduced alpha power and reduced dominant alpha frequency are both markers of “future” CNP. This NFB protocol was aiming to increase not only the alpha power but also the dominant alpha frequency. However, results of this study indicate that alpha amplitude rather than alpha frequency is related to the reduction in pain. We did not notice an issue with individual alpha bands when previously testing the NFB protocol on able-bodied [
45] because their average dominant alpha frequency was within 8–12 Hz.
The efficacy of NFB was 40% i.e. in 8 out of 20 initially recruited participants, the reduction in pain was clinically significant, i.e. larger than 30% [
61]. These are encouraging results, that should be used to estimate the effect size in a future larger randomised clinical trial. Although results of a single trial cannot be directly compared to the results of a meta-analysis, the meta-analysis results of other pharmacological and non-pharmacological treatments should serve as a desirable target. According to a Cochrane database study, gabapentin, a widely used pharmacological treatment of CNP [
71], has an efficacy of 50%. The efficiacy of NFB should also be compared to the other neuromodulatory treatments of CNP such as rTMS and tDCS, although rTMS is still not available for home use. A recent Cohrane review showed that rTMS on average results in 12%, while tDCS results in 17% short term relief in pain [
72]. That study adopted 15% as a clinically relevant reduction in pain, meaning that only tDCS may result in a clinically significant reduction in pain. If we addopted 15% as clinically significant reduction of pain in this study, then 13 out of 20 participants (65%), would achieve clinically significant reduction of pain.
Looking at the relationship between NFB and the reduction in pain, eight out of twelve participants whose pain was significantly reduced also had a significant increase in the alpha band power. Out of the remaining four patients who had a significant reduction in pain without significantly upregulating the alpha power, two significantly downregulated both theta and beta band power. A multivariate analysis, which was outwit the scope of this usability study may reveal more complex relations between different NFB parameters and the reduction in pain.
For six out of fifteen participants who could walk (ASIA D) pain was the major cause of restricted activity. Five of them used the system for 2 months and achieved statistically significant reduction of pain. This indicates that being able to self-administer NFB when and where needed might have a positive effect on the compliance and the effectiveness of NFB.
A limitation of this study is that there was no explicit test for placebo effect. There are two options to test for placebo: the first one is to occasionally blindfold participants to practice NFB with a pre-recorded session or from another area of the cortex. We tested both approaches in our previous laboratory based study, but this would have been hard to achieve in the present study because of infrequent direct contact with patients and because of transfer learning. The other option to test for placebo effect is to have a control group, which might be unethical in long-term studies due to the level of commitment expected by participants.
The presence of a trusted authority (a researcher or a therapist) may also have a placebo effect [
16]. In this study however, participants practiced NFB on their own. We cannot however exclude that shifting locus of control from external to internal did not have a placebo effect. On the other hand one can argue that sham neurofeedback provided from an active electrode might also induce a similar placebo effect.
In this study, instead of placebo test, we collected patient self-reported descriptions of sensations related to NFB (tingling, pleasant warm sensation etc) [
21] that accompany reduction in pain, to quickly assess whether there was any effect of NFB. We showed that out of 10 participants who significantly upregulated the alpha power, eight achieved significant reduction of pain. We also checked whether pain reduction lasted beyond the NFB training.
In this study NFB training was provided from C4 which is located over the primary motor cortex of the left hand. The same location that proved to be most effective for rTMS and tDCS treatments of CNP [
15]. It is believed that the mechanism of action of these treatments is through collateral neural branching [
73] which explains why it is not necessary to apply NFB for CNP in a somatotopic manner. In [
21] we showed that NFB from C4 also affects the motor cortex symmetrically over the contralateral side (electrode location C3).
Satisfaction is defined by the degree a product is giving contentment or making the user satisfied. Judging by the results of questionnaires assessing participant’s attitude towards novel technology and the perceived usefulness of BCI-FES all participants were “early adopters” of technology. These questionnaires were answered jointly by the participants with SCI and their caregivers and they jointly provided high scores. It shows that for adoption of new assistive or therapeutic technology it is important to motivate both patients and caregivers. An important issue emerged during interviews, that caregivers are typically not available for the whole day and that quick BCI donning is extremely important because of the overall limited time that caregivers could devote to the therapy. Previous studies investigating BCI priorities in people with SCI included only questionnaires without actual use of BCI [
74]. For this reason this study might provide useful information regarding the design of home based BCI studies with the SCI population, beyond NFB.
Results of QEUST showed that participants who completed the study were on average very satisfied with both the device and technical support (services) provided by the research team. According to QUEST the main priorities were functionality, ease of use, comfort and dimensions, which are similar to user priorities from other BCI usability studies. A very encouraging result is that most participants did not feel stigmatised by using a headset in front of other people at home. It should be however noted that not all participants completed the questionnaire and that three participants initially withdrew from the study as they perceived the BCI NFB system to be too complex. Fifteen out of twenty initially recruited participants, who stayed on the study were “early adopters” of technology and their level of commitment and satisfaction might not necessarily translate to more general users of this technology. However, feedback from “early adopters” is valuable for improving the technology to make it acceptable for more general patient population.
We identified four main themes in the semi-structured interviews: effect of NFB, usage pattern, problems with hardware and problems with software. The main effect of NFB was a reduction in pain but participants also reported other positive side effects such as reduced spasm and reduced foot drop, improved sensation and proprioception. Finnerup [
75] recently suggested similarities in neuronal origin of spasticity and CNP, such as differentiation of supraspinal neurons. Thus, a neuromodulatory intervention applied to the motor cortex, which affects one might affect the other phenomena. Tingling and a pleasant warm sensation were also reported frequently accompanying NFB, and these were also typically the first symptoms to NFB. This is indirect evidences of modulation of the sensory-motor cortex that is typically overactive in people with SCI and SCP [
6]. Similar side-effects were also reported in our previous study [
21]. Side effect reported in this study should be interpreted with caution, because they were not systematically monitored. Future, large scale, trials should incorporate sensory and spasticity test in the protocol. Alternatively, a NFB study that focuses solely on spasticity in people with SCI with preserved mobility would provide results that are more conclusive.
We did not separately assess satisfaction of people with SCI and their caregivers because only six out of fifteen participants had a caregiver and because caregivers’ roles varied from patient to patient. Only two tetraplegic patients required a caregiver to initiate NFB software.
An interesting observation was that patients often considered happy episodic memories during NFB, as a mind wandering strategy, indicating the involvement of the default mode network [
76]. This was unfortunately not possible to explore further due to single channel EEG recording.
The single main factor affecting the frequency of use of BCI was other health related issues which are fairly common in people with SCI (e.g. urinary tract infection, skin problems etc.). Our ethical permission did not allow us to get insight into all medications that participants took. However we have checked that during study they did not change dosage of pain medications, did not start any non- pharmacological treatment and did not aquire any neurological issues, that would affect their EEG. We also noticed that any disruption to the daily routine could potentially disrupt the usage pattern, indicating the need for regular monitoring in the early phase of adopting new technology. Other factors influencing frequency of use were available time and pain intensity. However, in some cases family doctors (general practitioners) were concerned about possible side effects of BCI as they were unfamiliar with it. This indicates the importance of educating healthcare professionals i.e. trusted persons for a wider user acceptance of BCI technology.
Problems with hardware mainly arose because the EEG headset had to be tilted, i.e. it was not placed on the head as it was designed for. Some previous studies also used this setup to record EEG from the sensory-motor cortex [
63]. Patients frequently used elastic bands to keep the hardware in place. Getting a good electrode contact with wet electrodes was also an issue mostly during the first few sessions but was a reason for abandoning the study by patients with thick or long hair. Forgetting to use a dongle or switch the device on were also frequent reasons for seeking help from the research group. Online monitoring of signal quality and automatic noise removal would be a bonus for any future BCI hardware designed for home use [
77].
Software was not a source of concern because it has been thoroughly tested on able-bodied volunteers [
45]. The largest problem during the first few sessions was remembering the instructions. Based on participants’ feedback, we plan to incorporate the following features in the next version of the NFB software: simpler measure of NFB performance with an electronic diary (current version does not have a diary); step-by-step instruction which could be switched off once participants become experienced with the software, larger font warning messages, a simplified one page software instruction and a car game GUI with better graphics. The NFB software has been designed in a way that it is independent of the hardware so in the future it could be used with custom-made headsets which are designed to cover the sensory-motor cortex.
Another feature that would be useful to incorporate is an audio warning related to the on-line monitoring of the signal quality. Interestingly, most participants reported that they could recognise from the dynamic of the visual feedback with bars whether they were successfully doing NFB or if the signal was of poor quality. That might explain why GUI with bars was preferred by all users, as it provided direct control of all features that should be controlled by NFB. We did not specifically test for the locus of control [
78], though “being in control of pain” was probably a major drivers behind this treatment.