Brain-Computer Interfaces (BCIs) are technologies that enable users to control applications such as video games (Kerous et al., 2018) or wheelchairs (Li et al., 2013), solely through their brain activity. Beyond these control applications, BCIs can be used for neurofeedback (NF) training with the objective of learning how to modulate our own cerebral activity, not in order to control something, but to improve or restore cognitive or motor skills. BCI-based post-stroke motor rehabilitations are in this second category and have demonstrated their efficacy to improve patients' motor and cognitive abilities (Cervera et al., 2018; Bai et al., 2020; Nojima et al., 2022). In the coming years, they are expected to substantially improve post-stroke subjects' quality of life (Nojima et al., 2022).

In classical motor rehabilitation, when subjects have no residual movement, i.e., when they cannot move their affected limb at all, physical practice is impossible and both subjects and therapists must rely on mental practice alone. Here, in mental practice, we include motor imagery (MI) as well as attempted movements. In concrete terms, therapists usually ask the subjects to perform MI or to try to move their arm (attempted movements), and simultaneously stimulate the limb by mobilizing it or, for instance, by using functional electrical stimulation (FES-which consists in stimulating peripheral motor nerves in order to artificially generate movements). While associated to encouraging results (Sharma et al., 2006), the difficulty encountered when trying to demonstrate the efficiency of this procedure might be related to the impossibility to assess the patients' compliance when they are asked to perform MI tasks (Sharma et al., 2006). In addition, we believe pure mental practice-based rehabilitation procedures present two main limitations. The first one is due to the impossibility of the therapist to know when, exactly, the patient imagines moving or tries to move. Therefore, the feedback patients are provided with will most likely not be synchronized with their MI or movement attempts. A second limitation concerns the constant reminder that the patient gets when the therapist asks them to move their arm and that they are unable to do so. Post-stroke subjects experiencing high anxiety levels (Burton et al., 2013), this method might also have detrimental psychological effects, potentially resulting in the patient disengaging from the rehabilitation procedure, and in the therapy being less efficient.

In this context, BCIs are very relevant as they enable the detection of MI/attempted movements of the impaired limb, which are underlain by modulations of the so-called sensori-motor rhythms (SMRs)—as defined in the BCI field by a large band covering mu (μ) and beta (β) rhythms (8–30 Hz) (Pfurtscheller et al., 2000)—, and provide the patient with a synchronized NF, for instance using FES that triggers an arm muscle contraction, or visual feedback [movement of a virtual hand on a screen (Pichiorri et al., 2015)]. Such a NF training enables the participants to train to voluntarily self-regulate their SMRs in a closed loop process, which should favor synaptic plasticity and motor recovery (Jeunet et al., 2019).

While this is encouraging, BCI efficiency is still far from the level required to achieve the clinical breakthrough expected by both clinicians and patients. Thus, BCIs remain barely used in clinical practice, outside laboratories (Kübler et al., 2014). BCI efficiency is known to be modulated by several factors. Many researchers are working on improving this efficiency either from a “technical” point of view (e.g., signal processing Lotte et al., 2018), or—less often—from the human learning standpoint (Pillette et al., 2020; Roc et al., 2021). This is an important step forward: reaching high efficiency is a necessary condition for BCI adoption. Nonetheless, it might not be sufficient for those technologies to be actually used in a clinical setting: fully optimized BCIs (in terms of sensors, signal processing, and training procedures) are pointless if patients and clinicians are not able or do not want to use them, i.e., if BCIs are not accepted (Blain-Moraes et al., 2012). For instance, misconceptions that patients and their entourage have regarding BCIs may have a detrimental effect on the acceptance of these technologies. BCI acceptance could also be altered by the fact that most stroke patients experience depression, and therefore high anxiety levels (Burton et al., 2013) that have a detrimental effect on BCI acceptance and learning (Jeunet et al., 2016). Thus, BCI acceptance is likely to have a major impact on the patients' learning processes and therefore on the efficiency of BCI-based stroke rehabilitation procedures. We hypothesize that identifying acceptability and acceptance factors will help us overcome these misconceptions and personalize the procedures, which will in turn result in reduced anxiety, and increased motivation and engagement levels for the patients. This should favor their learning and, ultimately, motor recovery. In other words, we expect that improving the acceptance levels of BCIs will result in an increased efficiency of these technologies and therefore contribute to their democratization.

Thus, it is crucial, when designing stroke rehabilitation procedures, to consider technology acceptance as a lever to optimize BCI efficiency—in terms of motor recovery. Yet, BCI acceptance remains an aspect that has been little studied to date. To the best of our knowledge, only (Morone et al., 2015) assessed the relevance of a BCI-based stroke rehabilitation procedure of the upper limb using acceptability and usability measures as primary criteria (pilot study, N = 8 patients). The acceptability and usability were measured in terms of mood, motivation, satisfaction, and perceived workload. Indeed, in the BCI field, acceptability is mostly assessed as an attribute of the user's satisfaction, itself being a dimension of user experience (Kübler et al., 2014; Nijboer, 2015). Morone et al. (2015) concludes that the BCI training was “accepted with a good compliance/adherence.” The same conclusion was drawn in the context of a BCI dedicated to a gamified cognitive training for the elderly (Lee et al., 2013) as well as in different studies dedicated to the acceptance of BCIs by Amyotrophic Lateral Sclerosis (ALS) patients (Huggins et al., 2011; Blain-Moraes et al., 2012; Nijboer, 2015), which is the clinical condition for which acceptance has been the most investigated (Nijboer, 2015). Using a focus group approach, Blain-Moraes et al. (2012) have shown that both personal and relational factors impacted BCI acceptance. The personal factors included physical (pain, discomfort), physiological (fatigue, endurance) and psychological (anxiety, attitude toward the technology) concerns, and the relational factors included corporeal (electrode type), technological (relationship between BCI and other type of software and hardware) and social (appearance, training and support personnel) factors. The relational factors had a stronger impact than the personal ones. In the same line, Huggins et al. (2011, 2015) led qualitative studies to assess the influence that different factors (physical interface, setup and training, acceptable performance, task and feature priorities) had on BCI acceptance in ALS patients and patients who had undergone a spinal cord injury. The functions provided by the BCI were rated as the most important feature, together with the ease of use of the system and the availability of a stand-by mode. Finally, Geronimo et al. (2015) have shown that behavioral impairments such as apathy and mental rigidity had a negative impact on ALS patients' BCI usage behavior. Furthermore, the fact that they performed a pilot study during which patients appeared to have a low perceived control over the system altered the perceived usefulness of the BCI. This latter study is the only one in the field of BCIs that assessed acceptance in terms of usage behavior and perceived usefulness. These are concepts from the field of psychology and ergonomics, which we draw on in the next paragraph (Kaleshtari et al., 2016). As claimed by Kaleshtari et al. (2016), who designed the first Model of Rehabilitation Technology Acceptance and Usability (RTAU), in order to be effective, rehabilitation technologies have to be used and therefore accepted by the patients and their families. According to Kaleshtari et al. (2016), this acceptance depends both on personal features, technology features, and social influence. The domain-specific literature indeed suggests that BCI acceptability and acceptance seem to rely on “subjective technical confidence and positive attitudes toward the use of technologies” (Morone et al., 2015). The EEG cap characteristics (gel, montage, and time to set up) seem important both for patients and caregivers (Morone et al., 2015; Nijboer, 2015). Generally speaking, the simpler the better for them (Huggins et al., 2011). Nonetheless, the EEG technology used and the reliability/discomfort trade-off, together with all the BCI-related characteristics, need to be thought of in light of the characteristics and requirements of the application (e.g., level of reliability required) as well as in light of the profile of the patient. This is why BCI-based stroke rehabilitation procedures should be carefully adapted to the training context of each patient. In this spirit, Kübler et al. (2014) proposed an inspiring approach, suggesting to “shift from focusing on single aspects, such as accuracy and information transfer rate, to a more holistic user experience” using User-Centered Design. User-Centered Design has since then been shown to contribute to the acceptability and usability of a BCI-based stroke rehabilitation procedure (Morone et al., 2015) and, more broadly, improved user experience has been suggested to enhance user acceptance and increase the performance of BCI systems (Gürkök et al., 2011).

The concepts of acceptability and acceptance were introduced in order to understand what led users to adopt or not a new system (Alexandre et al., 2018). The adoption of a technology refers to a use that is maintained over time, i.e., without abandonment. In concrete terms, acceptability measure is an evaluation of the user's behavioral intention (BI) i.e., their intention to use the studied technology. The main determinants of BI are perceived usefulness (PU) and perceived ease of use (PEOU). PU is the personal feeling about utility of the system, and PEOU the degree of belief to which using the system will require little or no effort. Acceptability and acceptance differ by the moment they are measured at: acceptability concerns the user's standpoint before any interaction with the system, while acceptance comes after at least one first use.

To the best of our knowledge, there is no model of BCI acceptability yet. Thus, our goals were to (i) create a theoretical model of acceptability—based on the literature—including the factors influencing BI toward BCIs, especially in the context or motor rehabilitation after stroke; (ii) implement a questionnaire from this model to assess BCI acceptability in general public; (iii) validate the structure of our model and questionnaire; and (iv) quantify the impact of the different factors included in the model on acceptability.

With this study we targeted the general public. This enabled us to collect the opinions and attitudes of a large sample of persons, representative of the adult population in France, and thereby to capture an estimation of BCI acceptability in the overall population that we will, in the future, compare with the one of patients and clinicians. Targeting the general public seems particularly relevant in this case for two reasons. First, due to the high prevalence of stroke that is one of the leading causes of disability in adults in France (Accident Vasculaire cérébral, 2019) and results in many of us being concerned, more or less directly, by this pathology and associated rehabilitation techniques. Second, due to the fact that the opinion and attitude of their close relatives will influence the patients' acceptability levels (Venkatesh et al., 2003).

In this paper, we first explain our methodology to achieve these objectives in Section 2: (i) the design of the model, (ii) the implementation of the associated questionnaire, (iii) the validation of these model and questionnaire, and (iv) the quantification of the influence of the different factors included in the model on BCI acceptability. Then, we present the results of this methodology in Section 3, which are based on data collected on a sample representative of the population of France (N = 753). We finish with a discussion about limitations and benefits of our research.

In this section, we present the acceptability model we built, the results of the questionnaire, our analyzes to validate these latter and finally those for quantifying the impact of the different factors on BCI acceptability.

We performed these analyzes using data from Qualtrics, after measuring the average score of each factor for every respondent (as explained in Section 2.2, a factor is measured by several questions which we had to average). For the qualitative/categorical variables, we calculated the number of occurrences of the sub-modalities.

This paper provides the following contributions. First, we designed a first-of-its-kind model of acceptability of BCIs for motor rehabilitation after stroke. This model is based on the literature, and notably on three validated models: TAM3 (Venkatesh and Bala, 2008), UTAUT2 (Venkatesh et al., 2012), and CUE (Thüring and Mahlke, 2007). Second, we created, based on this model, a questionnaire to assess acceptability. This questionnaire follows the structure of the model and includes 3 to 5 items to measure each of the factors. The quantitative items are represented as analog visual scales for which participants move a cursor from “do not agree at all” to “perfectly agree.” The position of the cursor is then translated into a score (from 0 to 10). The scores of the items measuring the same factor are averaged in order to obtain a robust estimation of this factor that is not (or at least as little as possible) dependent on the (mis)understanding of the item or on the state of the person when they answered the question. We distributed this questionnaire to a sample representative of the adult population in France (N = 753). This large and representative sample theoretically ensures the reliability of our results. Third, we performed analyzes on the data obtained to validate the structure of the model. More specifically, we assessed on the one hand the internal consistency of the factors using Cronbach's alpha analyzes. This enabled us to verify the relevance and complementarity of the items used to assess each factor.

On the other hand, we performed a confirmatory factor analysis to evaluate the internal consistency of the questionnaire, or in other words the relevance of the structure of the model. Finally, this is the fourth contribution, we quantified the impact that the different factors had on our target factors (PEOU, PU and BI) in order to identify the factors that influence the most BCI acceptability in the general public. To do so, we used two complementary methods: mediation analyzes and regressions based on random forest algorithms. The first one assessed this influence by taking into account the structure of the model while the second was independent from that structure. Our results show that BCIs are associated with high levels of acceptability for motor rehabilitation after stroke in the general public, and that the intention to use these technologies in that context is mainly driven by the perceived usefulness of the system, itself being mostly influenced by some characteristics of the system, and notably the benefits on risk ratio and scientific relevance. Facilitating conditions, and notably ease of learning and playfulness are the main determinants of the perceived ease of use. Finally, the subjective norm significantly influences the three target factors. With this methodology and results, our study is a first step toward an in-depth consideration of acceptability of BCIs for motor rehabilitation procedures after stroke.

For now, the model and questionnaire, while (we hope) insightful, are not really usable in practice due to their length and complexity. We voluntarily used an exploratory approach by including all the potential influential factors in our model, considering that the literature in the field did not enable us to have strong a priori. The extensive dataset collected enabled us to obtain first indications of the most influential, and therefore most relevant-to-assess factors. More data should now be collected to (un)validate those first results and refine the estimation of the impact that each factor has on BCI acceptability. Our objective is, ultimately, to design a shorter and more usable questionnaire that will enable the prediction of BCI acceptability based on a few factors (and so few items). This prediction could provide scientists/clinicians some indications on how to adapt the procedure, including the instructions, tasks, feedback and training environment, to favor BCI acceptability. As mentioned in the introduction, high acceptability levels could serve as levers to improve BCI efficiency. A main result of this study is that, globally, acceptability levels in terms of behavioral intention seem to be very high in the general public (with an average score of 8.23/10). This is consistent with other BCI acceptability studies (Al-Taleb et al., 2019; Voinea et al., 2019; Benaroch et al., 2021) who reported average scores of 8.0/10 (Al-Taleb et al., 2019) and 6.0/7 (Voinea et al., 2019) for perceived usefulness.

The analysis of Cronbach's alpha revealed that all the factors from the TAM3, UTAUT2 and CUE questionnaire were associated with high-quality internal consistency, i.e., scores were between 0.70 and 0.95 (Cortina, 1993; Tavakol and Dennick, 2011). It was not the case for some of the factors we added (in complement of those from TAM3, UTAUT2 and CUE) to fit with specificities of BCIs, namely, agency (0.50), autonomy (0.52), ease of learning (0.57) and benefits/risk ratio (0.62). This might be due to inadequate wording of the items. It should be noted though that the items used to assess autonomy were directly extracted, word-by-word, from the “Sociotropy-autonomy scale” (SAS, Husky et al., 2004), while those used to measure agency and ease of learning are reformulations (adapted to the context of BCIs) of items from the French adaptations of the Sense of Agency Scale (F-SoAS, Hurault et al., 2020) and of the System Usability Scale (SUS, Gronier and Baudet, 2021), respectively. In the future, it would be relevant to i) collect more data to assess the significance of this result, and ii) lead investigations regarding the comprehension of those items by potential BCI users, and maybe to reword them to increase the internal consistency to the associated factors.

Regarding the extreme score of internal consistency obtained for BI2 (0.97), we hypothesize that it might be due to the repetition of the items. Indeed, when participants saw the same items a second time, it might have happened that they automatically put some scores without really thinking about it, due to perceived redundancy. This hypothesis is supported by the fact that PU2 was also associated with very high, internal consistency scores (0.95) -while still in the “acceptable range.” This high score might also be due to a ceiling effect on those dimensions. Indeed, PU1 and BI1 were already rated with high scores (7.87+/1.63 and 7.88+/–1.73, respectively). After the second video, participants globally increased their rating and gave PU2 scores of 8.28+/1.57, and BI2 scores of 8.23+/–1.69. Thus, the range of values attributed to the items of PU2 and BI2 was narrow, resulting in low variability and thereby very high consistency within those dimensions.

To conclude on the validity of the questionnaire, while it is certainly not perfect yet -we hope that the community will help us improving it by collecting data and suggesting modifications- analyzes have globally revealed i) good internal consistency (as measured by Cronbach's alpha scores) for a large majority of the factors, and ii) a relevant structure of the model (as measured by the confirmatory factorial analysis).

If we have a closer look at the factors influencing the intention to use BCIs, thanks to the random forest-based regression analyzes, we notice that our different analyzes are consistent, notably in showing no significant impact of individual differences, including demographics (age, gender, socio-professional category) or cognitive/psychological profile (autonomy, anxiety, and self-efficacy). Yet, BCI studies have suggested an influence of those variables on BCI performance and learning (Burde and Blankertz, 2006; Nijboer et al., 2008, 2010; Witte et al., 2013; Jeunet et al., 2015). It might be possible that the weight of psychological variables such as anxiety or autonomy are stronger in persons with clinical conditions. It might also be the case that this influence is directly on efficiency as high levels of anxiety and low levels of autonomy and self-efficacy are detrimental for learning but does not alter acceptability. This reinforces the relevance of our approach consisting in optimizing acceptability in order to put the users/patients in the best conditions to favor learning despite their clinical condition, and thereby use acceptability as a lever to favor efficiency.

Behavioral intention is mainly influenced by the perceived usefulness of BCIs, itself being mainly determined by the perceived scientific relevance of the technology. This result highlights the importance of informing the population about BCIs, the way they function, and the level of scientific evidence regarding their clinical efficacy. This idea is strengthened by the significant increase of BI and PU scores following the second video in which the benefits of BCIs for motor rehabilitation after stroke are presented. In the same category of “system characteristics,” the benefits on risk ratio of the technology also seems to have a strong impact on acceptability. We hypothesize that this balance, as perceived by the user/patient, may have a moderator effect on the emphasis on scientific relevance that is objectively depicted by scientists and clinicians. Indeed, (irrational) fears or over-expectations may bias the balance making inaudible the scientific discourses.

Another main finding of this study is the influence that subjective norm has on the three target factors. This was expected for PU and BI. Nonetheless, if we refer to TAM3 and UTAUT2, this factor is not supposed to influence PEOU. In our results yet, this is on the latter that the subjective norm has the strongest impact. We hypothesize that the opinions of the patients' close ones, their technophilia and trust in science will, in the case of BCIs, not only play a role on the perceived usefulness, but will also contribute to emphasize or reduce apprehension toward the technology. This in turn may alter the perceived ease of use of the technology. In any case, the fact that social influence contributes in determining acceptability levels by acting on the three target factors reinforces the relevance of informing the general public, in which patients' relatives are included, to favor the acceptability and adoption of BCIs.

Finally, facilitating conditions, and especially ease of learning and playfulness, are the main determinants of PEOU that, while not influencing BI directly, significantly impacts PU. We believe that this result should encourage us to keep in mind that instructions should be clear and training motivating when we design BCI procedures. This will enable patients to feel confident in their ability to use a BCI. Providing an engaging environment can also be a way to make training more accessible. These results are consistent with the guidelines for successful MI-BCI training (Roc et al., 2021). The question of the transferability of this result to populations of patients could be raised. Indeed, the general population, while they do not need to use BCIs for rehabilitation, may perceive BCIs as a “toy,” which could explain this result. In fact, playfulness has also been shown to increase the compliance of patients in the rehabilitation process in other fields (Burke et al., 2009; Korn and Tietz, 2017; Lopes et al., 2018).

This question of differences between populations is definitely relevant. While we can assume some similarities and differences based on the literature, it will be necessary to lead the same approach with patients and clinicians in order to confront the results, deepen our knowledge and increase our ability to adapt BCIs accordingly. Once more, this will be a lever to improve BCI efficiency. Beyond the differences depending on the status of the respondents (patients, clinicians, and general public), there might also be differences related to their culture (Straub et al., 1997). Therefore, it also seems necessary to apply this approach on different populations around the world.

Collecting more data on diverse populations will enable us to refine our model. It is classic for acceptability models to evolve and to be adjusted to the time and context. The two versions of the UTAUT give us a perfect illustration of the necessity of adaptations. Indeed, whereas the first one was rather adapted to technologies for organizations (Venkatesh et al., 2003), the second one gravitates toward individual consumers/users (Venkatesh et al., 2012). For appropriate adaptations to be made, an open science approach will be necessary. Indeed we think that it will be possible only if people collect data, share their findings and work together on improving the soundness and reliability of the model.

This paper is dedicated to the general public acceptability of BCI-based post-stroke rehabilitation procedures. We are conscious that collecting the opinions of post-stroke subjects and caregivers is also essential. We are currently working on this, conducting questionnaires and semi-structured interviews with post-stroke subjects and caregivers. This will allow us to investigate whether the acceptability factors that stand out the most are similar to those of the general public, and if not, to try to understand what could be the cause of these differences and how to move toward more personalized acceptability models, adapted to the targeted users.

The raw data supporting the conclusions of this article will be made available by the authors upon request, without undue reservation.

The studies involving human participants were reviewed and approved by Institutional Review Board of Toulouse Federal University (N°2019-140). The patients/participants provided their written informed consent to participate in this study.

CJ-K, EG, KF, ST, FA, DG, JP, and LB conceived and designed the experiments. EG, KF, ST, and CJ-K performed the survey and analyzed the data. EG wrote the first manuscript. All authors contributed to the article and approved the submitted version.