Introduction
Paralysis is a devastating condition that affects approximately 5.4 million people in the US alone [
1]. Among the various causes that result in paralysis, stroke is the most common cause followed by spinal cord injury (SCI) and multiple sclerosis [
2]. For SCI, incidence is highest in the US and the prevalence of cervical SCI is rising [
3]. Paralysis imposes a significant economic and social burden on the individual, their families, caregivers, and public health, as the costs for individuals with high tetraplegia can exceed $1 M in the 1st year after injury [
4]. For 40% of stroke subjects, and most SCI subjects, functional deficits are typically permanent, and no treatment yet exists. Thus, the need to address functional improvements and restoration of movement and independence in these subjects remains a critical challenge.
Despite a lack of available treatment, recent advances in intracortical brain–computer interfaces (BCI) have shown promising results in restoring functional reaching and grasping in individuals with paralysis [
5‐
8]. BCIs create an external link between the brain and body by means other than the body’s nervous system [
9]. While BCIs may target a variety of consumers, this technology yields promising outcomes for subjects with sensorimotor deficits, in which BCIs can circumvent lost function [
10].
Research efforts investigating invasive BCIs have highlighted their potential to restore function lost due to neurological disorder and injury. Invasive neural recordings have enabled human subjects to control virtual cursors [
11‐
16], computers [
17,
18], spellers [
19], virtual [
20‐
22] and robotic prosthetics [
6‐
8,
15,
23], exoskeletons [
24], and their own paralyzed limbs via functional electrical stimulation (FES) [
5,
25,
26]. These achievements highlight many remarkable improvements in BCIs, such as the ability of neural-control devices in multiple degrees of freedom [
8,
15,
23,
24]. These demonstrations generally require subjects to be tethered to the neural data acquisition hardware to allow for high-resolution data streaming [
6‐
8,
12‐
15,
17,
20‐
22,
25,
26]. However, the constant need to tether the subject to the data acquisition hardware limits the subject’s ability to use the BCI in their home and community, subsequently providing no real way for the subjects to control or configure how they interact with the device. Therefore, progress in the field to enable BCIs to function outside of a research environment remains a critical milestone [
27,
28].
The development of portable BCIs for out-of-lab use, or BCI systems that are not tethered to large stationary research equipment and can be easily transported anywhere, remains a significant challenge for intracortical BCIs as they must consider the constraints that current wireless technologies impose on high-data-rate wireless telemetry, ultra-low-power electronics, and space that the physical components occupy, both inside and on the surface of the body [
29,
30].
Research in animal studies continue to help solve these problems by developing wireless intracortical interfaces [
31] and modifying device parameters that decrease power consumption without significant impacts on decoder quality [
32]. Recently, Simeral and colleagues demonstrated an untethered approach of an intracortical BCI for controlling a computer cursor—relying on a short-range radio frequency (RF) receiver tower placed in the same room as the subject to capture high-bandwidth data transmission from a wireless RF head-mount [
33].
Notably, many studies that utilize electroencephalography (EEG) have accomplished home use in a variety of applications [
34‐
38]; however, EEG-based BCIs lack the spatial resolution captured by more invasive techniques [
11]. Moreover, the use of non-invasive technologies, such as EEG, for BCI control signals can complicate set up procedures (e.g., appropriate electrode placement) limiting the subject’s independence and the ease of BCI use at home.
Though invasive intracortical microelectrode arrays have excellent signal quality and have demonstrated many of the remarkable strengths of BCI capablities, these arrays suffer from diminished signal quality over time, typically as a result of the inflammation triggered by the foreign body response to the electrode shanks puncturing the blood–brain barrier [
39‐
42]. These devices are often percutaneous implants, where a portion of the implant is fully exposed increasing the risk of infection [
43].
Similar to intracortical micro arrays are the use of invasive electrodes on the surface of the brain via electrocorticography (ECoG). This method has enabled long-term and stable signal acquisition [
44] with higher spatial resolution and signal-to-noise ratio [
45]. Such applications have successfully controlled cursors [
11,
46], computers [
18], spellers [
47], and recently exoskeletons [
24] with multiple degrees of freedom. Chronic, fully implanted implementations have enabled the translation of research out of the lab and into the home [
19,
47]. While fully implantable BCIs require the need for invasive brain surgery, they have the potential to move applicable BCI technology into the home setting, by minimizing set up time for caregivers and minimizing percutaneous infections since no part of the implanted device is exposed, as is the case in intracortical microelectrode arrays.
Potential users of BCI and their caregivers indicated that portability, simple system configurations, and minimized setup times are important components of BCI system implementations [
48]. Addressing these design considerations will better facilitate transitions into the home. To accomplish this goal of designing at-home BCI systems, we used a minicomputer mounted on the subject’s wheelchair that wirelessly recorded ECoG data from the implanted Activa PC + S device and performed neural decoding. Our design included a modular approach, which enabled the subject to select from a variety of output devices and allowed for add-on peripheral devices in the future by utilizing a nearly plug-and-play ready system.
Discussion
Our design aimed to allow the BCI to function both in and out of the home and give the subject control over device selection, data collection, and system settings and preferences, all while minimizing the need for caregiver assisted setup. Although this system was implemented in a subject with SCI with an implanted neural interface, applications can be extended to other forms of paralysis, or wherever BCI can be employed, for example in use with EEG headsets or intracortical spike signals.
Mobile phone application
In our design, the App functioned as a GUI. However, other implementations have used smartphones for signal processing—using the phone as the sole processing unit for the system [
55‐
58]—as opposed to using an on-board minicomputer. Others have used smartphones to drive communication with home appliances [
35,
36]. Here, the use of NativeScript limited the development of the phone application to Bluetooth Low Energy protocols, which limits our system’s ability for high bandwidth data streaming that Bluetooth classic radio frequency communication (RFCOMM) could otherwise achieve for raw signal transmission. The alternative solution of using classic RFCOMM, however, would likely drain the battery of the phone fairly quickly, as reported in Campbell et al. [
56]. However, the implementation of RFCOMM would need to be considered for continuous and reliable data collection and streaming if the system were to be moved entirely onto the phone in future iterations. Additionally, migrating to phone-use warrants increased security measures. As [
59] points out, widespread and trusted use of such systems will require care to ensure the integrity and security of the system.
As an important note, our subject was able to navigate use of the App using the residual movements available with their bicep muscles. However, the App is simple enough to enable a caregiver to configure or alter system settings where needed, and doesn’t limit the application of our system to subjects with unique residual movements.
Portable system
For more practical use of BCIs outside of the laboratory, system portability is a significant component [
27]. One method for increasing portability, is using smaller computational equipment that has become more readily available in personal tablets, phones, and minicomputers. Several previous studies have used portable devices such as tablets and phones for command [
35] and signal processing [
55,
57,
58], subject interaction [
38], and as the end effector itself [
56]. For our design, we used a minicomputer and the subject’s smartphone, which allowed us to give the subject control of the BCI while ensuring that the BCI software could continue running independently of the phone. Additionally, by way of cost, these components are already consumer grade products that can readily be purchased. The short-range Bluetooth communication used by the phone and motorized orthotic allowed for constant control of the system, which permitted the subject to continue using the BCI outside of the home. Because the motorized orthotic communicated with the system over wireless Bluetooth radio, WiFi was only necessary for uploading data needed for offline analysis, the use of the BCI system itself ran independently of internet connectivity.
Modular system
Modular BCI designs for research development have helped improve scientific reproducibility and have reduced the time required to develop new BCI software and configurations. Several software tools, packages, and pipelines have been developed to help with this effort [
60‐
64]. Such applications have been used successfully in the lab and at home using some of these systems [
65]. In addition to these tools, other studies have shown a variety of successful output controls including, cursor control, spellers, home appliance control, exoskeletons, prosthetics, and FES [
6‐
8,
11‐
24,
47]. Our design focused on end-user interaction while enabling the software to scale across OS and processor architectures. While we only utilized a motorized orthotic when testing our device, we ensured the system could be modular, including dynamic changes to input and output devices at runtime. This modularity was another purpose for employing the subject’s mobile device. It allowed the subject to swap end effector devices, initiate training sessions, and adjust available system settings.
Input devices (e.g., EEG system) have plug-n-play functionality, when a device class is provided. Implementing a device class requires some mechanism for sampling digitalized data from the device to the computer; e.g., through serial port, socket, or device-provided API. A decoder from our work can be used or custom made that transforms, classifies, or regresses the input signal into a meaningful value. The device class must inherit the device base class and define a get_input method that then (1) collects the data as previously determines, (2) decodes that data into a meaningful value, then (3) returns that value. Lastly, serial port, socket, or API properties can be defined on the class that the platform will utilize to automatically detect the device during start up. With these in place, a new device can be added to the platform, and will be available for selection on the mobile app for use.
Subject interaction with the at-home BCI platform
Together, the BCI platform and its implementation not only enabled a subject with cervical SCI to reanimate hand grasp at home and swap modular components, but it also gave the subject control over the settings and preferences for each device individually and the BCI system as a whole. More specifically, these settings included options for changing the response time of the motorized orthotic and restarting the BCI application. Researcher assistance or transport to the laboratory became unnecessary because the App allowed the subject to initiate data collection sessions on their schedule.
The deployment of our system for use at home minimized the need for complex donning and doffing procedures. Over several sessions, we measured the average setup time, from start until the subject could control the glove, to be around 5 min. This is well within range of previously recommended setup times of 10–20 min desired by surveyed potential users [
48]. A large part of this simplification is likely a result of using a fully implanted device. This setup avoids the use of EEG caps that might require wire connections, gel application, accurate placement, and configuration of the EEG cap [
66,
67]. In comparison, a recent study was able to achieve an average setup time of 20 min in 8 caretaker-subject pairs over several sessions [
68]. In general, setup times are dependent on the setup time of each component of a BCI system. Continued developments in non-invasive technologies will help drive down these setup times and configuration complexities. Although, user-evaluation was not systematically evaluated the subject did comment: “It’s easy to use and control and only sometimes take a bit of time to connect over Bluetooth”.
Though our implementation only utilizes 4 ECoG channels, potentially limiting the functional output of the system in comparison to many EEG channels, the purpose of this implementation was to minimize the setup time and complexity to use at home, while still delivering some functionality to the subject. Still, even fully-implanted systems can require a wired setup that requires plugging in the system [
6‐
8,
19,
47].
In our implementation, however, switching on the battery fixed to the back of the wheelchair was sufficient to start up the system. From here, aligning the telemetry antenna to the subcutaneous transmitter enabled neural control. Notably, the ease of setup because of using an implanted device such as the Activa PC + S is a matter of implementation of this BCI platform rather than a requirement to use the platform. The platform could be implemented using an EEG headset. For the implementation presented here, the fully implanted Activa PC + S was used to improve at-home setup and use of the system as a whole.
Limitations
Python as the development language may not result in a completely OS agnostic platform. This limitation is mostly a result of methods the language uses to interface with the OS and hardware. An alternative solution could be to utilize already available software such as BCI2000 [
60]. However, using an interpreted language such as Python for development removed the need for compile time during remote system updates.
The subject in the implementation presented here was able to use the App GUI on their own using their remnant bicep function. This generally limits the GUI’s use for less abled subjects. However, the App GUI still provides convenient system access for the caregiver. Additionally, future iterations are planned to enable voice activation and/or gaze control to enable better control in subjects with varying disability and control of the hand and arm.
Using the Activa PC + S and Nexus telemeter isn’t completely wireless. Data sample collection between the Nexus telemeter and the Activa PC + S requires an external antenna to be placed close to the implanted generator. However, Newer implementations of implantable generators (Medtronic Percept, St. Jude, Boston Scientific, Clinatec, etc.) are building wireless capabilities into their generators. Additionally, the Activa PC + S, used in cases of deep brain stimulation, can last up to 5 years, but the battery life of the device used in this context where stimulation is not used, is unknown. Implants that can be inductively charged [
69] and wirelessly transmit their signals [
70] will provide a more robust system that further minimizes system maintenance and caregiver assistance while maximizing portability.
Like many invasive studies, our implementation of an invasive device is limited to one subject, and thus difficult to generalize. Fortunately, using event-related desynchronizations as a method to perform binary classification of motor imagery is well established [
68,
71‐
74], and a few studies have employed it in the home [
34,
68,
75,
76]. Although, we employed this BCI system for use in a subject with cervical SCI, it has provided insights in to limitations how the software and App can be improved for easier incorporation into other BCI implementations.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.