Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The brain-computer interface researcher’s questionnaire: from research to application

The brain-computer interface researcher’s questionnaire: from research to application Brain-Computer interfa Ces, 2017 VoL. 4, no . 4, 236– 247 https://doi.org/10.1080/2326263X.2017.1366237 OPEN ACCESS The brain-computer interface researcher’s questionnaire: from research to application M. J. Vansteensel  , G. Kristo, E. J. Aarnoutse  and N. F. Ramsey  Department of neurosurgery, Brain Center rudolf m agnus, university m edical Center utrecht, utrecht, t he netherlands ABSTRACT ARTICLE HISTORY received 3 f ebruary 2017 Over the last decades, the brain-computer interface (BCI) research field has seen steep progress a ccepted 7 July 2017 and now recognizes several types of applications, each with its own target population. Despite this evolution, BCIs seem to remain a research application, and are hardly used in daily life at home, work KEYWORDS or in the clinic. Therefore, as a field, we need to take the next step and make sure that we develop Brain-computer interface; BCIs that can, and will, be used outside the laboratory. To this purpose, BCI researchers worldwide survey; questionnaire; were approached by email with the request to fill out a questionnaire, aimed at laying out the near research; future and far future of the BCI field. We show that researchers are quite optimistic about the feasibility of having BCIs applied in real life. Also, we identify the hurdles that need to be taken before these applications become commercial products and the research activities required to accomplish this. 1. Introduction output that is lost due to injury or disease. A BCI may also be used to control an electrical stimulation device Since the pioneering work of Fetz [1], Vidal [2] and others, that targets dysfunctional muscles or nerves and thereby the field of brain-computer interface (BCI) research has restores their function, or to improve limb control in reha- seen steep progress, as evidenced by the steady increase bilitation settings or the symptoms of neurological and in the number of peer-reviewed BCI articles over the last psychiatric problems using neurofeedback approaches. 15 to 20 years [3]. Over time, numerous studies have suc- Other BCI application scenarios mainly target healthy cessfully or less successfully attempted to maximize both people, and may enhance their day-to-day functioning, performance and multi-dimensionality of BCIs, by opti- for example using BCI-based detection of attention lapses mizing brain signal acquisition methods, signal processing during prolonged driving, or supplement people’s physical and decoding techniques as well as feedback design, and abilities with an extra actuator such as a BCI-controlled by evaluating different concepts or strategies for control. robotic third arm. BCIs can also be used as a research tool At the same time, the definition of a BCI application as to study brain function, such as the neuronal processes of well as that of the target population for BCIs have evolved learning or decision-making. substantially. Whereas early studies were mainly oriented Brain signals for BCI can be recorded non-invasively towards enabling brain-based communication for patients with sensors outside the head (non-invasive BCIs), or suffering from severe paralysis or the locked-in syndrome with implanted electrodes (implanted BCIs). See for (LIS), the BCI field currently recognizes several types of review, [5]. The most well-known non-invasive record- application scenarios for BCIs (i.e. types of devices that ing technique is electroencephalography (EEG), which a BCI can control), each with their own range of target records neuroelectrical activity with electrodes placed populations. A subdivision of five application scenarios on the scalp. Other non-invasive recording techniques was originally proposed by Wolpaw and Wolpaw in 2012 include functional MRI (fMRI) and functional near-in- [3], and was extended with a sixth application scenario frared spectroscopy (fNIRS), both of which measure the (research tool, see below) by the BNCI Horizon 2020 project [4]. In the classic application of a communication hemodynamic responses that occur in active brain areas, tool for locked-in patients, or for example in the case of and magnetoencephalography (MEG). Whereas fMRI BCI control of a wheelchair, the BCI replaces natural CNS and MEG involve large equipment, and are therefore CONTACT m. J. Vansteensel m.j.vansteensel@umcutrecht.nl t he supplemental material for this paper is available online at https://doi.org/10.1080/2326263X.2017.1366237. © 2017 t he a uthor(s). published by informa uK Limited, trading as t aylor & f rancis Group. t his is an open a ccess article distributed under the terms of the Creative Commons a ttribution-nonCommercial-noDerivatives License (http://creativecommons.org/licenses/by-nc- nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way. BRAIN-COMPUTER INTERFACES 237 mainly used as a research tool, fNIRS has the potential to invitation to fill out the questionnaire. This second round be used for BCI applications with end-users. Implanted was closed at the beginning of January 2015. electrodes for BCI may be placed epidurally or subdur- ally (i.e. electrocorticography, ECoG) or may penetrate 2.2. Structure of the BCI researchers’ questionnaire the cortical surface (multi-electrode arrays, MEA). Both e Th BCI researchers’ questionnaire consisted of three sec - recording techniques require surgery to precisely place the tions (see also the Supplemental Material). electrodes, but have the advantage of high temporal and spatial resolution and potential 24/7 availability, making 2.2.1. Section A (Respondents) them interesting candidates for certain BCI application In section A, respondents were asked to answer a list scenarios. of questions, which was set up to characterize each Despite the research progress that we have witnessed researcher by several criteria, such as their background over the past years, BCIs are only sparsely available for and what type of BCIs they work on. Questions within the large range of diseased and healthy target populations this section were multiple-choice questions, and allowed that may be interested in using them in home, work or respondents to select one or more items (depending on clinical settings, and only a very limited number of people the question). Where applicable, one of the choices was use BCI tools outside research settings. In order to make ‘Other’, which was followed by a free-text field, allowing sure that potential end-users, in time, will start to benefit people to describe their answer in words. from the extensive knowledge gained in BCI research, it is important to lay out the future of the BCI field, identi- 2.2.2. Section B (Near future) fying the most promising BCI niches and signaling topics In section B, we asked participants to shortly describe in that need more attention. From November 2013 to May one sentence a potential BCI application that could or 2015, the BNCI Horizon 2020 project, funded within should be feasible within the next 5–10 years (e.g. EEG- the European Commission’s Framework Programme 7, BCI for motor rehabilitation after stroke ), and to assign this worked on the development of a roadmap for the BCI application to one of the six scenarios (replace, restore, field, laying out the near and far future of BCIs [6 ]. One enhance, improve, supplement, and research tool), by of the efforts of the consortium to reach this goal was to selecting the respective item in a multiple-choice list. Also, acquire insight into the opinion of BCI researchers about participants were asked to indicate whether they would the current status and the future of their field. To this want to see this application developed using an implanted purpose, BCI researchers worldwide were approached or a non-invasive approach (multiple choice). Next, par- by email with the request to fill out a questionnaire, ticipants were asked to keep their specific application in asking them for their view on the most promising BCI mind, and indicate to what extent each of a given list of applications, the hurdles that need to be taken before bottlenecks would apply to their application. Each item these applications become commercial products and the on the list (i.e. each possible bottleneck, such as Potential research activities that are needed to accomplish this. A users do not know about this BCI tool) could be rated on a concise summary of the results of this questionnaire was 5-point scale (strongly agree, agree, neutral, disagree, and described in one of the Appendices of the BNCI Horizon strongly disagree), by selecting the associated box. The 2020 roadmap. In the current manuscript, the result of sixth possible answer was ‘not applicable’. A separate list of the research is presented in more detail and we quantified items was used to ask participants what, in their opinion, the qualitative answers respondents gave about the far should be the focus of BCI research in the next 5–10 years future of BCI research. to reach the envisioned application. Here, each item on the list was a possible research direction (e.g. Developing and 2. Methods testing advanced signal-processing techniques for improved system performance), and rating options were the same as 2.1. General for the bottlenecks. e q Th uestionnaire was sent around to 3291 BCI researchers in May of 2014. Each researcher received an email describ- 2.2.3. Section C (Far future) ing the purpose of the questionnaire and containing a link In section C, participants were asked to think out of the to fill it out on a website (Google Form). Email addresses box and into the far future and write in a free-text field a were obtained from several BCI research groups and soci- short statement or description of a potential ‘killer’ appli- eties. Two reminders were sent until the questionnaire was cation or major research breakthrough. This was done for closed on 10 July 2014. Early December 2014, researchers both the implanted and the non-invasive BCI approaches, who did not respond in the first round received a second resulting in two Far future statements per respondent. 238 M . J. VANSTEENSEL ET AL. 2.3. Data analysis to an overall disagreement or strong disagreement with the statement. Statements that received a COM value 2.3.1. Section A (Respondents) between 2.5 and 3.5 can be considered as neither particu- e a Th nswers respondents gave in section A (Respondents) larly relevant nor irrelevant, according to the respond- served to characterize background, expertise, and working ents of our questionnaire. area of the participants. Data are presented as percent- ages in the results section. For many questions, multiple 2.3.3. Section C (Far future) answers were allowed, resulting in more entries than the Responses in the Far future section were analyzed using total number of respondents. To make sure that this fact ‘content analysis’, a highly suitable approach to analyze is represented accurately in the results, percentages were text data [7,8]. The technique has a long history espe- computed relative to the total number of respondents, and cially within the social sciences and health studies. We the total per question could therefore exceed 100%, unless used a conventional or inductive approach, where codes stated otherwise. are directly derived from the text. More specifically, responses of the Far future section of the questionnaire 2.3.2. Section B (Near future) were checked for completeness and clarity. Incomplete For section B (Near future), our aim was to extract from or absent responses in this section, as well as unclear the questionnaire the most important hurdles in the devel- statements or descriptions, were excluded from analy- opment of BCI applications in each of the six scenarios, as sis. Subsequently, in a data-driven approach, the content well as the most urgent research directions to overcome of the statements given by the respondents was used to these. As the first step of the analysis, two of the authors assemble a list of numbered topics (codes) that, together, (MV and GK) evaluated the clarity of the descriptions of covered all issues described by the respondents. Where the BCI applications that were given by the respondents, relevant, codes were grouped into larger categories. Next, as well as the correctness of the assignment to each to the in an iterative process, all statements were annotated with scenarios. Unclear descriptions were excluded from fur- one or more codes from that list by two raters (MV and ther analysis. Descriptions that were clear but incorrectly EA). In three rounds of updating codes and a fourth round assigned to a certain scenario were re-assigned. Where the of discussion, the raters reached agreement about code two raters disagreed, they discussed the application until assignments. For each code, the numbers of instances in they reached consensus. the list of implanted and non-invasive responses were For subsequent analysis, we considered all combina- counted. tions of approach (implanted, non-invasive) and appli- cation scenario (replace, restore, enhance, improve, 3. Results supplement, and research tool), and refer to these combi- nations as ‘scenario-approach combinations’ (e.g. implant- 3.1. Respondents ed-replace or non-invasive-supplement). We counted the Two rounds of the questionnaire resulted in 298 (9.1%) number of responses assigned to each scenario-approach responses. Most respondents (56%) worked in Europe, but combination, and only included in the analysis the sce- we also received a substantial number of responses from nario-approach combinations that received at least 10 people working in North America and Asia (Figure 1(A)). responses. e m Th ajority of BCI researchers had some form of back- To quantify the importance of the each of the bot- ground or training in the field of engineering (62%). tlenecks and research directions for the scenario-ap- Computer science and neuroscience came second and proach combinations, we first labeled each rating with a third with 37% and 29%, respectively (Figure 1(B)). certain weight: ‘not applicable’ with 0, ‘totally disagree’ Notably, many (46%) respondents reported two or more with 1, ‘disagree’ with 2, ‘neutral’ with 3, ‘agree’ with 4, training backgrounds. Respondents covered a wide range and ‘totally agree’ with 5. Subsequently, the center of of positions in their institute, such as post-doc, PhD stu- mass (COM) of the given answers was computed, per dent or head of research and R&D departments. bottleneck/research direction statement, by summing Brain functions used for decoding were mainly motor the product of the number of subjects who assigned a function (74%), but other functions, such as attention and certain rating to a statement and the weight given to visual perception, were also addressed by many respond- that rating. This sum was divided by the total number ents (47% and 45%, respectively, Figure 1(C)). Signals of respondents within that scenario-approach combina- were mainly decoded from healthy subjects (85%) and tion. A COM value of 3.5 or higher meant that most patients (48%), and only a few people worked with ani- respondents agreed or strongly agreed with the state- mals (7%, Figure 1(D)). ment, whereas a COM value of 2.5 or lower was related BRAIN-COMPUTER INTERFACES 239 Figure 1.  respondents’ characteristics. ( a ) Continent where respondents perform BCi research, (B) background or training of the respondents, (C) brain functions respondents use for decoding, (D) type of subjects respondents use in BCi studies. A large majority of the respondents (n = 264, 89%) and 2005, and a decrease in the number of people with predominantly used non-invasive BCI systems, almost all a more recent starting year (i.e. within the last 10 years). based on EEG (95% of researchers working on non-inva- er Th e was an interesting distribution over continents sive BCIs). Gel electrodes and electrodes in a cap were the of respondents working on implanted and non-invasive most popular, but most respondents reported the use of BCIs: only 8% (13/167) of respondents from Europe and other approaches as well (water-based or dry electrodes, 4% (2/57) of respondents from Asia worked on implanted separate electrodes or electrodes in a headset). The distri - BCIs, whereas this percentage was 26% (18/68) for bution of active versus passive EEG electrodes was almost respondents of North America and 18% (2/11) for South equal: 88 respondents only used active electrodes, 70 only America (Figure 2(B)). passive and 94 used both. e r Th espondents who used implanted BCI systems (n = 3.2. Near future 34, 11%) recorded signals mostly with surface electrodes (62% of researchers working on implanted BCIs) that were e 298 r Th espondents of the questionnaire together sub- placed subdurally. Respondents using penetrating elec- mitted 363 BCI applications: 233 people submitted one trodes mostly recorded from cortical structures. A few application, and 65 submitted two applications. The respondents (also) reported recording from subcortical descriptions were unclear in 46 cases and these responses structures. were excluded from analysis, leaving 317 usable descrip- Overall, more than 70% of the respondents entered the tions of BCI applications. In 69 cases, we reassigned the field less than 10 years ago. This was true for the entire BCI application to a scenario different from the one to population of respondents, as well as for the respondents which it was originally assigned by the respondent. working on non-invasive BCIs (Figure 2(A)). The popula - When asked which BCI devices could be feasible within tion of respondents working on implanted BCIs, however, the next 5 to 10 years, most respondents (33%) chose devices showed a very clear peak in starting year between 2001 that replace natural CNS output. Other respondents described 240 M . J. VANSTEENSEL ET AL. Figure 2. implanted versus non-invasive. (a ) Distribution of the years in which respondents started working in BCi research. Gray bars indicate all respondents together. Green and red lines indicate the starting year of researchers working mainly on non-invasive and implanted BCis, respectively. (B) percentage of respondents in implanted (red) and non-invasive (green) BCi research, per continent. Table 1. number of respondents describing an implanted and a non-invasive solution for an application within one of the six scenarios in section B of the questionnaire. Replace Restore Improve Enhance Research Supplement Total implanted 21 6 9 3 4 1 44 (14%) non-invasive 83 11 75 47 19 38 273 (86%) t otal 104 (33%) 17 (5%) 84 (26%) 50 (16%) 23 (7%) 39 (12%) 317 BCI applications that may improve (26%), enhance (16%), 3.2.1. Bottlenecks supplement (12%), and restore (5%) natural CNS output or Interestingly, the ratings given by the respondents to that can be used as a research tool (7%). Most respondents the bottleneck statements indicate that several issues described their feasible application as being developed by apply to all or almost all of the six non-invasive scenar- using a non-invasive BCI system (86%, see Table 1). ios (Figure 3). In fact, participants did not agree with the Most of the replace applications that were mentioned statements that the long-term risks are too high (COM would be used for communication in general and for com- value  <  2.5 for six out of six non-invasive scenarios) munication in locked-in patients in particular (BCI for and that there is insufficient evidence for user safety communication in locked-in patients). Other typical exam- (6/6), indicating that the long-term risks of non-inva- ples of feasible BCI applications that were suggested aimed sive applications are considered acceptable and there to enhance cognitive functions (BCI as an improved tool to is sufficient evidence for this conclusion. Most people enhance daily brain activity, thus reducing the risks of los- agreed or strongly agreed, however, that (long-term) ing attention, memory and health), improve rehabilitation system performance of non-invasive BCIs is not yet (BCI to modulate plasticity in recovery from neural injury), good enough (COM value > 3.5 for 6/6 scenarios), that supplement during gaming or home automation (A sepa- potential users are unaware of the BCI tools (5/6), that rate control for routine tasks allowing users to direct extra current systems are too complicated for home use (5/6), attention to a productive use), restore lost movement or and that the wishes and needs of end- users are not met speech (BCI as a control signal for a FES neuroprosthesis), appropriately (4/6). or be used as a research tool for cognitive assessment or For the implanted replace scenario, participants believe technique-improvement purposes (BCI as a research tool that this approach has clear advantages over other non- to study decision-making, perception or learning). BCI tools and that there is sufficient evidence for this In the subsequent two sections, only groups of 10 or (Figure 3). They also agree that current system durability more scenario-approach applications are included in the and performance are insufficient and that there is insuf- analysis. Whereas all non-invasive scenarios could be ficient evidence for system performance, durability, and analyzed, only the replace scenario was included for the the risk/benefit ratio for end-users. The lack of aware- implanted approach, since only for this scenario a suf- ness about BCIs, which was a general bottleneck for all ficient number of respondents described an implanted non-invasive solutions, was also reported by participants approach (Table 1). addressing implanted replace solutions. BRAIN-COMPUTER INTERFACES 241 Figure 4. BCi research directions. t he opinion of the respondents of the questionnaire on the relevance of each of the given research directions for their near future applications. Colored horizontal bars indicate the center of mass of the responses, with 1 representing ‘totally disagree’ and 5 ‘totally agree’. t he gray shaded area indicates a center of mass between 2.5 and 3.5. research directions with bars that fall completely within this area can be considered as neither particularly relevant nor irrelevant. note that for the implanted approach, only the ‘replace’ scenario could be analyzed, because of lack of responses for the other scenarios. improve system performance. Also, the BCI community Figure 3. BCi bottlenecks. t he opinion of the respondents of the needs to investigate the wishes and requirements of the questionnaire on the relevance of each of the given bottlenecks end-users, and perform clinical trials on system durability, for their near future applications. Colored horizontal bars indicate performance, safety, efficacy (compared to non-invasive), the center of mass of the responses, with 1 representing ‘totally and risk-benefit ratio for end-users. disagree’ and 5 ‘totally agree’. t he gray shaded area indicates a center of mass between 2.5 and 3.5. Bottlenecks with bars that fall completely within this area can be considered as neither 3.3. Far future particularly relevant nor irrelevant. note that for the implanted approach, only the ‘replace’ scenario could be analyzed, because Over 60% of respondents described a far future scenario, of lack of responses for the other scenarios. defining what they considered a major breakthrough in the implanted or non-invasive BCI area, such as a killer 3.2.2. Research directions application or major research advance. Some answers When asked about the focus of research related to were considered unclear by the raters (n = 15 and n = 16 non-invasive BCI applications in the coming 5–10 years for implanted and non-invasive, respectively) and were (Figure 4), participants agreed on the need for the devel- removed from further analysis, resulting in 169 and 178 opment and testing of new sensors (6/6) and new signal- far future statements for implanted and non-invasive processing techniques for improving system performance approaches, respectively. These statements were used to (6/6). Participants also agree that there is a need for clini- define the list of topics (Table 2 ) and were labeled accord- cal trials to demonstrate system performance (5/6) and for ing to these topics. The topics were mostly not mutually identifying the wishes and needs of the end-users (6/6). exclusive, but refer to different aspects of BCIs (e.g. tar - For the research directions on implanted BCI appli- get group, hardware). Some topics are mutually exclu- cations, all statements received a score of 3.5 or higher. sive, but not completely (e.g. healthy subjects, patients). This means that new sensor and amplifier techniques Depending on the description, responses received one are needed, as well as signal-processing techniques that or more labels. 242 M . J. VANSTEENSEL ET AL. Statements most oen ft addressed ‘user friendliness’ environmental control for patients’ (n = 42 occurrences (n = 73 individual occurrences in the total of 347 implanted in 347 statements), ‘prostheses and artificial limbs for and non-invasive BCI statements), ‘communication and patients’ (n = 53), ‘hardware: sensors’ (n = 70), and ‘accu- racy and reliability of signal processing and decoding’ (n Table 2. f ar future topics identified from the statements given by = 52), indicating that these topics represent areas of par- the respondents in the f ar future section. ticular interest for the far future (Figure 5). Topic Subtopic Description erTh e were interesting differences between the 1 User friendliness, daily use, home use, portability implanted and non-invasive far future statements. First, 2 Critical statement about non-invasive or implanted the topics mentioned most oen f ft or these approaches BCIs 3 Ethics, public opinion, privacy differed, in that ‘user friendliness’ and ‘hardware: sen- 4 Commercialization sors’ were described most frequently in the non-inva- a number of end-users sive far future statements, whereas ‘communication and b n iche 5 Applications for patients environmental control for patients’ and ‘prosthesis and a psychiatry, behavior, neurofeedback artificial limbs for patients’ occurred most oen in t ft he b rehabilitation, functional electrical stimulation c epilepsy implanted far future statements (Figure 5). Second, within d1 Communication and environmental control the implanted far future statements, the ‘applications for d2 prosthesis, artificial limbs, exoskeletons, robotic arm e any/other patients’ topic (i.e. taking all of its subtopics together) was 6 Applications for healthy people referred to by 92 respondents, compared to ‘applications a Behavior, learning, stress, skills, neurofeedback b monitoring of health, alertness for healthy subjects’ by 55 respondents. These numbers c Gaming practically mirrored the situation for non-invasive far d military, security, forensic future statements (n = 51 for ‘applications for patients’, e enhanced/augmented reality f enhanced/extra actors and n = 85 for ‘applications for healthy subjects’). In con- g mood, environmental control, smart home, domotics trast, the topics ‘hardware’ (including all its subtopics) was h Car i any/other represented equally in the statements for implanted and 7 Brain research for non-invasive approaches (n = 94 in both cases). Also 8 Robotics, telepresence 9 Hardware the topic ‘signal processing/decoding’ (including all sub- a s ensors topics) was similarly represented in the responses for both b Wireless c Battery life implanted (n = 47) and non-invasive (n = 60) approaches. d s afety For both approaches, quite some truly out-of-the box e price, size, esthetics, comfort statements were given, a selection of which is indicated f signal-to-noise ratio g Brain stimulation, neuromodulation in Table 3. h Durability of system 10 Signal processing, decoding a a ccuracy, reliability, performance, multidimensionality 4. Discussion b optimal brain function for control c speed, information transfer d Calibration, coregistration In the current study, we describe the view of BCI research- e resolution ers about the future of their field, as obtained through 11 Hybrids 12 Relevant for both patients and healthy users the responses of the BNCI Horizon 2020 Researcher’s Figure 5.  f ar future topics. f or each of the topics identified from the f ar future statements (see t able 2), the number of statements referring to these topics is indicated on the y-axis. red indicates counts from the implanted far future statements, green from non- invasive far future statements. BRAIN-COMPUTER INTERFACES 243 Table 3. examples of out-of-the box f ar future statements. Non-invasive Far future statements a domotics device that you would put on when you arrive at your place, giving you full control of all your gadgets and appliances with minimal effort. t his implementation would also be useful for both healthy and disabled subjects BCi for enhanced creativity or logical thinking, e.g. by using a BCi as a feedback loop to help users reaching a mental state in which they are more creative or more logical Due to the common necessity of having to drive the brain activity via specific paradigms in order to get the brain signals into BCi mode (to infer the end user’s intention) compared to other instances when the user does not want to use the BCi, a major research advancement would be to find a brain signature that indisputably could be labeled as the mental switch to turn on a BCi, so it is not paradigm nor stimuli dependent a small signal reader (perhaps magnetic) that will allow to get a clear reading of the head electric potentials in terms of an aa C device, full integration into a device that can function using multiple input modalities. f or example, typing/cursor control, single switch, eye tracking, emG, eoG, and eeG. such a system would provide a solution that an end-user can learn to use while relatively normal function remains. as control becomes less reliable, the control modality can be changed to the most reliable method merging two neural networks for better decision-making Implanted Far future statements s afe wireless system that could be implemented “without” surgery like “nanobots” that would find their way to problematic regions of Cns to replace their functionality regenerating damaged brain areas by planting new brain cells, and train them with external stimulating impulses, to restore basic functionality such as mov - ing extremities in locked-in patients Chronic BCi usage for enhance scenario: answer suggestions pop up if a user raises a question, whatever they are Latest technological developments have reduced the hardware size to a headphone which can be worn easily by users at home. t he real potential application for BCis would be people communicating and controlling electronic equipment through thoughts alone. signals can be read by the home appliances and based on user moods and needs can alter the environment like if a user is sleepy lights can be automatically shut down without the user taking any action. if the end user is thirsty, cool water is served automatically. t his is only one huge potential the BCis could have in 20 years. t omorrow people might communi- cate on mobiles through thoughts alone rather than texts or sms Creating a high-resolution neural spike recording technology that does not require surgery a ccurate decoding from invasive brain signals of the markers of intention/volition/high level controls of the muscles, such high level inputs being used by lower systems (Central pattern Generators) to control the low-levels movements of the limbs. Current BCi do not work this way (they try to decode low-level control) but that’s not what the cortex does Questionnaire in 2014. As far as the authors are aware, of metabolic signals has been suggested as the main reason this is the first survey in which BCI researchers are specifi- that fNIRS and fMRI are not (yet) oen u ft sed for real-time cally asked about what they see as the most promising BCI applications [11], but especially fMRI has a significant applications and to identify the most important hurdles in value in earlier stages of BCI development, such as in getting these products available for end-users. finding and localizing optimal brain functions and areas In the first section of the survey, respondents were for control [12,13]. In this respect, it is interesting to note asked to characterize themselves and their current BCI that, although imagined, attempted, or executed move- work, in order to obtain a view of the state of affairs in BCI ment remain the brain functions that are most oen u ft sed research. The background and training of the respondents in BCI studies, more cognitive brain functions currently was quite variable, with a strong representation of techni- receive a considerable amount of attention. Almost half cal disciplines and neurosciences. This corresponds with of our respondents indicated they investigate attention or a previous survey among BCI researchers about ethical visual perception for decoding. issues [9] and with the multidisciplinary character of the Another conclusion drawn from the characterization BCI research and development process, in which in-depth section of the survey is that the field remains dominated knowledge about brain function needs to be combined by non-invasive approaches. This may not be surprising with advanced mathematical and engineering solutions, considering the practical difficulties of implanted BCI in order to develop products that can be used in the daily research, such as the limited number of available (human) life of patients or healthy subjects. Almost 10% of BCI subjects and access to the required medical context. In researchers had a medical training, which may be indica- this respect, the large difference between continents in tive of a promising interest from the treatment and reha- the ratio of implanted/non-invasive researchers is strik- bilitation professions. ing. Whereas the percentage of respondents working Although many studies use metabolic signals for BCI on implanted BCIs is 8% (13/167) within Europe, this research (e.g. fNIRS, fMRI), the most prevailing signal- number is three times as high among North American acquisition methods in the field involve electrical signals. respondents (26%, 18/68). This discrepancy is well- In fact, more than 95% of respondents indicated that they known in the BCI field [14], and may be attributed use EEG, MEG, or implanted electrodes as their main sig- to a different perception of the risk of implants, or to nal-acquisition technique. Also a recent survey among different regulations for implanted human and animal BCI game developers, researchers, and potential users studies. Lack of interest of the target population is not indicated that EEG is by far the most oen u ft sed non-inva - likely to be associated with these numbers, since studies sive acquisition technique [10]. The poor time resolution in both Europe and North America have indicated that 244 M . J. VANSTEENSEL ET AL. the majority of paralyzed patients have a positive attitude researchers and less well-funded by funding agencies, and towards implanted BCIs [15–17], especially if such a sys- may therefore have generated fewer descriptions. tem were wireless [18]. er Th e was a remarkable consistency in the reported Overall, the relative proportion of implanted-BCI least and most important bottlenecks identified for researchers seems to be decreasing over the last 10 years, the non-invasive applications. Across all six scenarios, despite substantial growth of the BCI field as a whole: more respondents indicated that safety issues are not a bottle- than 70% of all respondents started within the last 10 years, neck, indicating that the risks of non-invasive BCI sys- but for the implanted BCI researchers this number is only tems are considered negligible and that there is sufficient slightly higher than 30%. Whether these numbers suggest evidence for this. Notably, this result corresponds with a decreasing interest of researchers or funding agencies in the conclusions on the safety of non-invasive BCIs in implanted BCIs, or whether practicalities, regulations, or the Asilomar researchers’ survey on ethical issues [9]. A other issues stand in the way, remains to be determined. significant hurdle for non-invasive BCI tools, however, One possible explanation is that with an increasing inter- is long-term system performance, which is considered est in BCI (partly due to the significant BCI funding by the insufficient for most applications described. In fact, for EU until 2014), combined with ao ff rdable measurement 209 of the 273 submitted non-invasive applications, the and computing tools (notably EEG and PCs), it is more respondent agreed or strongly agreed with the statement feasible for new researchers to get engaged in non-invasive (Long-term) system performance is not yet good enough, than in implanted BCI research. With the recent surge in whereas only 25 disagreed or strongly disagreed. Related, funding for implanted BCI technology in North America respondents identified a clear need for improved sensors (DARPA, Brain Initiative), we may well see a shift towards and signal-processing techniques to improve system per- this field in the coming years. formance. Interestingly, this finding confirms the results Interestingly, the percentages of implanted and from a recent survey among BCI game researchers [10], non-invasive BCI researchers (11% vs 89%) within the and extends it to other non-invasive BCI applications. current sample corresponded largely with the percent- Other generally applicable issues that reportedly need to ages of implanted and non-invasive BCI applications be addressed are the complexity of the systems, which that were suggested in section B (14% vs 86%). Indeed, currently renders them not usable in the home environ- more than 90% of the applications described in section B ment of end-users, and the limited knowledge of end-us- used the same approach (implanted/non-invasive) as the ers about BCI applications, together with the insufficient researcher used for his/her own research. This suggests incorporation of their needs and wishes in the developed that the opinion of respondents about the bottlenecks and systems. requirements for future research for this application is Addressing the increasing recognition of the need for likely to be based on actual expertise and knowledge about a user-centered design [19], the bottlenecks and research these issues, and supports the validity of the results of this topics identified by researchers should be compared questionnaire. with the opinions of users. There are a number of studies describing the view on BCIs expressed by people who have tried using a BCI system in a research setting, or who are 4.1. Near future currently using such a system at home, at work, or in the When asked to describe a BCI application that may be clinic. To our knowledge these studies only involved the feasible within the next 5–10  years, many researchers use of non-invasive BCIs by people with ALS or disabilities (about one in three) described an application to replace with other etiologies. Interestingly, the topics that can be lost CNS function, such as BCI for communication with extracted from these reports correspond largely with the locked-in patients. Also applications aimed to improve issues identified in the current study, in that performance CNS function, for instance BCI for rehabilitation after (accuracy, speed), sensor issues (discomfort and cumber- stroke, were described oen (a ft bout one-quarter of the someness of using wet electrodes), and system complex- responses within this section). Together, the replace and ity (for both user and caregiver) are recurring themes improve scenario applications covered almost 60% of [20–24]. A significant discrepancy between these studies all descriptions submitted. One may conclude that BCI and the current report, however, is the importance attrib- researchers view these kinds of applications as the most uted to esthetics and user stigmatization. Although in the promising within the near future. It cannot be excluded, current study there was some agreement for the enhance, however, that many researchers are most familiar with supplement and restore scenarios, with the statement The applications that are currently eligible for funding, and currently available hardware is cosmetically unappealing, that the other four scenarios, which represent relatively this was not the case for replace, improve and research tool. ‘new’ BCI directions, may be less well-known among BCI For the statement The end -user image is stigmatized by BRAIN-COMPUTER INTERFACES 245 wearing a BCI system, responses for all scenarios were of statements, respondents indicated that systems have between 2.5 and 3.5, reflecting no dominant agreement to become easy-to-use, wearable, and durable, and that or disagreement with the statement. Overall, it can be it should be possible to use them in any environment. inferred from these results that BCI researchers consider Although about two out of three of these statements came the esthetics/stigmatization aspects of currently available from the responses to the non-invasive out-of-the-box systems acceptable, at least for a large part of the applica- question, it is unlikely that user friendliness is considered tions. This, however, does not correspond with reports in better for existing implanted BCI approaches, since only which users are asked for their opinion on this topic. In very few people have so far been implanted with electrodes fact, the physical appearance of BCI systems is considered purely for BCI purposes. Instead, BCI researchers may a relevant issue by most users, and is not appropriately consider the stage of implanted BCI research too prema- addressed by current (EEG-based) systems [20,21,23,25]. ture to even consider user friendliness. Indeed, implanted e l Th argest number of descriptions for implanted appli- far future statements most oen co ft ntained referrals to cations were focused on replacing lost CNS function. Over ‘communication and environmental control for patients’ 20% of the described BCI tools within the replace sce- and ‘prosthesis and artificial limbs for patients’ (together nario were based on implants, suggesting that a signif- in >40% of implanted far future statements), suggesting icant number of scientists consider implanted solutions that implanted BCI research needs a breakthrough in for replacement of lost brain function as feasible within applying neuroscientific knowledge into actual applica- the coming 5–10 years. Moreover, the implanted approach tions to replace lost brain function. is clearly marked as relevant for this type of application, e Th second most oen ft mentioned topic in the non-in - in that respondents disagree with the statements The vasive far-future statements was ‘hardware: sensors’ (n = advantages over existing non-BCI solutions are too small 45, >25% of non-invasive far future statements). Many and It is unclear what the advantages of the BCI system of these statements indicated a need for dry electrodes are, compared to existing non-BCI solutions. Performance or headsets (Dry electrodes that actually work well… or (long-term) and durability were identified as important Acquisition with unobtrusive dry-electrode headsets and technical bottlenecks. Much research is still needed (all high signal quality and stability) and for easier-to-use sen- possible research directions are considered relevant) to sors (Wireless, thin sensors that can be placed on the head overcome the current hurdles and bring implanted replace without preparation time or e Th only possible breakthrough solutions to the market, which may be considered a log- for non-invasive BCI is the possibility to have sensors that ical consequence of the early stage in which this type of could be used so easy to use that you could pair them to your research finds itself. mobile phone). In fact, 19 of 45 non-invasive statements Interestingly, the statement Potential users do not that were tagged with ‘hardware: sensors’ were also tagged know about this BCI tool was agreed upon for almost all with ‘user friendliness’, which may suggest that the user implanted and non-invasive application scenarios in the friendliness of non-invasive sensors is an important niche current study. Studies involving both healthy and disabled for future development and eventual user acceptance. users, however, have found that 50–80% of the interviewed Although most implanted statements were patient-ori- potential users were aware of BCIs before they completed ented and most non-invasive statements were oriented the respective surveys [10,16,17]. Although it is conceiva- towards healthy subjects, it was interesting to note that ble that participants (i.e. potential users) of these previous people also foresee a role of implanted systems for healthy surveys on BCIs are more likely to participate if they have end-users, indicating that implanted BCI applications heard of BCIs before (potential survey recruitment bias), may, in the long run, find a niche outside the patient-ori- it is also possible that the respondents of the current study ented ‘replace’ scenario. Implanted applications for healthy (i.e. BCI researchers) underestimate the awareness among end-users were oen r ft elated to futuristic scenarios, such potential users. as brain-to-brain communication or enhancing brain capacity, but also approaches to enhance daily functioning or to monitor health were mentioned (Direct behavioral 4.2. Far future modic fi ation. Using implants to enhance or suppress desira - Since the list of topics used for the labeling of the Far ble or undesirable habits or cravings or e d Th evelopment of future statements was derived from these statements technologies that can monitor and care for the user beyond themselves, the list can be considered as an exhaustive the perception of it). overview of research areas that have not been addressed Besides being an overview of topics where break- adequately in the past decades of BCI research, and where throughs are needed for currently known BCI approaches major breakthroughs are needed. The topic of ‘user friend - to be actually used by the target populations, the list of far liness’ was addressed the most frequently: in over 20% future statements also contained many descriptions that 246 M . J. VANSTEENSEL ET AL. go far beyond that. A selection of these truly out-of-the system performance, and user friendliness is crucial for box statements, given in Table 3, may act as an inspiration non-invasive BCIs, as well as an increased incorpora- for current and future BCI researchers. In addition, these tion of the wishes and needs of end-users. For implanted statements may serve the ethical debate that will necessar- BCI applications, steps have to be taken to develop com- ily accompany further BCI development, in defining the pletely implantable systems with adequate performance currently foreseeable concepts and technology. and durability, and to make these systems available for patients. Moreover, the results warrant a recommendation for the field to move towards clinical trials to demonstrate 4.3. Limitations that implanted BCIs can be safe and durable, and yield With two rounds of the questionnaire, 298 respond- good performance. Finally, the field stands to benefit from ents completed the form, which is 9.1% of the 3291 BCI identifying the benefits of implanted over non-invasive researchers who were invited to do so. Notably, this solutions, and vice versa, recognizing the individual users’ percentage may be an underestimation, since a subset needs and circumstances. of the contacts may not be active in BCI research any more. In fact, in the second round of invitations to fill Acknowledgements out the questionnaire, we added the option to select I am e a Th uthors would like to thank the people from the BNCI Ho- no longer involved in BCI research, without further filling rizon 2020 consortium for their valuable contribution during out the questionnaire. Upon this second invitation, 21 the development of the questionnaire. people selected this option, and 69 people did complete the questionnaire (the other 229 responses were obtained in the first round). This suggests that possibly more than Disclosure statement 20% of the people addressed were no longer working in e a Th uthors have no conflicts of interest to disclose. BCI research. Nevertheless, the incomplete response rate is an obvious limitation of the current study. Other recent questionnaires with email invitations among biomedical Funding researchers have obtained response rates between 16 and This work was supported by the FP7 project ‘e Th Future of 53% [26–28]. Reasons for the relatively low response Brain/Neural Computer Interaction: Horizon 2020’ under rate may be the length of the questionnaire, as well as Grant 609593; ERC Advanced Grant ‘Intracranial Connection its complexity (i.e. some open questions, describing with Neural Networks for Enabling Communication in Total paralysis’ under Grant 320708; and STW project ‘Pilot study of ideas etc.). Another limitation is that we potentially have an implantable brain-computer interface system with people a bias towards European respondents: more than 50% with locked-in syndrome’ under Grant STW12803. of the responses came from people working in Europe. Considering the context of the questionnaire, being part of the BNCI Horizon 2020 roadmap activities, this large per- ORCID centage of European respondents may not be surprising. M. J. Vansteensel   http://orcid.org/0000-0002-9252-5116 e Th y may have more incentive to give their opinion about E. J. Aarnoutse   http://orcid.org/0000-0001-7648-250X where the BCI field should head in the future. Despite N. F. Ramsey   http://orcid.org/0000-0002-7136-259X these constraints, several of our findings correspond with previous reports (see above) and give us confidence that References the outcome of this questionnaire and the topics that are identified reasonably reflect the BCI research field. [1] Fetz EE. Operant conditioning of cortical unit activity. Science. 1969;63:955–958. [2] Vidal JJ. Toward direct brain-computer communication. 5. Conclusions Annu Rev Biophys Bioeng. 1973;2:157–180. [3] Wolpaw JR, Winter Wolpaw E. Brain–computer From the current study, we conclude that BCI research- interfaces. principles and practice. New York (NY): ers are quite optimistic about the feasibility of BCI appli- Oxford University Press; 2012. cations for both patients and healthy end-users. They [4] Brunner C, Birbaumer N, Blankertz B, et al. BNCI Horizon described many types of applications that would be via- 2020: towards a roadmap for the BCI community. Brain- Computer Interfaces. 2015;2:1–10. ble in the near and far future. Especially applications for [5] Nicolas-Alonso LF, Gomez-Gil J. Brain computer patients, to replace or improve lost brain function, are interfaces, a review. Sensors. 2012;12:1211–1279. considered promising by many BCI researchers. However, [6] BNCI Horizon2020 Consortium. Roadmap. the future of before these products can penetrate the market, a num- brain/neural computer interaction: Horizon2020. Graz: ber of issues have to be addressed. Improving sensors, University of Technology; 2015. BRAIN-COMPUTER INTERFACES 247 [7] Elo S, Kyngäs H. The qualitative content analysis process. [19] K übler A, Holz EM, Riccio A, et al. The user-centered J Adv Nurs. 2008;62:107–115. design as novel perspective for evaluating the usability [8] Hsieh HF, Shannon SE. Three approaches to qualitative of BCI-controlled applications. PLoS One. 2014;9: content analysis. Qual Health Res. 2005;15:1277–1288. e112392. [9] Nijboer F, Clausen J, Allison BZ, et al. The Asilomar [20] Blain-Moraes S, Schaff R, Gruis KL, et al. Barriers to and survey: Stakeholders’ opinions on ethical issues related to mediators of brain-computer interface user acceptance: brain-computer interfacing. Neuroethics. 2013;6:541–578. user group findings. Ergonomics. 2012;55:516–525. [10] Ahn M, Lee M, Choi J, et al. A review of brain-computer [21] Holz EM, Höhne J, Staiger-Sälzer P, et al. Brain-computer interface games and an opinion survey from researchers, interface controlled gaming: evaluation of usability by developers and users. Sensors. 2014;14:14601–14633. severely motor restricted end-users. Artif Intell Med. [11] S itaram R, Lee, S, Birbaumer N. BCIs that use brain 2013;59:111–120. metabolic signals. In: Wolpaw JR, Winter Wolpaw E, [22] Leeb R, Perdikis S, Tonin L, et al. Transferring brain- editors. Brain-computer interfaces. principles and practice. computer interfaces beyond the laboratory: successful New York (NY): Oxford University Press; 2012. p. 301–314. application control for motor-disabled users. Artif Intell [12] Vansteensel MJ, Hermes D, Aarnoutse EJ, et al. Brain- Med. 2013;59:121–132. computer interfacing based on cognitive control. Ann [23] Zickler C, Halder S, Kleih SC, et al. Brain Painting: Neurol. 2010;67:809–816. usability testing according to the user-centered design in [13] Bleichner MG, Jansma JM, Salari E, et al. Classification end users with severe motor paralysis. Artif Intell Med. of mouth movements using 7T fMRI. J Neural Eng. 2013;59:99–110. 2015;12:066026. [24] Peters B, Bieker G, Heckman SM, et al. Brain-computer [14] Berger TW, Chapin JK, Gerhardt GA, et al. International interface users speak up: e Th virtual users’ forum at the assessment of research and development in brain- 2013 international brain-computer interface meeting. computer interfaces. Maryland: World Technology Arch Phys Med Rehabil. 2015;96:S33–S37. Evaluation Center, Inc; 2007. [25] Nijboer F. Technology transfer of brain-computer [15] Huggins JE, Wren PA, Gruis KL. What would brain- interfaces as assistive technology: barriers and computer interface users want? Opinions and priorities opportunities. Ann Phys Rehabil Med. 2015;58:35–38. of potential users with amyotrophic lateral sclerosis. [26] Oushy MH, Palacios R, Holden A, et al. To share or Amyotroph Lateral Scler. 2011;12:318–324. not to share? A survey of biomedical researchers in the [16] Collinger JL, Boninger ML, Bruns TM, et al. Functional U.S. Southwest, an ethnically diverse region. PLoS One. priorities, assistive technology, and brain-computer 2015;10:e0138239. interfaces aer s ft pinal cord injury. J Rehabil Res Dev. [27] Teunis T, Nota SP, Schwab JH. Do corresponding authors 2013;50:145–160. take responsibility for their work? A covert survey. Clin [17] Lahr J, Schwartz C, Heimbach B, et al. Invasive brain- Orthop Relat Res. 2015;473:729–735. machine interfaces: a survey of paralyzed patients’ [28] Al-Herz W, Haider H, Al-Bahhar M, et al. Honorary attitudes, knowledge and methods of information authorship in biomedical journals: how common is it and retrieval. J Neural Eng. 2015;12:043001. why does it exist? J Med Ethics. 2014;40:346–348. [18] Blabe CH, Gilja V, Chestek CA, et al. Assessment of brain-machine interfaces from the perspective of people with paralysis. J Neural Eng. 2015;12:043002. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Brain-Computer Interfaces Taylor & Francis

The brain-computer interface researcher’s questionnaire: from research to application

The brain-computer interface researcher’s questionnaire: from research to application

Abstract

AbstractOver the last decades, the brain-computer interface (BCI) research field has seen steep progress and now recognizes several types of applications, each with its own target population. Despite this evolution, BCIs seem to remain a research application, and are hardly used in daily life at home, work or in the clinic. Therefore, as a field, we need to take the next step and make sure that we develop BCIs that can, and will, be used outside the laboratory. To this purpose, BCI...
Loading next page...
 
/lp/taylor-francis/the-brain-computer-interface-researcher-s-questionnaire-from-research-mVDX01rfME
Publisher
Taylor & Francis
Copyright
© 2017 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
ISSN
2326-2621
eISSN
2326-263x
DOI
10.1080/2326263X.2017.1366237
Publisher site
See Article on Publisher Site

Abstract

Brain-Computer interfa Ces, 2017 VoL. 4, no . 4, 236– 247 https://doi.org/10.1080/2326263X.2017.1366237 OPEN ACCESS The brain-computer interface researcher’s questionnaire: from research to application M. J. Vansteensel  , G. Kristo, E. J. Aarnoutse  and N. F. Ramsey  Department of neurosurgery, Brain Center rudolf m agnus, university m edical Center utrecht, utrecht, t he netherlands ABSTRACT ARTICLE HISTORY received 3 f ebruary 2017 Over the last decades, the brain-computer interface (BCI) research field has seen steep progress a ccepted 7 July 2017 and now recognizes several types of applications, each with its own target population. Despite this evolution, BCIs seem to remain a research application, and are hardly used in daily life at home, work KEYWORDS or in the clinic. Therefore, as a field, we need to take the next step and make sure that we develop Brain-computer interface; BCIs that can, and will, be used outside the laboratory. To this purpose, BCI researchers worldwide survey; questionnaire; were approached by email with the request to fill out a questionnaire, aimed at laying out the near research; future and far future of the BCI field. We show that researchers are quite optimistic about the feasibility of having BCIs applied in real life. Also, we identify the hurdles that need to be taken before these applications become commercial products and the research activities required to accomplish this. 1. Introduction output that is lost due to injury or disease. A BCI may also be used to control an electrical stimulation device Since the pioneering work of Fetz [1], Vidal [2] and others, that targets dysfunctional muscles or nerves and thereby the field of brain-computer interface (BCI) research has restores their function, or to improve limb control in reha- seen steep progress, as evidenced by the steady increase bilitation settings or the symptoms of neurological and in the number of peer-reviewed BCI articles over the last psychiatric problems using neurofeedback approaches. 15 to 20 years [3]. Over time, numerous studies have suc- Other BCI application scenarios mainly target healthy cessfully or less successfully attempted to maximize both people, and may enhance their day-to-day functioning, performance and multi-dimensionality of BCIs, by opti- for example using BCI-based detection of attention lapses mizing brain signal acquisition methods, signal processing during prolonged driving, or supplement people’s physical and decoding techniques as well as feedback design, and abilities with an extra actuator such as a BCI-controlled by evaluating different concepts or strategies for control. robotic third arm. BCIs can also be used as a research tool At the same time, the definition of a BCI application as to study brain function, such as the neuronal processes of well as that of the target population for BCIs have evolved learning or decision-making. substantially. Whereas early studies were mainly oriented Brain signals for BCI can be recorded non-invasively towards enabling brain-based communication for patients with sensors outside the head (non-invasive BCIs), or suffering from severe paralysis or the locked-in syndrome with implanted electrodes (implanted BCIs). See for (LIS), the BCI field currently recognizes several types of review, [5]. The most well-known non-invasive record- application scenarios for BCIs (i.e. types of devices that ing technique is electroencephalography (EEG), which a BCI can control), each with their own range of target records neuroelectrical activity with electrodes placed populations. A subdivision of five application scenarios on the scalp. Other non-invasive recording techniques was originally proposed by Wolpaw and Wolpaw in 2012 include functional MRI (fMRI) and functional near-in- [3], and was extended with a sixth application scenario frared spectroscopy (fNIRS), both of which measure the (research tool, see below) by the BNCI Horizon 2020 project [4]. In the classic application of a communication hemodynamic responses that occur in active brain areas, tool for locked-in patients, or for example in the case of and magnetoencephalography (MEG). Whereas fMRI BCI control of a wheelchair, the BCI replaces natural CNS and MEG involve large equipment, and are therefore CONTACT m. J. Vansteensel m.j.vansteensel@umcutrecht.nl t he supplemental material for this paper is available online at https://doi.org/10.1080/2326263X.2017.1366237. © 2017 t he a uthor(s). published by informa uK Limited, trading as t aylor & f rancis Group. t his is an open a ccess article distributed under the terms of the Creative Commons a ttribution-nonCommercial-noDerivatives License (http://creativecommons.org/licenses/by-nc- nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way. BRAIN-COMPUTER INTERFACES 237 mainly used as a research tool, fNIRS has the potential to invitation to fill out the questionnaire. This second round be used for BCI applications with end-users. Implanted was closed at the beginning of January 2015. electrodes for BCI may be placed epidurally or subdur- ally (i.e. electrocorticography, ECoG) or may penetrate 2.2. Structure of the BCI researchers’ questionnaire the cortical surface (multi-electrode arrays, MEA). Both e Th BCI researchers’ questionnaire consisted of three sec - recording techniques require surgery to precisely place the tions (see also the Supplemental Material). electrodes, but have the advantage of high temporal and spatial resolution and potential 24/7 availability, making 2.2.1. Section A (Respondents) them interesting candidates for certain BCI application In section A, respondents were asked to answer a list scenarios. of questions, which was set up to characterize each Despite the research progress that we have witnessed researcher by several criteria, such as their background over the past years, BCIs are only sparsely available for and what type of BCIs they work on. Questions within the large range of diseased and healthy target populations this section were multiple-choice questions, and allowed that may be interested in using them in home, work or respondents to select one or more items (depending on clinical settings, and only a very limited number of people the question). Where applicable, one of the choices was use BCI tools outside research settings. In order to make ‘Other’, which was followed by a free-text field, allowing sure that potential end-users, in time, will start to benefit people to describe their answer in words. from the extensive knowledge gained in BCI research, it is important to lay out the future of the BCI field, identi- 2.2.2. Section B (Near future) fying the most promising BCI niches and signaling topics In section B, we asked participants to shortly describe in that need more attention. From November 2013 to May one sentence a potential BCI application that could or 2015, the BNCI Horizon 2020 project, funded within should be feasible within the next 5–10 years (e.g. EEG- the European Commission’s Framework Programme 7, BCI for motor rehabilitation after stroke ), and to assign this worked on the development of a roadmap for the BCI application to one of the six scenarios (replace, restore, field, laying out the near and far future of BCIs [6 ]. One enhance, improve, supplement, and research tool), by of the efforts of the consortium to reach this goal was to selecting the respective item in a multiple-choice list. Also, acquire insight into the opinion of BCI researchers about participants were asked to indicate whether they would the current status and the future of their field. To this want to see this application developed using an implanted purpose, BCI researchers worldwide were approached or a non-invasive approach (multiple choice). Next, par- by email with the request to fill out a questionnaire, ticipants were asked to keep their specific application in asking them for their view on the most promising BCI mind, and indicate to what extent each of a given list of applications, the hurdles that need to be taken before bottlenecks would apply to their application. Each item these applications become commercial products and the on the list (i.e. each possible bottleneck, such as Potential research activities that are needed to accomplish this. A users do not know about this BCI tool) could be rated on a concise summary of the results of this questionnaire was 5-point scale (strongly agree, agree, neutral, disagree, and described in one of the Appendices of the BNCI Horizon strongly disagree), by selecting the associated box. The 2020 roadmap. In the current manuscript, the result of sixth possible answer was ‘not applicable’. A separate list of the research is presented in more detail and we quantified items was used to ask participants what, in their opinion, the qualitative answers respondents gave about the far should be the focus of BCI research in the next 5–10 years future of BCI research. to reach the envisioned application. Here, each item on the list was a possible research direction (e.g. Developing and 2. Methods testing advanced signal-processing techniques for improved system performance), and rating options were the same as 2.1. General for the bottlenecks. e q Th uestionnaire was sent around to 3291 BCI researchers in May of 2014. Each researcher received an email describ- 2.2.3. Section C (Far future) ing the purpose of the questionnaire and containing a link In section C, participants were asked to think out of the to fill it out on a website (Google Form). Email addresses box and into the far future and write in a free-text field a were obtained from several BCI research groups and soci- short statement or description of a potential ‘killer’ appli- eties. Two reminders were sent until the questionnaire was cation or major research breakthrough. This was done for closed on 10 July 2014. Early December 2014, researchers both the implanted and the non-invasive BCI approaches, who did not respond in the first round received a second resulting in two Far future statements per respondent. 238 M . J. VANSTEENSEL ET AL. 2.3. Data analysis to an overall disagreement or strong disagreement with the statement. Statements that received a COM value 2.3.1. Section A (Respondents) between 2.5 and 3.5 can be considered as neither particu- e a Th nswers respondents gave in section A (Respondents) larly relevant nor irrelevant, according to the respond- served to characterize background, expertise, and working ents of our questionnaire. area of the participants. Data are presented as percent- ages in the results section. For many questions, multiple 2.3.3. Section C (Far future) answers were allowed, resulting in more entries than the Responses in the Far future section were analyzed using total number of respondents. To make sure that this fact ‘content analysis’, a highly suitable approach to analyze is represented accurately in the results, percentages were text data [7,8]. The technique has a long history espe- computed relative to the total number of respondents, and cially within the social sciences and health studies. We the total per question could therefore exceed 100%, unless used a conventional or inductive approach, where codes stated otherwise. are directly derived from the text. More specifically, responses of the Far future section of the questionnaire 2.3.2. Section B (Near future) were checked for completeness and clarity. Incomplete For section B (Near future), our aim was to extract from or absent responses in this section, as well as unclear the questionnaire the most important hurdles in the devel- statements or descriptions, were excluded from analy- opment of BCI applications in each of the six scenarios, as sis. Subsequently, in a data-driven approach, the content well as the most urgent research directions to overcome of the statements given by the respondents was used to these. As the first step of the analysis, two of the authors assemble a list of numbered topics (codes) that, together, (MV and GK) evaluated the clarity of the descriptions of covered all issues described by the respondents. Where the BCI applications that were given by the respondents, relevant, codes were grouped into larger categories. Next, as well as the correctness of the assignment to each to the in an iterative process, all statements were annotated with scenarios. Unclear descriptions were excluded from fur- one or more codes from that list by two raters (MV and ther analysis. Descriptions that were clear but incorrectly EA). In three rounds of updating codes and a fourth round assigned to a certain scenario were re-assigned. Where the of discussion, the raters reached agreement about code two raters disagreed, they discussed the application until assignments. For each code, the numbers of instances in they reached consensus. the list of implanted and non-invasive responses were For subsequent analysis, we considered all combina- counted. tions of approach (implanted, non-invasive) and appli- cation scenario (replace, restore, enhance, improve, 3. Results supplement, and research tool), and refer to these combi- nations as ‘scenario-approach combinations’ (e.g. implant- 3.1. Respondents ed-replace or non-invasive-supplement). We counted the Two rounds of the questionnaire resulted in 298 (9.1%) number of responses assigned to each scenario-approach responses. Most respondents (56%) worked in Europe, but combination, and only included in the analysis the sce- we also received a substantial number of responses from nario-approach combinations that received at least 10 people working in North America and Asia (Figure 1(A)). responses. e m Th ajority of BCI researchers had some form of back- To quantify the importance of the each of the bot- ground or training in the field of engineering (62%). tlenecks and research directions for the scenario-ap- Computer science and neuroscience came second and proach combinations, we first labeled each rating with a third with 37% and 29%, respectively (Figure 1(B)). certain weight: ‘not applicable’ with 0, ‘totally disagree’ Notably, many (46%) respondents reported two or more with 1, ‘disagree’ with 2, ‘neutral’ with 3, ‘agree’ with 4, training backgrounds. Respondents covered a wide range and ‘totally agree’ with 5. Subsequently, the center of of positions in their institute, such as post-doc, PhD stu- mass (COM) of the given answers was computed, per dent or head of research and R&D departments. bottleneck/research direction statement, by summing Brain functions used for decoding were mainly motor the product of the number of subjects who assigned a function (74%), but other functions, such as attention and certain rating to a statement and the weight given to visual perception, were also addressed by many respond- that rating. This sum was divided by the total number ents (47% and 45%, respectively, Figure 1(C)). Signals of respondents within that scenario-approach combina- were mainly decoded from healthy subjects (85%) and tion. A COM value of 3.5 or higher meant that most patients (48%), and only a few people worked with ani- respondents agreed or strongly agreed with the state- mals (7%, Figure 1(D)). ment, whereas a COM value of 2.5 or lower was related BRAIN-COMPUTER INTERFACES 239 Figure 1.  respondents’ characteristics. ( a ) Continent where respondents perform BCi research, (B) background or training of the respondents, (C) brain functions respondents use for decoding, (D) type of subjects respondents use in BCi studies. A large majority of the respondents (n = 264, 89%) and 2005, and a decrease in the number of people with predominantly used non-invasive BCI systems, almost all a more recent starting year (i.e. within the last 10 years). based on EEG (95% of researchers working on non-inva- er Th e was an interesting distribution over continents sive BCIs). Gel electrodes and electrodes in a cap were the of respondents working on implanted and non-invasive most popular, but most respondents reported the use of BCIs: only 8% (13/167) of respondents from Europe and other approaches as well (water-based or dry electrodes, 4% (2/57) of respondents from Asia worked on implanted separate electrodes or electrodes in a headset). The distri - BCIs, whereas this percentage was 26% (18/68) for bution of active versus passive EEG electrodes was almost respondents of North America and 18% (2/11) for South equal: 88 respondents only used active electrodes, 70 only America (Figure 2(B)). passive and 94 used both. e r Th espondents who used implanted BCI systems (n = 3.2. Near future 34, 11%) recorded signals mostly with surface electrodes (62% of researchers working on implanted BCIs) that were e 298 r Th espondents of the questionnaire together sub- placed subdurally. Respondents using penetrating elec- mitted 363 BCI applications: 233 people submitted one trodes mostly recorded from cortical structures. A few application, and 65 submitted two applications. The respondents (also) reported recording from subcortical descriptions were unclear in 46 cases and these responses structures. were excluded from analysis, leaving 317 usable descrip- Overall, more than 70% of the respondents entered the tions of BCI applications. In 69 cases, we reassigned the field less than 10 years ago. This was true for the entire BCI application to a scenario different from the one to population of respondents, as well as for the respondents which it was originally assigned by the respondent. working on non-invasive BCIs (Figure 2(A)). The popula - When asked which BCI devices could be feasible within tion of respondents working on implanted BCIs, however, the next 5 to 10 years, most respondents (33%) chose devices showed a very clear peak in starting year between 2001 that replace natural CNS output. Other respondents described 240 M . J. VANSTEENSEL ET AL. Figure 2. implanted versus non-invasive. (a ) Distribution of the years in which respondents started working in BCi research. Gray bars indicate all respondents together. Green and red lines indicate the starting year of researchers working mainly on non-invasive and implanted BCis, respectively. (B) percentage of respondents in implanted (red) and non-invasive (green) BCi research, per continent. Table 1. number of respondents describing an implanted and a non-invasive solution for an application within one of the six scenarios in section B of the questionnaire. Replace Restore Improve Enhance Research Supplement Total implanted 21 6 9 3 4 1 44 (14%) non-invasive 83 11 75 47 19 38 273 (86%) t otal 104 (33%) 17 (5%) 84 (26%) 50 (16%) 23 (7%) 39 (12%) 317 BCI applications that may improve (26%), enhance (16%), 3.2.1. Bottlenecks supplement (12%), and restore (5%) natural CNS output or Interestingly, the ratings given by the respondents to that can be used as a research tool (7%). Most respondents the bottleneck statements indicate that several issues described their feasible application as being developed by apply to all or almost all of the six non-invasive scenar- using a non-invasive BCI system (86%, see Table 1). ios (Figure 3). In fact, participants did not agree with the Most of the replace applications that were mentioned statements that the long-term risks are too high (COM would be used for communication in general and for com- value  <  2.5 for six out of six non-invasive scenarios) munication in locked-in patients in particular (BCI for and that there is insufficient evidence for user safety communication in locked-in patients). Other typical exam- (6/6), indicating that the long-term risks of non-inva- ples of feasible BCI applications that were suggested aimed sive applications are considered acceptable and there to enhance cognitive functions (BCI as an improved tool to is sufficient evidence for this conclusion. Most people enhance daily brain activity, thus reducing the risks of los- agreed or strongly agreed, however, that (long-term) ing attention, memory and health), improve rehabilitation system performance of non-invasive BCIs is not yet (BCI to modulate plasticity in recovery from neural injury), good enough (COM value > 3.5 for 6/6 scenarios), that supplement during gaming or home automation (A sepa- potential users are unaware of the BCI tools (5/6), that rate control for routine tasks allowing users to direct extra current systems are too complicated for home use (5/6), attention to a productive use), restore lost movement or and that the wishes and needs of end- users are not met speech (BCI as a control signal for a FES neuroprosthesis), appropriately (4/6). or be used as a research tool for cognitive assessment or For the implanted replace scenario, participants believe technique-improvement purposes (BCI as a research tool that this approach has clear advantages over other non- to study decision-making, perception or learning). BCI tools and that there is sufficient evidence for this In the subsequent two sections, only groups of 10 or (Figure 3). They also agree that current system durability more scenario-approach applications are included in the and performance are insufficient and that there is insuf- analysis. Whereas all non-invasive scenarios could be ficient evidence for system performance, durability, and analyzed, only the replace scenario was included for the the risk/benefit ratio for end-users. The lack of aware- implanted approach, since only for this scenario a suf- ness about BCIs, which was a general bottleneck for all ficient number of respondents described an implanted non-invasive solutions, was also reported by participants approach (Table 1). addressing implanted replace solutions. BRAIN-COMPUTER INTERFACES 241 Figure 4. BCi research directions. t he opinion of the respondents of the questionnaire on the relevance of each of the given research directions for their near future applications. Colored horizontal bars indicate the center of mass of the responses, with 1 representing ‘totally disagree’ and 5 ‘totally agree’. t he gray shaded area indicates a center of mass between 2.5 and 3.5. research directions with bars that fall completely within this area can be considered as neither particularly relevant nor irrelevant. note that for the implanted approach, only the ‘replace’ scenario could be analyzed, because of lack of responses for the other scenarios. improve system performance. Also, the BCI community Figure 3. BCi bottlenecks. t he opinion of the respondents of the needs to investigate the wishes and requirements of the questionnaire on the relevance of each of the given bottlenecks end-users, and perform clinical trials on system durability, for their near future applications. Colored horizontal bars indicate performance, safety, efficacy (compared to non-invasive), the center of mass of the responses, with 1 representing ‘totally and risk-benefit ratio for end-users. disagree’ and 5 ‘totally agree’. t he gray shaded area indicates a center of mass between 2.5 and 3.5. Bottlenecks with bars that fall completely within this area can be considered as neither 3.3. Far future particularly relevant nor irrelevant. note that for the implanted approach, only the ‘replace’ scenario could be analyzed, because Over 60% of respondents described a far future scenario, of lack of responses for the other scenarios. defining what they considered a major breakthrough in the implanted or non-invasive BCI area, such as a killer 3.2.2. Research directions application or major research advance. Some answers When asked about the focus of research related to were considered unclear by the raters (n = 15 and n = 16 non-invasive BCI applications in the coming 5–10 years for implanted and non-invasive, respectively) and were (Figure 4), participants agreed on the need for the devel- removed from further analysis, resulting in 169 and 178 opment and testing of new sensors (6/6) and new signal- far future statements for implanted and non-invasive processing techniques for improving system performance approaches, respectively. These statements were used to (6/6). Participants also agree that there is a need for clini- define the list of topics (Table 2 ) and were labeled accord- cal trials to demonstrate system performance (5/6) and for ing to these topics. The topics were mostly not mutually identifying the wishes and needs of the end-users (6/6). exclusive, but refer to different aspects of BCIs (e.g. tar - For the research directions on implanted BCI appli- get group, hardware). Some topics are mutually exclu- cations, all statements received a score of 3.5 or higher. sive, but not completely (e.g. healthy subjects, patients). This means that new sensor and amplifier techniques Depending on the description, responses received one are needed, as well as signal-processing techniques that or more labels. 242 M . J. VANSTEENSEL ET AL. Statements most oen ft addressed ‘user friendliness’ environmental control for patients’ (n = 42 occurrences (n = 73 individual occurrences in the total of 347 implanted in 347 statements), ‘prostheses and artificial limbs for and non-invasive BCI statements), ‘communication and patients’ (n = 53), ‘hardware: sensors’ (n = 70), and ‘accu- racy and reliability of signal processing and decoding’ (n Table 2. f ar future topics identified from the statements given by = 52), indicating that these topics represent areas of par- the respondents in the f ar future section. ticular interest for the far future (Figure 5). Topic Subtopic Description erTh e were interesting differences between the 1 User friendliness, daily use, home use, portability implanted and non-invasive far future statements. First, 2 Critical statement about non-invasive or implanted the topics mentioned most oen f ft or these approaches BCIs 3 Ethics, public opinion, privacy differed, in that ‘user friendliness’ and ‘hardware: sen- 4 Commercialization sors’ were described most frequently in the non-inva- a number of end-users sive far future statements, whereas ‘communication and b n iche 5 Applications for patients environmental control for patients’ and ‘prosthesis and a psychiatry, behavior, neurofeedback artificial limbs for patients’ occurred most oen in t ft he b rehabilitation, functional electrical stimulation c epilepsy implanted far future statements (Figure 5). Second, within d1 Communication and environmental control the implanted far future statements, the ‘applications for d2 prosthesis, artificial limbs, exoskeletons, robotic arm e any/other patients’ topic (i.e. taking all of its subtopics together) was 6 Applications for healthy people referred to by 92 respondents, compared to ‘applications a Behavior, learning, stress, skills, neurofeedback b monitoring of health, alertness for healthy subjects’ by 55 respondents. These numbers c Gaming practically mirrored the situation for non-invasive far d military, security, forensic future statements (n = 51 for ‘applications for patients’, e enhanced/augmented reality f enhanced/extra actors and n = 85 for ‘applications for healthy subjects’). In con- g mood, environmental control, smart home, domotics trast, the topics ‘hardware’ (including all its subtopics) was h Car i any/other represented equally in the statements for implanted and 7 Brain research for non-invasive approaches (n = 94 in both cases). Also 8 Robotics, telepresence 9 Hardware the topic ‘signal processing/decoding’ (including all sub- a s ensors topics) was similarly represented in the responses for both b Wireless c Battery life implanted (n = 47) and non-invasive (n = 60) approaches. d s afety For both approaches, quite some truly out-of-the box e price, size, esthetics, comfort statements were given, a selection of which is indicated f signal-to-noise ratio g Brain stimulation, neuromodulation in Table 3. h Durability of system 10 Signal processing, decoding a a ccuracy, reliability, performance, multidimensionality 4. Discussion b optimal brain function for control c speed, information transfer d Calibration, coregistration In the current study, we describe the view of BCI research- e resolution ers about the future of their field, as obtained through 11 Hybrids 12 Relevant for both patients and healthy users the responses of the BNCI Horizon 2020 Researcher’s Figure 5.  f ar future topics. f or each of the topics identified from the f ar future statements (see t able 2), the number of statements referring to these topics is indicated on the y-axis. red indicates counts from the implanted far future statements, green from non- invasive far future statements. BRAIN-COMPUTER INTERFACES 243 Table 3. examples of out-of-the box f ar future statements. Non-invasive Far future statements a domotics device that you would put on when you arrive at your place, giving you full control of all your gadgets and appliances with minimal effort. t his implementation would also be useful for both healthy and disabled subjects BCi for enhanced creativity or logical thinking, e.g. by using a BCi as a feedback loop to help users reaching a mental state in which they are more creative or more logical Due to the common necessity of having to drive the brain activity via specific paradigms in order to get the brain signals into BCi mode (to infer the end user’s intention) compared to other instances when the user does not want to use the BCi, a major research advancement would be to find a brain signature that indisputably could be labeled as the mental switch to turn on a BCi, so it is not paradigm nor stimuli dependent a small signal reader (perhaps magnetic) that will allow to get a clear reading of the head electric potentials in terms of an aa C device, full integration into a device that can function using multiple input modalities. f or example, typing/cursor control, single switch, eye tracking, emG, eoG, and eeG. such a system would provide a solution that an end-user can learn to use while relatively normal function remains. as control becomes less reliable, the control modality can be changed to the most reliable method merging two neural networks for better decision-making Implanted Far future statements s afe wireless system that could be implemented “without” surgery like “nanobots” that would find their way to problematic regions of Cns to replace their functionality regenerating damaged brain areas by planting new brain cells, and train them with external stimulating impulses, to restore basic functionality such as mov - ing extremities in locked-in patients Chronic BCi usage for enhance scenario: answer suggestions pop up if a user raises a question, whatever they are Latest technological developments have reduced the hardware size to a headphone which can be worn easily by users at home. t he real potential application for BCis would be people communicating and controlling electronic equipment through thoughts alone. signals can be read by the home appliances and based on user moods and needs can alter the environment like if a user is sleepy lights can be automatically shut down without the user taking any action. if the end user is thirsty, cool water is served automatically. t his is only one huge potential the BCis could have in 20 years. t omorrow people might communi- cate on mobiles through thoughts alone rather than texts or sms Creating a high-resolution neural spike recording technology that does not require surgery a ccurate decoding from invasive brain signals of the markers of intention/volition/high level controls of the muscles, such high level inputs being used by lower systems (Central pattern Generators) to control the low-levels movements of the limbs. Current BCi do not work this way (they try to decode low-level control) but that’s not what the cortex does Questionnaire in 2014. As far as the authors are aware, of metabolic signals has been suggested as the main reason this is the first survey in which BCI researchers are specifi- that fNIRS and fMRI are not (yet) oen u ft sed for real-time cally asked about what they see as the most promising BCI applications [11], but especially fMRI has a significant applications and to identify the most important hurdles in value in earlier stages of BCI development, such as in getting these products available for end-users. finding and localizing optimal brain functions and areas In the first section of the survey, respondents were for control [12,13]. In this respect, it is interesting to note asked to characterize themselves and their current BCI that, although imagined, attempted, or executed move- work, in order to obtain a view of the state of affairs in BCI ment remain the brain functions that are most oen u ft sed research. The background and training of the respondents in BCI studies, more cognitive brain functions currently was quite variable, with a strong representation of techni- receive a considerable amount of attention. Almost half cal disciplines and neurosciences. This corresponds with of our respondents indicated they investigate attention or a previous survey among BCI researchers about ethical visual perception for decoding. issues [9] and with the multidisciplinary character of the Another conclusion drawn from the characterization BCI research and development process, in which in-depth section of the survey is that the field remains dominated knowledge about brain function needs to be combined by non-invasive approaches. This may not be surprising with advanced mathematical and engineering solutions, considering the practical difficulties of implanted BCI in order to develop products that can be used in the daily research, such as the limited number of available (human) life of patients or healthy subjects. Almost 10% of BCI subjects and access to the required medical context. In researchers had a medical training, which may be indica- this respect, the large difference between continents in tive of a promising interest from the treatment and reha- the ratio of implanted/non-invasive researchers is strik- bilitation professions. ing. Whereas the percentage of respondents working Although many studies use metabolic signals for BCI on implanted BCIs is 8% (13/167) within Europe, this research (e.g. fNIRS, fMRI), the most prevailing signal- number is three times as high among North American acquisition methods in the field involve electrical signals. respondents (26%, 18/68). This discrepancy is well- In fact, more than 95% of respondents indicated that they known in the BCI field [14], and may be attributed use EEG, MEG, or implanted electrodes as their main sig- to a different perception of the risk of implants, or to nal-acquisition technique. Also a recent survey among different regulations for implanted human and animal BCI game developers, researchers, and potential users studies. Lack of interest of the target population is not indicated that EEG is by far the most oen u ft sed non-inva - likely to be associated with these numbers, since studies sive acquisition technique [10]. The poor time resolution in both Europe and North America have indicated that 244 M . J. VANSTEENSEL ET AL. the majority of paralyzed patients have a positive attitude researchers and less well-funded by funding agencies, and towards implanted BCIs [15–17], especially if such a sys- may therefore have generated fewer descriptions. tem were wireless [18]. er Th e was a remarkable consistency in the reported Overall, the relative proportion of implanted-BCI least and most important bottlenecks identified for researchers seems to be decreasing over the last 10 years, the non-invasive applications. Across all six scenarios, despite substantial growth of the BCI field as a whole: more respondents indicated that safety issues are not a bottle- than 70% of all respondents started within the last 10 years, neck, indicating that the risks of non-invasive BCI sys- but for the implanted BCI researchers this number is only tems are considered negligible and that there is sufficient slightly higher than 30%. Whether these numbers suggest evidence for this. Notably, this result corresponds with a decreasing interest of researchers or funding agencies in the conclusions on the safety of non-invasive BCIs in implanted BCIs, or whether practicalities, regulations, or the Asilomar researchers’ survey on ethical issues [9]. A other issues stand in the way, remains to be determined. significant hurdle for non-invasive BCI tools, however, One possible explanation is that with an increasing inter- is long-term system performance, which is considered est in BCI (partly due to the significant BCI funding by the insufficient for most applications described. In fact, for EU until 2014), combined with ao ff rdable measurement 209 of the 273 submitted non-invasive applications, the and computing tools (notably EEG and PCs), it is more respondent agreed or strongly agreed with the statement feasible for new researchers to get engaged in non-invasive (Long-term) system performance is not yet good enough, than in implanted BCI research. With the recent surge in whereas only 25 disagreed or strongly disagreed. Related, funding for implanted BCI technology in North America respondents identified a clear need for improved sensors (DARPA, Brain Initiative), we may well see a shift towards and signal-processing techniques to improve system per- this field in the coming years. formance. Interestingly, this finding confirms the results Interestingly, the percentages of implanted and from a recent survey among BCI game researchers [10], non-invasive BCI researchers (11% vs 89%) within the and extends it to other non-invasive BCI applications. current sample corresponded largely with the percent- Other generally applicable issues that reportedly need to ages of implanted and non-invasive BCI applications be addressed are the complexity of the systems, which that were suggested in section B (14% vs 86%). Indeed, currently renders them not usable in the home environ- more than 90% of the applications described in section B ment of end-users, and the limited knowledge of end-us- used the same approach (implanted/non-invasive) as the ers about BCI applications, together with the insufficient researcher used for his/her own research. This suggests incorporation of their needs and wishes in the developed that the opinion of respondents about the bottlenecks and systems. requirements for future research for this application is Addressing the increasing recognition of the need for likely to be based on actual expertise and knowledge about a user-centered design [19], the bottlenecks and research these issues, and supports the validity of the results of this topics identified by researchers should be compared questionnaire. with the opinions of users. There are a number of studies describing the view on BCIs expressed by people who have tried using a BCI system in a research setting, or who are 4.1. Near future currently using such a system at home, at work, or in the When asked to describe a BCI application that may be clinic. To our knowledge these studies only involved the feasible within the next 5–10  years, many researchers use of non-invasive BCIs by people with ALS or disabilities (about one in three) described an application to replace with other etiologies. Interestingly, the topics that can be lost CNS function, such as BCI for communication with extracted from these reports correspond largely with the locked-in patients. Also applications aimed to improve issues identified in the current study, in that performance CNS function, for instance BCI for rehabilitation after (accuracy, speed), sensor issues (discomfort and cumber- stroke, were described oen (a ft bout one-quarter of the someness of using wet electrodes), and system complex- responses within this section). Together, the replace and ity (for both user and caregiver) are recurring themes improve scenario applications covered almost 60% of [20–24]. A significant discrepancy between these studies all descriptions submitted. One may conclude that BCI and the current report, however, is the importance attrib- researchers view these kinds of applications as the most uted to esthetics and user stigmatization. Although in the promising within the near future. It cannot be excluded, current study there was some agreement for the enhance, however, that many researchers are most familiar with supplement and restore scenarios, with the statement The applications that are currently eligible for funding, and currently available hardware is cosmetically unappealing, that the other four scenarios, which represent relatively this was not the case for replace, improve and research tool. ‘new’ BCI directions, may be less well-known among BCI For the statement The end -user image is stigmatized by BRAIN-COMPUTER INTERFACES 245 wearing a BCI system, responses for all scenarios were of statements, respondents indicated that systems have between 2.5 and 3.5, reflecting no dominant agreement to become easy-to-use, wearable, and durable, and that or disagreement with the statement. Overall, it can be it should be possible to use them in any environment. inferred from these results that BCI researchers consider Although about two out of three of these statements came the esthetics/stigmatization aspects of currently available from the responses to the non-invasive out-of-the-box systems acceptable, at least for a large part of the applica- question, it is unlikely that user friendliness is considered tions. This, however, does not correspond with reports in better for existing implanted BCI approaches, since only which users are asked for their opinion on this topic. In very few people have so far been implanted with electrodes fact, the physical appearance of BCI systems is considered purely for BCI purposes. Instead, BCI researchers may a relevant issue by most users, and is not appropriately consider the stage of implanted BCI research too prema- addressed by current (EEG-based) systems [20,21,23,25]. ture to even consider user friendliness. Indeed, implanted e l Th argest number of descriptions for implanted appli- far future statements most oen co ft ntained referrals to cations were focused on replacing lost CNS function. Over ‘communication and environmental control for patients’ 20% of the described BCI tools within the replace sce- and ‘prosthesis and artificial limbs for patients’ (together nario were based on implants, suggesting that a signif- in >40% of implanted far future statements), suggesting icant number of scientists consider implanted solutions that implanted BCI research needs a breakthrough in for replacement of lost brain function as feasible within applying neuroscientific knowledge into actual applica- the coming 5–10 years. Moreover, the implanted approach tions to replace lost brain function. is clearly marked as relevant for this type of application, e Th second most oen ft mentioned topic in the non-in - in that respondents disagree with the statements The vasive far-future statements was ‘hardware: sensors’ (n = advantages over existing non-BCI solutions are too small 45, >25% of non-invasive far future statements). Many and It is unclear what the advantages of the BCI system of these statements indicated a need for dry electrodes are, compared to existing non-BCI solutions. Performance or headsets (Dry electrodes that actually work well… or (long-term) and durability were identified as important Acquisition with unobtrusive dry-electrode headsets and technical bottlenecks. Much research is still needed (all high signal quality and stability) and for easier-to-use sen- possible research directions are considered relevant) to sors (Wireless, thin sensors that can be placed on the head overcome the current hurdles and bring implanted replace without preparation time or e Th only possible breakthrough solutions to the market, which may be considered a log- for non-invasive BCI is the possibility to have sensors that ical consequence of the early stage in which this type of could be used so easy to use that you could pair them to your research finds itself. mobile phone). In fact, 19 of 45 non-invasive statements Interestingly, the statement Potential users do not that were tagged with ‘hardware: sensors’ were also tagged know about this BCI tool was agreed upon for almost all with ‘user friendliness’, which may suggest that the user implanted and non-invasive application scenarios in the friendliness of non-invasive sensors is an important niche current study. Studies involving both healthy and disabled for future development and eventual user acceptance. users, however, have found that 50–80% of the interviewed Although most implanted statements were patient-ori- potential users were aware of BCIs before they completed ented and most non-invasive statements were oriented the respective surveys [10,16,17]. Although it is conceiva- towards healthy subjects, it was interesting to note that ble that participants (i.e. potential users) of these previous people also foresee a role of implanted systems for healthy surveys on BCIs are more likely to participate if they have end-users, indicating that implanted BCI applications heard of BCIs before (potential survey recruitment bias), may, in the long run, find a niche outside the patient-ori- it is also possible that the respondents of the current study ented ‘replace’ scenario. Implanted applications for healthy (i.e. BCI researchers) underestimate the awareness among end-users were oen r ft elated to futuristic scenarios, such potential users. as brain-to-brain communication or enhancing brain capacity, but also approaches to enhance daily functioning or to monitor health were mentioned (Direct behavioral 4.2. Far future modic fi ation. Using implants to enhance or suppress desira - Since the list of topics used for the labeling of the Far ble or undesirable habits or cravings or e d Th evelopment of future statements was derived from these statements technologies that can monitor and care for the user beyond themselves, the list can be considered as an exhaustive the perception of it). overview of research areas that have not been addressed Besides being an overview of topics where break- adequately in the past decades of BCI research, and where throughs are needed for currently known BCI approaches major breakthroughs are needed. The topic of ‘user friend - to be actually used by the target populations, the list of far liness’ was addressed the most frequently: in over 20% future statements also contained many descriptions that 246 M . J. VANSTEENSEL ET AL. go far beyond that. A selection of these truly out-of-the system performance, and user friendliness is crucial for box statements, given in Table 3, may act as an inspiration non-invasive BCIs, as well as an increased incorpora- for current and future BCI researchers. In addition, these tion of the wishes and needs of end-users. For implanted statements may serve the ethical debate that will necessar- BCI applications, steps have to be taken to develop com- ily accompany further BCI development, in defining the pletely implantable systems with adequate performance currently foreseeable concepts and technology. and durability, and to make these systems available for patients. Moreover, the results warrant a recommendation for the field to move towards clinical trials to demonstrate 4.3. Limitations that implanted BCIs can be safe and durable, and yield With two rounds of the questionnaire, 298 respond- good performance. Finally, the field stands to benefit from ents completed the form, which is 9.1% of the 3291 BCI identifying the benefits of implanted over non-invasive researchers who were invited to do so. Notably, this solutions, and vice versa, recognizing the individual users’ percentage may be an underestimation, since a subset needs and circumstances. of the contacts may not be active in BCI research any more. In fact, in the second round of invitations to fill Acknowledgements out the questionnaire, we added the option to select I am e a Th uthors would like to thank the people from the BNCI Ho- no longer involved in BCI research, without further filling rizon 2020 consortium for their valuable contribution during out the questionnaire. Upon this second invitation, 21 the development of the questionnaire. people selected this option, and 69 people did complete the questionnaire (the other 229 responses were obtained in the first round). This suggests that possibly more than Disclosure statement 20% of the people addressed were no longer working in e a Th uthors have no conflicts of interest to disclose. BCI research. Nevertheless, the incomplete response rate is an obvious limitation of the current study. Other recent questionnaires with email invitations among biomedical Funding researchers have obtained response rates between 16 and This work was supported by the FP7 project ‘e Th Future of 53% [26–28]. Reasons for the relatively low response Brain/Neural Computer Interaction: Horizon 2020’ under rate may be the length of the questionnaire, as well as Grant 609593; ERC Advanced Grant ‘Intracranial Connection its complexity (i.e. some open questions, describing with Neural Networks for Enabling Communication in Total paralysis’ under Grant 320708; and STW project ‘Pilot study of ideas etc.). Another limitation is that we potentially have an implantable brain-computer interface system with people a bias towards European respondents: more than 50% with locked-in syndrome’ under Grant STW12803. of the responses came from people working in Europe. Considering the context of the questionnaire, being part of the BNCI Horizon 2020 roadmap activities, this large per- ORCID centage of European respondents may not be surprising. M. J. Vansteensel   http://orcid.org/0000-0002-9252-5116 e Th y may have more incentive to give their opinion about E. J. Aarnoutse   http://orcid.org/0000-0001-7648-250X where the BCI field should head in the future. Despite N. F. Ramsey   http://orcid.org/0000-0002-7136-259X these constraints, several of our findings correspond with previous reports (see above) and give us confidence that References the outcome of this questionnaire and the topics that are identified reasonably reflect the BCI research field. [1] Fetz EE. Operant conditioning of cortical unit activity. Science. 1969;63:955–958. [2] Vidal JJ. Toward direct brain-computer communication. 5. Conclusions Annu Rev Biophys Bioeng. 1973;2:157–180. [3] Wolpaw JR, Winter Wolpaw E. Brain–computer From the current study, we conclude that BCI research- interfaces. principles and practice. New York (NY): ers are quite optimistic about the feasibility of BCI appli- Oxford University Press; 2012. cations for both patients and healthy end-users. They [4] Brunner C, Birbaumer N, Blankertz B, et al. BNCI Horizon described many types of applications that would be via- 2020: towards a roadmap for the BCI community. Brain- Computer Interfaces. 2015;2:1–10. ble in the near and far future. Especially applications for [5] Nicolas-Alonso LF, Gomez-Gil J. Brain computer patients, to replace or improve lost brain function, are interfaces, a review. Sensors. 2012;12:1211–1279. considered promising by many BCI researchers. However, [6] BNCI Horizon2020 Consortium. Roadmap. the future of before these products can penetrate the market, a num- brain/neural computer interaction: Horizon2020. Graz: ber of issues have to be addressed. Improving sensors, University of Technology; 2015. BRAIN-COMPUTER INTERFACES 247 [7] Elo S, Kyngäs H. The qualitative content analysis process. [19] K übler A, Holz EM, Riccio A, et al. The user-centered J Adv Nurs. 2008;62:107–115. design as novel perspective for evaluating the usability [8] Hsieh HF, Shannon SE. Three approaches to qualitative of BCI-controlled applications. PLoS One. 2014;9: content analysis. Qual Health Res. 2005;15:1277–1288. e112392. [9] Nijboer F, Clausen J, Allison BZ, et al. The Asilomar [20] Blain-Moraes S, Schaff R, Gruis KL, et al. Barriers to and survey: Stakeholders’ opinions on ethical issues related to mediators of brain-computer interface user acceptance: brain-computer interfacing. Neuroethics. 2013;6:541–578. user group findings. Ergonomics. 2012;55:516–525. [10] Ahn M, Lee M, Choi J, et al. A review of brain-computer [21] Holz EM, Höhne J, Staiger-Sälzer P, et al. Brain-computer interface games and an opinion survey from researchers, interface controlled gaming: evaluation of usability by developers and users. Sensors. 2014;14:14601–14633. severely motor restricted end-users. Artif Intell Med. [11] S itaram R, Lee, S, Birbaumer N. BCIs that use brain 2013;59:111–120. metabolic signals. In: Wolpaw JR, Winter Wolpaw E, [22] Leeb R, Perdikis S, Tonin L, et al. Transferring brain- editors. Brain-computer interfaces. principles and practice. computer interfaces beyond the laboratory: successful New York (NY): Oxford University Press; 2012. p. 301–314. application control for motor-disabled users. Artif Intell [12] Vansteensel MJ, Hermes D, Aarnoutse EJ, et al. Brain- Med. 2013;59:121–132. computer interfacing based on cognitive control. Ann [23] Zickler C, Halder S, Kleih SC, et al. Brain Painting: Neurol. 2010;67:809–816. usability testing according to the user-centered design in [13] Bleichner MG, Jansma JM, Salari E, et al. Classification end users with severe motor paralysis. Artif Intell Med. of mouth movements using 7T fMRI. J Neural Eng. 2013;59:99–110. 2015;12:066026. [24] Peters B, Bieker G, Heckman SM, et al. Brain-computer [14] Berger TW, Chapin JK, Gerhardt GA, et al. International interface users speak up: e Th virtual users’ forum at the assessment of research and development in brain- 2013 international brain-computer interface meeting. computer interfaces. Maryland: World Technology Arch Phys Med Rehabil. 2015;96:S33–S37. Evaluation Center, Inc; 2007. [25] Nijboer F. Technology transfer of brain-computer [15] Huggins JE, Wren PA, Gruis KL. What would brain- interfaces as assistive technology: barriers and computer interface users want? Opinions and priorities opportunities. Ann Phys Rehabil Med. 2015;58:35–38. of potential users with amyotrophic lateral sclerosis. [26] Oushy MH, Palacios R, Holden A, et al. To share or Amyotroph Lateral Scler. 2011;12:318–324. not to share? A survey of biomedical researchers in the [16] Collinger JL, Boninger ML, Bruns TM, et al. Functional U.S. Southwest, an ethnically diverse region. PLoS One. priorities, assistive technology, and brain-computer 2015;10:e0138239. interfaces aer s ft pinal cord injury. J Rehabil Res Dev. [27] Teunis T, Nota SP, Schwab JH. Do corresponding authors 2013;50:145–160. take responsibility for their work? A covert survey. Clin [17] Lahr J, Schwartz C, Heimbach B, et al. Invasive brain- Orthop Relat Res. 2015;473:729–735. machine interfaces: a survey of paralyzed patients’ [28] Al-Herz W, Haider H, Al-Bahhar M, et al. Honorary attitudes, knowledge and methods of information authorship in biomedical journals: how common is it and retrieval. J Neural Eng. 2015;12:043001. why does it exist? J Med Ethics. 2014;40:346–348. [18] Blabe CH, Gilja V, Chestek CA, et al. Assessment of brain-machine interfaces from the perspective of people with paralysis. J Neural Eng. 2015;12:043002.

Journal

Brain-Computer InterfacesTaylor & Francis

Published: Oct 2, 2017

Keywords: Brain-computer interface; survey; questionnaire; research; future

References