Consumer Brain-Computer Interface: Challenges & Opportunities

Active algorithms, hybrid BCI, Scalability, Bandwith, and Future of non-invasive BCI headsets

Photo by on

Brain-Computer Interfaces are poised to become the next trendy consumer device. As such, most consumer electronics companies have started to integrate them into their roadmap in some way or another.

Some companies have already developed and released consumer wireless EEG-based BCI with interesting non-medical applications. Now, measuring brain activity is no longer limited to medical diagnostics but includes more applications aiming to change users’ lifestyles.

As part of a team in charge of innovation, I was involved in the development of several BCI prototypes. This article will present some of the current technical and commercial challenges of wireless BCI systems and discuss a few possible future research directions to enable wireless BCI systems to become scalable.

Brain-Computer Interface: The basics

Image by Author

Current Challenges and Directions

Non-invasive / Invasive BCI system

Existing recording devices are divided into two main categories: invasive and non-invasive. The invasive category, which requires implanting surgery, is usually needed for critical paralyzed situations because of their higher accuracy rates.

Many wired clinical-grades devices with high accuracy currently use wet electrodes dipped in gel or saline solution, which is not suitable for everyday life. Nevertheless, recent research shows that dry electrodes are capable of producing similar results compared to wet ones [].

On the other hand, the non-invasive category is the one most suited for immediate commercial applications. Non-invasive EEG-based BCIs have been widely spread in different application fields due to its advantages over the invasive one.

Non-invasive BCI cannot compare with invasive BCI in the senses of accuracy and directness.

However, non-invasive systems based on EEG lack the spatial resolution [] to record detailed activity at the level of the neuronal circuit, and so can only be used for very simple low bandwidth interfaces.

Bandwidth describes the maximum data transfer rate of a network or Internet connection. It measures how much data can be sent over a specific connection in a given amount of time.

Consumer wireless EEG-based non-invasive BCI

On top of the elements previously mentioned, the number of electrodes on consumer wearables is limited compared to the clinical-grade devices. Moreover, electrodes are usually focused on a specific area of the brain, and their resolution is lower than those of high-clinical density [].

This, alongside the low signal-to-noise ratio (SNR) of consumer consumer-grade devices, questions their validity as reliable neuroscience solutions.

Training process

Another issue when it comes to the scalability of consumer BCI systems is related to the necessary training process. First, BCI users must be trained to use such devices. In some cases, it might take many distinct sessions before accurately using a BCI.

For non-commercial systems, training the user is a time-consuming activity to guide the user through the process or in the number of recorded sessions. This situation is a real issue because if users do not see tangible results within the first training sessions, they will likely lose motivation to continue investing effort into trying to control the BCI.

In the future, we would need to reduce the necessary training time of consumer BCI by perhaps relying on various Adaptive and zero training classifiers .

Nonstationarity and noise

BCI systems also suffer from the fact that mental and emotional state background through different sessions can contribute in EEG signals variability.

The same can be said about fatigue and concentration levels which are considered part of internal nonstationarity factors. Noise is also a big contributor in the challenges facing the BCI technology and causing the nonstationarity issue. Future scalable consumer BCI systems will have to reduce the negative impact of EEG signals variability on accuracy.

Training sets

Another significant challenge in designing a commercial BCI is to balance the trade-off between the technological complexity of interpreting the user’s brain signals and the amount of training needed for successful operation of the interface .

In general, more targeted work needs to be done around the amount of data required to fully exploit the potential advantages of Deep Learning in EEG processing.

Such work could explore the relationship between performance and the amount of data, the relationship between performance and data augmentation, the use of Generative Adversarial Networks, the relationship between performance, the amount of data, and the network’s depth.

EEG Data, Image by Author

Hardware

The development of consumer BCI systems highly depends on the hardware aspect. However, substantial challenges with existing BCI implementations have prevented its widespread adoption []. Therefore, the priority is to develop comfortable, convenient, and stable signal-acquisition hardware.

Clinical-grade BCI systems are often wired. With just three electrodes positioned at the occipital lobe, the acquisition part of wired BCI systems generally comes with bulky and heavy amplifiers and preprocessing units. In addition, connection wiring is usually complicated with many cables between the electrodes and the acquisition part. For these reasons, preparation time for measuring EEG signals is typically very long. In addition, the user’s movement is limited due to cable constraints.

Despite these limits, most teams involved in BCI competition use gelled electrodes, indicating a certain level of distrust towards dry electrodes reliable enough for use in uncontrolled environments. Similarly, most teams I encountered used laboratory-grade EEG amplifiers, suggesting that no team trusted consumer-grade devices to reach high accuracy.

Issues with wireless BCI

In terms of scalability, wireless BCIs are an improvement compared to conventional wired BCI. However, “battery life, device weight, and form factors” [] are still issues.

In the future, a scalable non-invasive wireless BCI will result from a perfect balance between the cost of high sampling rates, an ‘always-connected’ state, and improved battery life to sustain the long hours required for daily usage. Additionally, users should not have to wear cumbersome headsets or experience long setup/training times.

Wireless BCI hardware should also have advanced electrodes for measuring clean EEG signals.

Another reality about wireless non-invasive BCI is that “the more data you collect, the slower you become, and the faster you get, the less data you can grab” []. So future hardware will have to find the perfect balance between speed and accuracy.

Lack of applications

Several companies introduced various BCI applications related to entertainment and health care. But such applications are still not enough to make any significant change in our life.
Is sleep quality monitoring using a non-invasive BCI enough to convince people to purchase a 300$ device? Some experts might argue that EEG does not have the potential to make huge differences in our daily lives…

One of the main challenges for non-invasive wireless BCIs is to provide value for many different user populations. One solution could be to integrate BCIs into the microcosm of the user’s devices (smart home) by providing mobile APIs and Bluetooth connectivity. Another one could be to improve the relationship between smartphones and BCI, for instance, visual keyboards with messaging apps [].

The interaction with other devices using BCI could be a real reason to trigger a non-invasive wireless BCI purchase. There is another exciting category of BCI: passive BCI, for which “the mental state of the user is passively estimated, without any voluntary mental command from the user, to adapt the application in real-time to this mental state” []. I expect this type of BCI to become popular in a few years combined with other wearables.

Another solution could be to develop further applications dedicated to human-machine collaboration for improved decision making, assisted-human operations, and efficiency at work. Neuro-ergonomics (the study of human brain with performance at work and everyday setting) is a promising field that will benefit from BCI.

Perhaps, the real breakthrough for consumers will be bidirectional BCI. For instance, we can imagine that thanks to an invasive BCI, when a robotic arm touches an object, the sensor can feedback this information to the brain by electrically stimulating the cortex of the participant so that the user can genuinely feel that he or she is touching the object.

User Experience

Beyond relevant applications, consumer experience is also a strategic topic for brands willing to invest in BCI. At an individual level, the use of EEG-based wireless non-invasive BCI to monitor attention level or quality of sleep becomes either tedious or frustrating over time [].

Task execution time should be as low as possible.

I believe the future of EEG-based non-invasive BCI lies in collaborative endeavors. Since human beings are inherently social creatures, advanced EEG technologies could foster their interactions.

For example, consumers could be empowered and motivated by bringing them into large-scale interactive projects or programs in which users may communicate within the legal constraints. In addition, “the user experience may be significantly improved by detecting users’ affective states to adapt individual and collaborative features” ().

Training users to control BCI has been scarcely studied in the BCI literature so far, and the best way to train users to master BCI control skills is still unknown []. Moreover, it is estimated that, with current systems, 10 to 30% of BCI users cannot reach BCI control at all (so-called BCI deficiency)[].

New electrodes & Data generalization

According to various researchers, “to provide clear EEG signal acquisition at the electrode-skin interface” [], the development of outstanding electrodes is a critical issue. For this reason, many research groups have recently been interested in developing advanced electrodes which can provide low-noise recording, convenience in installation, and comfort even in long-term wearing.

Recent wireless non-invasive BCI systems are equipped with active dry electrodes to combine the advantages of active and dry electrodes, such as convenient installation and high fidelity signals. Because these electrodes provide more stable signal quality in mobile environments, they are suitable for wireless non-invasive BCI systems. Soon, I expect to see more improved dry electrodes for commercial non-invasive BCI.

EEG is also a non-stationary signal []. As a result, a classifier trained on a temporally limited amount of user data might generalize poorly to data recorded at a different time on the same individual.

Non-stationary signal is the kind of signal where time period, frequency are not constant but variable.

This issue is another critical challenge for real-life applications of EEG, which often need to work with limited amounts of data.

Design

Depending on the application and target users, several hardware designs have been used. For example, existing wireless non-invasive BCI systems include headsets, headbands, baseball caps, and military helmets.

Image by Author

While most existing systems may not be socially accepted, everyday life BCIs should be more discrete, for example, by integrating into hats, ice caps, glasses, or other head-worn garments. In this way, users can wear the BCI daily. Moreover, EEG-based BCI must be comfortable to wear for several hours.

Wireless non-invasive BCI should also try to “minimize power consumption, enabling a user’s brain activity to be recorded in their own home for 24 hours continuously” [].

I expect companies to continue investing in the development of portable, miniature, and discrete wireless consumer BCI. The majority of existing wireless non-invasive BCI systems are too cumbersome to become scalable everyday solutions used outside of the medical field.

Accuracy & Applications

While working on various prototypes for my company, I realized that current wireless EEG-based non-invasive BCI systems are not reliable enough to be used in some accuracy-critical applications.

In wireless BCI systems, many features of the EEG signals, such as cognitive states, event-related potentials (ERPs) in P300, and steady-state visual evoked potentials (SSVEP), are used for BCI-based controls. However, as mentioned before, these features are easily negatively impacted by various noise and inference sources.

For these reasons, most commercial wireless BCI systems are developed for less accuracy-critical applications such as computer games and home appliances. However, the future of commercial BCI also depends on solving this issue of accuracy in critical applications.

Hybrid signal acquisition

One way to solve the accuracy issue could be through hybrid signal acquisition for higher accuracy and fast brain-computer interaction. Indeed, some experts believe that the limit of non-invasive BCI systems comes from the nature of EEG data.

Hybrid signal acquisition through simultaneous recording of multiple brain signals has been shown to ensure higher accuracy thanks to the complementary analysis of user motivations [].

The three primary purposes of hybridization are:

  • Increasing the number of control commands
  • Improving classification accuracy
  • Reducing the signal detection time

Currently, combinations of EEG + fNIRS and EEG + EOG are most commonly employed.

In my opinion, hBCI’s (hybrid BCI) main application is safety. It can be useful in monitoring the vigilance levels of specific workers (pilots, drivers, doctors, etc.). Although consumer BCI systems that can monitor brain activity and alert drowsy drivers do not yet exist, hBCI might contribute to the development of such a commercial system.

Suppose wireless BCI systems employ additional features of EEG signals which are not generated by the same motivation. In that case, the accuracy of the applications will be improved by adopting complementary classification.

I believe the success of BCI will depend on our ability to go beyond brain waves collected using scalp EEG or combining several EEG protocols and combining them with other ways to measure brain activity.

The new advances we can expect are primarily in the field of non-invasive signals collection. Signal processing & computation also needs to be improved to better deal with a massive (multichannels, multimodalities,…) flux of varied signals used by the BCI, without relying on the traditional signal processing techniques & BCI scheme.

Adaptive algorithms

Adaptive algorithms are needed to reduce training time and achieve higher accuracy.

An adaptive algorithm is an algorithm that changes its behavior at the time it is run, based on information available and on a priori defined reward mechanism.

Because every person has a unique set of their own EEG characteristics, most applications require dedicated training procedures for learning the user’s EEG patterns. However, the EEG patterns typically change continuously, affected by many factors such as the mental state of the users and the circumstance surrounding the users. Furthermore, “long-term training can make the users tired and induce degradation in accuracy” [].

For these reasons, the reduction of training in applications is a crucial issue in BCI researches. To reduce training, additional signal processing schemes likes adaptive classification algorithms can be added to BCI systems []. Because these schemes can discern changes in EEG features, the accuracy of applications can be greatly improved.

Scalable consumer non-invasive wireless BCI systems will provide high-fidelity data acquisition, and fast onboard signal processing are available at a low cost.

Data Privacy and Cybersecurity

Clinical-based EEG-based BCI can monitor a patient’s emotional and cognitive reactions based on specific elements. These metrics allow the medical team to obtain a patient’s feelings toward about sensitive topics.

It is safe to assume that many companies will see this lucrative brain data market and start collecting this sensitive data to sell it to third parties without the users’ consent. This situation can be exploited by individuals who sell personal data on the Dark Web [].

Therefore, it is crucial for BCI companies to address the trust deficit, as cultural barriers among potential buyers will get higher as more BCI enter the market.

The vulnerability of BCI commercial applications is also a topic of concern. It has been demonstrated that EEG-based BCI can be “hacked” to output anything the attacker wants with a high success rate []. Suppose these attacks may be targeted in other scenarios such as automatic driving where the feedback plays an important role, and the cost of a one-step mistake can have a significant impact on the user. In that case, the security of EEG commercial applications should be reconsidered before deployment.

The widespread adoption of commercial BCIs will depend explaining how safe the device is and how users’ brain data is being used. This element is crucial since EEG can directly extract sensitive information from the brain without the user being aware of it.

Therefore, companies and manufacturers of BCI devices will have to obtain the valid consent of the consumer by providing specific, informed, and detailed information on the processing of his sensitive data.

Bandwith

Another critical aspect for the success of consumer BCI will be the increase of the bandwidth between the brain and computer. The future of commercial BCI must increase the rate of data transfer across a given path.

I also believe BCIs could benefit from “increased computational platforms. Indeed, the computational power available on mobile platforms is limited; hence the signal-processing pipeline often needs to be optimized to suit a platform architecture” [].

For EEG-based wireless non-invasive BCI to be mass-marketed, some breakthroughs must be found (accuracy, speed, etc.). From a cost-benefit standpoint, EEG may not be the best option to invest in at the moment due to its minimal commercial viability and limited applications. However, I remain convinced that the next trendy consumer electronic will be brain-computer interfaces.

AI Consultant @Philips | I write about AI and BCI

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store