Brain–computer interfaces in 2023–2024
Shugeng Chen and Mingyi Chen contributed equally to this work.
Abstract
Brain–computer interfaces (BCIs) have advanced at a rapid pace in recent years, particularly in the medical domain. This review provides a comprehensive summary of the progress made in medical BCIs during the 2023–2024 period, covering a wide range of topics from invasive to non-invasive techniques, and from fundamental mechanisms to clinical applications. The 2023–2024 period saw numerous research breakthroughs and clinical applications of BCI technology. As BCI hardware and software continue to evolve, and as the understanding of basic medical principles deepens, the expectation is that innovative BCI inventions will increasingly be introduced in clinical practice. Both invasive and non-invasive BCI technologies are paving the way for broader clinical applications. It is anticipated that BCI technologies will offer greater hope for disease treatment, provide additional methods of enhancing human bodily functions, and ultimately improve the quality of life.
Key points
What is already known about this topic?
-
While Brain–computer interfaces (BCIs) are developing rapidly and in various directions, there is a need to systematically understand their specific hardware and software, clinical applications, and underlying mechanisms in order to better promote their development.
What does this study add?
-
This study reviews the advancements in BCI technology from 2023 to 2024, covering hardware and software progress, neural mechanisms, and specific clinical applications, to provide valuable insights for the field's development.
1 INTRODUCTION
Brain–computer interfaces (BCIs), as revolutionary neuroengineering technologies, decode neural signals to provide rehabilitation solutions for patients with amputations or neurological deficits. By bypassing peripheral neuromuscular pathways, BCI systems translate electroencephalographic activity into control commands for external devices (e.g., computers, assistive tools, and neural prostheses), significantly enhancing the quality of life of individuals with severe motor disabilities while reducing healthcare costs. A BCI operates through five stages: signal acquisition (noise reduction and artifact processing), preprocessing (signal optimization), feature extraction (discriminative neural pattern identification), classification (intent recognition), and control interfacing (command execution). While invasive BCIs demonstrate superior signal-to-noise ratios, sensitivity, and resolution through direct neural–electrode contact,1 non-invasive BCIs hold exceptional clinical value due to their non-surgical nature, safety, and scalability. For instance, systems based on electroencephalography (EEG) eliminate implantation risks and are widely applicable in pediatric/geriatric care and short-term interventions, showing promise in neurofeedback therapy for depression, attention modulation in attention deficit hyperactivity disorder (ADHD), and social skills training for autism. Advances in high-density electrode arrays and signal processing algorithms (e.g., spatial filtering and deep learning) have narrowed performance gaps with invasive counterparts, particularly in motor imagery (MI) paradigms, achieving sufficient accuracy for wheelchair control or virtual keyboard operation while avoiding surgical complications—making them ideal for mild-to-moderate impairments.
As of 2023–2024, BCIs have achieved breakthroughs across three domains: therapeutic management of linguistic/motor deficits, mental navigation research, and emerging technology development. In language rehabilitation, invasive BCIs enable real-time linguistic signal decoding with tonal analysis, whereas non-invasive systems leverage dry electrodes and portable designs to enable home-based personalized training. In motor recovery, invasive BCIs assist patients with paralysis in walking with minimal calibration, and promote neuroplasticity; non-invasive systems induce neural reorganization in spinal cord injuries through closed-loop cortical modulation. Neuroscientific advances, such as hippocampal neural coding mechanisms,2 drive BCI innovation, complemented by minimally invasive frontiers like functional ultrasound (fUS) and endovascular BCIs that balance performance and safety.
Clinically, BCIs now address stroke, spinal cord injuries, amyotrophic lateral sclerosis (ALS), and neuropsychiatric disorders (e.g., depression, ADHD, and autism), forming a complementary framework wherein invasive BCIs target precision interventions for severe disabilities and non-invasive systems expand accessibility through low-risk deployment in primary care and large-scale screening. This synergy, enhanced by integration with AI and nanosensing, deepens the understanding of neuroplasticity and accelerates clinical paradigm shifts. Figure 1 shows the applications of different types of BCI systems based on brain regions and signal sources.
Applications of different types of BCI systems based on brain regions and signal sources.
2 PROGRESS IN BCI SOFTWARE AND HARDWARE
BCI translates complex brain signals or activities into computer commands, offering a promising method for restoring the capabilities of individuals with paralysis. Brain signals or activities can be recorded through numerous methods, including implantable neural recording probes (NPs),3, 4 multi-electrode arrays (MEAs),5, 6 electrocorticography (ECoG),7, 8 EEG,9, 10 functional near-infrared spectroscopy (fNIRS),11, 12 functional magnetic resonance imaging (fMRI)13, 14 and fUS.15 Broadly, BCIs can be classified as invasive or non-invasive. Figure 2 compares the spatial coverage and resolution between different types of BCIs.15 Non-invasive BCIs (EEG, fNIRS, fMRI) generally have greater spatial coverage, whereas invasive BCIs (NPs, MEAs, ECoG, fUS) offer improved spatial resolution. Aside from the necessity of surgical implantation, various performance trade-offs between signal quality, spatiotemporal resolution and spatial coverage must be considered to achieve an optimum BCI solution.
Spatial coverage and resolution of BCIs.
Due to its non-invasive nature, EEG has been one of the most implemented techniques in the neuroscientific field since its invention by Hans Berger in 1929.16 The EEG signal typically has an extremely low frequency (below 100 Hz) with a small amplitude (usually below 100 μV).17 Conventional EEG-based BCI mainly uses recording electrodes placed on different fixed positions on the head. To achieve good electrode–skin contact, a wet electrode fixed to the scalp with a conductive gel is conventionally used. Dry electrode recording is preferred for daily monitoring as it enables portability and provides stable data quality during long recording sessions. However, dry electrodes have much higher contact impedance (up to tens of MΩ) compared with wet electrodes (typically below 5 kΩ),18 that presents challenges to low-noise EEG-acquisition integrated circuits (ICs) with ultra-high input impedance. As shown in Figure 3A, a conventional EEG-acquisition IC consists of an analog frontend (AFE) followed by an analog-to-digital converter (ADC). The AFE includes a low-noise instrumental amplifier (IA) as well as a programmable-gain amplifier (PGA). It performs signal conditioning, which includes amplification, filtering, noise reduction, and so forth, on a weak EEG signal. The ADC then converts the preprocessed EEG signal into its digital counterpart, which is further processed by the back-end digital signal processing, and finally transmitted by the wireless part.
Typical IC architectures of BCIs.
Considerable developments have been made in non-invasive EEG-acquisition ICs in the following domains: 1) High input-impedance. State-of-the-art AFE architectures using impedance-boosting techniques have achieved input impedances exceeding the GΩ range within the EEG signal bandwidth.19-22 2) Increasing the common-mode rejection ratio (CMRR). Improving the CMRR is key to mitigating ambient common mode interference, such as the 50/60 Hz main interference. Several techniques, including the common-mode feedback,23 common-mode feedforward,24 common-mode replication,25 ping-pong auto zeroing (AZ) and chopper-stabilized26 techniques, have been proposed to improve the CMRR. State-of-the-art AFEs have boosted the CMRR of IAs up to 140 dB, and the system-level CMRR to more than 80 dB.27, 28 3) Innovative IA architectures for a better noise-efficient factor (NEF). The main-stream IA can have two types of architectures: driect current (DC)-coupled architecture based on the classic three-op-amp IA,29, 30 and Alternating Current (AC)-coupled architecture based on a capacitively coupled IA (CCIA).27, 31-35 The CCIA was invented in 2003 for neuro acquisition by Harrison.31 Compared to the classic three-op-amp IA, CCIAs offer a better NEF as less amplification is required. AZ and chopping techniques can be employed in CCIA to further reduce noise and thus improve the NEF.27, 32-35 In addition to the basic architectures, current feedback IA36-39 and current balance IA40, 41 have been proposed in recent years to achieve a better trade-off between input-impedance, NEF, and chip area.
Invasive BCIs record neural local field potentials (0.5–1000 Hz) or neural action potentials (0.3–10 kHz) with a slightly higher amplitude (typically from 100 μV to 1 mV). Kruger42 traces the earliest simultaneous recording to Matthews, 1929 from the frog peroneal nerve. Over the past 5 decades, advancements in neural recording techniques have enabled the number of neurons that can be recorded simultaneously to approximately double every 7 years, following Moore's law.42 Benefiting from modern IC fabrication technology, fully integrated high-density NPs deeply implanted into the brain cortex, or MEAs placed on the exposed surface of the brain, have become powerful neuroscience tools for performing single-cell neural recording across different brain regions. As shown in Figure 3(B), the most common neural-recording integrated circuit (NRIC) architecture consists of multiple AFE channels multiplexed into one ADC.43-59 The amplification stage is typically implemented as an AC-coupled or DC-coupled IA followed by a PGA. State-of-the-art NRICs have recorded thousands of sites; the performance and robustness of these NRICs have been well validated by large-scale and long-term recordings. However, compared with the huge number of neurons (∼80 billion) in the human brain, neural recording still has a long way to go. Aside from the number of recording sites, important specifications such as noise, power, area, input range, and electrode DC offset (EDO) tolerance are yet to be co-optimized in specific BCI applications.
In addition to neural recording, invasive BCIs are necessary for performing deep brain stimulation (DBS) to treat spinal cord injuries, Parkinson's disease, epilepsy, and other debilitating neurological conditions. The stimulation can be implemented in voltage or current modes. Concurrent DBS with recording is required to enable the stimulation parameters to be adaptively adjusted in a bidirectional closed-loop manner. Figure 3(C) shows the typical architecture of the bidirectional BCI.60 During DBS phases, the recording electrodes may be subject to concurrent stimulation artifacts (SAs). The amplitude of the SAs depends on several factors such as the amplitude of the stimulation current, and the distance between the recording and stimulation electrodes. In extreme scenarios, up to hundreds of mVs could saturate the entire recording channel. Thus, the dynamic range (DR) of the recording channel should be sufficiently high to accommodate small neural signals superposed onto large SAs. Conventionally, the IA's gain should be reduced or adaptively controlled to extend the DR.45, 61 Feedback techniques can be adopted to prevent saturation while improving the maximum input range.62, 63
Direct-conversion front-ends (Direct-FEs) have become a front-runner in artifacts-tolerant bio-potential acquisition front-ends in the recent years.64-78 Unlike conventional architecture (IA + ADC), the Direct-FE records bio-potentials without a front-end amplifier. Larger artifacts could thus be accommodated as no gain stage is employed. As background noise is typically on the order of 10 μVRMS,74 the higher input-referred circuit noise due to the lack of a front-end amplifier is usually inconsequential in real applications. By incorporating other advanced spectrum shaping techniques, such as ∆-modulation (∆M) or ∆-∆Σ-modulation (∆-∆ΣM), the irect-FE demonstrates superior DR than conventional architecture, placing it among the most promising candidates for wearable bio-potential acquisition. State-of-the-art Direct-FEs have three main architectures (see Figure 4): (A) ∆M; (B) single-loop ∆-∆ΣM, either DC- or AC- coupled; and (C) two-step ∆M. The ∆M was invented for multi-channel implantable recording due to its compact dimensions as well as low power dissipation.71-73 Nevertheless, its DR and signal-to-noise-and-distortion ratio (SNDR) are too limited for wearable devices where large artifact tolerance is a main concern.72, 73 In order to improve the SNDR, ∆-∆ΣM has been proposed74; however, the low input-impedance caused by the charge and discharge of the input switched-capacitor would not be satisfactory for wearable devices. Other AC-coupled ∆-∆ΣM Direct-Fes also have insufficient input-impedance for wearable devices.75, 76 Dc-coupled 2nd-order ∆-∆ΣM Direct-FE achieves high input impedance77; however, a large EDO might saturate the amplifier, and its linear-input-range fundamentally limits the DR. A two-step Direct-FE with improved ∆-modulation was proposed.78 It achieves a peak input range of 3.56 VPP, an input-impedance of 26 GΩ and a ±1.8 V EDO tolerance while consuming only 63 μW power. It is the first reported Direct-FE with a DR of over 110 dB, enabling the acquisition of weak bio-potentials superposed onto very large artifacts.
Architectures of state-of-the-art Direct-conversion front-ends (Direct-FEs) (A) ∆M; (B) ∆-∆ΣM; and (C) two-step78.
In conclusion, non-invasive BCIs are well suited for daily health monitoring due to their portability, while invasive BCIs can be widely used in medical therapy. To address challenges such as input-impedance, common-mode interference, artifacts, channel density, and so forth in real applications, multichannel bio-potential acquisition ICs with high DR, high CMRR, low noise, low power, and small area are yet to be well explored.
In addition to signal acquisition and processing equipment, data transmission, external control, and power systems are also key supporting modules that constitute a complete closed loop in BCI systems. The data transmission system establishes a physical connection between the biological entity and external device through wired means (such as high-precision medical cables) or wireless methods (such as low-latency Bluetooth 5.0). The stability and interference resistance of this system directly impact the real-time performance of the BCI. The control execution layer drives terminal devices such as prosthetic limbs and virtual interfaces using decoded neural signals while incorporating multi-modal feedback devices (such as tactile vibrations and visual projections) to form a sensory closed loop. The power management system must balance power efficiency and biocompatibility by employing wireless charging (Qi standard) and miniaturized solid-state battery technology to address the continuous power supply challenges of implanted devices. These three subsystems collectively form the complete chain from signal acquisition to functional output in BCI systems.
3 INVASIVE BCI AND FUNDAMENTAL MECHANISMS
Invasive BCIs have demonstrated a superior performance profile relative to their non-invasive counterparts, particularly in aspects such as signal-to-noise ratio, sensitivity, and resolution.1 As of 2023, BCIs have made significant achievements in the therapeutic management of patients with compromised linguistic and motor functions,8, 79-85 contributed to research in mental navigation,2, 86, 87 and been instrumental in the advancement of emerging technologies.15, 88-90
For patients with impaired language function, high-performance BCIs facilitate the accurate and real-time decoding of linguistic signals. To enrich the range of languages decoded, these interfaces also incorporate linguistic tonality analysis. In the context of motor function impairment, BCIs not only enable assisted walking for patients with paralysis but also showcase notable features such as minimal calibration, real-time operation, and the promotion of neurological recovery during the course of treatment.8, 82-84 Improvements in fundamental neuroscience research are crucial for developing novel BCIs. Investigations into the mechanisms of neural information coding in the hippocampus have provided new methods for the implementation of BCI navigation systems.2 Furthermore, emerging technologies such as fUS and endovascular BCI technologies represent the next frontier in minimally invasive brain–machine interfaces. These technologies offer a synergistic blend of high performance and safety, positioning themselves as a pivotal area of focus for future investigations.
3.1 Speech neuroprosthesis
A standard speech BCI involves the decoding of human brain signals and the generation of complete sentences via speech synthesis. However, the decoded output is typically in the form of text, which differs from conventional communication methods, and the rate of decoded output lags behind the pace of natural language processing. To enhance user experience, synthesized speech audio and facial-avatar animation can be employed to visualize the text. Some 2023 developments in neural prosthetics have not only facilitated real-time decoding but also attained a decoding speed of 78 words per minute (WPM) with a notably low error rate.8 Another study completed in 2023 reported a decoding rate of 62 WPM, demonstrating that high decoding performance is not excessively reliant on the language model. This research further revealed that patients retain the articulatory code for phonemes, which elucidates why the speech signals of patients who have been without the ability to speak for several years can still be successfully decoded.79
In 2024, a deeper exploration of the brain's encoding and decoding mechanisms was undertaken through various case studies in motor, visual, and language processing. These studies demonstrate the brain's ability to transform neural dynamics across distributed circuits into meaningful information about sensory and other task stimuli, which is critical for the development of effective BCIs. Techniques such as deep learning and mathematical tools have been applied to measure and enhance the fidelity of these processes.91
Recent advances in BCIs, particularly those achieved in 2024, have significantly impacted the field of speech restoration, particularly in patients with severe disabilities such as ALS. These interfaces, known as speech neuroprostheses, harness the brain's cortical signals related to speech attempts and decode them into text or synthesized speech using advanced computational algorithms. This emerging technology provides a vital communication lifeline to patients with severe paralysis and speech impairments, transforming theoretical concepts into practical life-enhancing applications.92
In the realm of language decoding, challenges extend beyond vowels and consonants, especially for tonal languages. For instance, a speech BCI tailored to Chinese is capable of translating decoded signal sequences into coherent Mandarin sentences. This facilitates understanding by native Chinese speakers, although decoding inaccuracies may occur.80 By modularizing the neural decoding of tones and basic syllables, coupled with speech synthesis, the complexity of signal processing can be significantly reduced, thereby enhancing performance.81 Moreover, research indicates that BCIs require minimal initial model training and calibration to maintain high levels of decoding accuracy, and this accuracy remains stable regardless of the duration of implantation.82
Further, the use of a large language model for error correction in BCI output allows for online recalibration, eliminating the necessity for frequent device interruptions, and enabling extended periods of continuous use.83
Further investigation of decoding speed and accuracy remains a critical area of focus. The current signal-to-text decoding speeds achieved by BCIs fall short of 160 WPM, a benchmark set by natural spoken language.8, 79, 93, 94 This highlights the necessity of reducing the error margins in BCI decoding. Additionally, current studies have yet to attain a high level of accuracy across extensive vocabularies, and the goal of enabling users to articulate unrestricted phrases with precision remains an aspirational objective. Table 1 shows more research progress in speech neuroprosthesis.
| Researcher | BCI types | BCI signals | Country/Region | Subjects enrollment | Application | Clinical efficacy | Reference |
|---|---|---|---|---|---|---|---|
| Metzger et al. | Invasive BCI | ECoG | USA | Quadriplegia and anarthria (n = 1) | Designed a multimodal speech neuroprosthesis to decode text and audio–visual speech outputs. | Achieved accurate and rapid large-vocabulary decoding with a median rate of 78 WPM and median word error rate of 25% for text. | [8] |
| Willett et al. | Invasive BCI | Intracortical microelectrode arrays | USA | ALS (n = 1) | Developed microelectrode arrays to record neural activity at single-neuron resolution. | The participant's attempted speech was decoded at 62 WPM, achieving a 9.1% word error rate on a 50-word vocabulary, and a 23.8% word error rate on a 125,000-word vocabulary. | [79] |
| Feng et al. | Invasive BCI | sEEG | China/USA | Epilepsy (males = 2, females = 2) | Developed a language model to transform the predicted syllable elements. | Achieved a high-performance decoding with a median character error rate of 29%. | [80] |
| Liu et al. | Invasive BCI | ECoG | China | Brain tumors (males = 4,female = 1) | Designed a modularized multi-stream neural network. | The mean accuracy of tonal syllable decoding increased from 55.7% to 75.6%, and the highest accuracy was 91.4%. | [81] |
| Luo et al. | Invasive BCI | ECoG | USA/the Netherlands | ALS (n = 1) | Designed a “plug-and-play” BCI control system from a chronic ECoG implant. | Achieved a median accuracy of 90.59%. | [82] |
| Fan et al. | Invasive BCI | Silicon microelectrode arrays | USA | ALS (n = 1) | Leveraged large language models to automatically correct errors in iBCI outputs. | Achieved a decoding accuracy of 93.84%. | [83] |
| Thomas et al. | Invasive BCI | sEEG | USA | Epilepsy (n = 8) | Trained linear classifiers to decode distinct speech components and evaluated the decoding performance using nested five-fold cross-validation. | Achieved an accuracy of 18.7% across 9 places of articulation, 26.5% across 5 manners of articulation labels, and 4.81% across 38 phonemes. | [95] |
| Soroush et al. | Invasive BCI | sEEG | USA/the Netherlands/Germany | Epilepsy (n = 7) | Used spatial, spectral, and temporal brain activity features to construct speech activity detection models. | Provided insights towards developing effective imagined speech decoding models. | [96] |
| Duraivel et al. | Invasive BCI | μECoG | USA | Movement disorder (n = 3), brain tumor (n = 1), epilepsy (males = 4, females = 7) | Performed high-resolution, micro-electrocorticographic neural recordings during intra-operative speech production. | Obtained neural signals with 57 × higher spatial resolution and 48% higher signal-to-noise ratio compared to macro-ECoG and SEEG, improved decoding by 35%. | [97] |
| Card et al. | Invasive BCI | fMRI | USA | ALS (n = 1) | Achieved an intracortical speech neuroprosthesis. | Achieved 99.6% accuracy with a 50-word vocabulary and 90.2% accuracy using a 125,000-word vocabulary. | [98] |
- Abbreviations: ALS, amyotrophic lateral sclerosis; BCIs, Brain–computer interfaces; ECoG, electrocorticography; fMRI, functional magnetic resonance imaging; SEEG, stereoelectroencephalogram; WPM, words per minute.
3.2 BCI motor control
Approximately 200 million individuals worldwide live with severe disabilities and functional impairments.99 Previous studies have employed spinal epidural electrical stimulation (EES) as a remedial approach in patients with spinal cord injuries. However, the patients encountered challenges in actively controlling muscle activity.100-102 Progressing beyond conventional EES methods, researchers developed a digital bridge connecting the brain and spinal cord, thereby affording patients enhanced flexibility in muscle control and facilitating more naturalistic walking abilities. Concurrently, this approach has shown promise in nerve restoration, as evidenced by a patient's regained ability to ambulate with the aid of crutches even when the brain–spine interface was switched off.84
Interestingly, a system initially designed for the restoration of upper limb movement has been adapted to control lower limb motion by anticipating the movement intentions of the upper limbs.85 This study underscores the potential for versatile applications of the same signal and system, a development that could significantly propel future advancements in the field.
When motor function in the extremities is compromised, BCIs play a pivotal role in restoring some degree of function, assisting in nerve connectivity, and potentially facilitating self-recovery. Nonetheless, there are critical areas for improvement, notably in reducing device latency and enhancing coordination to improve user experience. The time lapse between the detection of motor intention and the corresponding action by the prosthetic device may lead to discomfort with use from the patient's perspective.84 Table 2 shows more research progress in BCI motor control.
| Researcher | BCI types | BCI signals | Country/Region | Subjects enrollment | Application | Clinical efficacy | Reference |
|---|---|---|---|---|---|---|---|
| Lorach et al. | Invasive BCI | ECoG | Switzerland/France/UK/the Netherlands | Chronic tetraplegia (n = 1) | Developed a wireless, digital bridge between the brain and spinal cord. | Enabled the patient to stand and walk naturally in community settings. | [84] |
| Cajigas et al. | Invasive BCI | ECoG | USA/Sri Lanka | Cervical spinal cord injuries (n = 1) | Invasive BCI systems designed for upper-extremity motor control. | Achieved an accuracy of 84.15%. | [85] |
| Shah et al. | Invasive BCI | Silicon microelectrode arrays | USA | C4 AIS C spinal cord injury (n = 1) | Designed a virtual keyboard for typing with three finger groups. | The keyboard enabled the patient to type with 31 symbols at 90% accuracy, and at approximately 2.3 s/symbol. | [103] |
| Borra et al. | Invasive BCI | Dataset of V6A | Italy | Macaques | CNNs are effective neural decoders for invasive BCIs | A CNN-based data-driven analysis may provide insights for the encoding and function of brain regions. | [104] |
| Peterson et al. | Invasive BCI | STN-LFP and ECoG | USA/Germany | Parkinson disease (n = 11) | Spatial filters and patterns could be used to obtain neurophysiological information about the brain networks. | Adding spatial information to the model achieved a 6% gain in decoding. | [105] |
| Mender et al. | Invasive BCI | EMG | USA | Rhesus macaques (n = 2) | Neural activity shifted trajectories proportional to the required muscle activation. | 20.9%–61.7% of tuned SBP channels modulating activity with context. | [106] |
| Lim er al. | Invasive BCI | ECoG and MEA | USA | Epilepsy (n = 2) | PWNP could act as an efficient method of artifact suppression to biomimetically restore motor function. | Suppressed artifacts by 78%–80% and 85% in ECoG and MEA signals. | [107] |
- Abbreviations: AIS, ASIA impairment scale; BCIs, Brain–computer interfaces; ECoG, electrocorticography; EMG, electromyography; MEAs, multi-electrode arrays; PWNP, pre-whitening and null projection; SBP, spiking band power; STN-LFP, subthalamic nucleus-local field potentials.
3.3 BCI navigation
Research in mental navigation has potential applications in enhancing the quality of life of patients with paralysis, particularly in the context of wheelchair control. Traditional research has often relied on EEG or tracking eye movements, resulting in issues such as poor quality of decoded information and limited user movement. A 2023 study successfully decoded the speed of movement during real-world ambulation in real time through the hippocampus. This breakthrough suggests the feasibility of decoding a range of speeds from hippocampal activity, thereby laying the groundwork for the development of invasive hippocampal BCIs for wheelchair control.86
The hippocampus is known for its role in cognitive processing, particularly spatial information.108 Studies have revealed that rats, akin to humans, possess the capacity for imagination, retrospection, and future envisioning.
Similar to humans, rats can visualize places and objects not currently before them, and can imagine walking to a certain place or moving distant objects to a specific location without engaging in physical movement. These findings offer insight into the underlying mechanisms of episodic memory recall, mental simulation, and planning.2
A significant future challenge lies in improving the safety of brain region implantation for human research. The hippocampus, located deep within the temporal lobe, presents considerable difficulties for electrode implantation. The lateral prefrontal cortex (LPFC) offers a more accessible implantation site for decoding motor intentions. Investigations have found that there is significant neuronal activity in the LPFCs of mice while performing spatial movement-related tasks.87
Further exploration of brain functions, such as the neural mechanisms of navigation, is essential for developing novel BCI systems. The hippocampus plays a pivotal role in functions such as spatial navigation and place representation; understanding and harnessing these neural mechanisms are key to developing more sophisticated and effective BCI technologies. Table 3 shows more research progress in BCI navigation.
| Researcher | BCI types | BCI signals | Country/Region | Subjects enrollment | Application | Clinical efficacy | Reference |
|---|---|---|---|---|---|---|---|
| Lai et al. | Invasive BCI | LFP | USA | Rats | Provided insight into the mechanisms underlying episodic memory recall, etc. | High-level neural prosthetics using hippocampal representations. | [2] |
| Saal et al. | Invasive BCI | sEEG | The Netherlands/USA | Pharmaco-resistant epilepsy (males = 2, female = 1) | Developed an invasive neural prosthetic. | A decoder to classify virtual-movement speeds from hippocampal signals recorded during a virtual navigation task. | [86] |
| Johnston et al. | Invasive BCI | LFP | Canada/USA | Rhesus macaques (males = 2) | Multi-unit spiking activity recorded from the LPFC of non-human primates can be used to predict the location of a subject in a virtual maze. | LPFC could be a valuable implant location for an intracortical BCI system used for spatial navigation. | [87] |
- Abbreviations: BCIs, Brain–computer interfaces; LPFC, lateral prefrontal cortex; SEEG, stereoelectroencephalogram.
3.4 Emerging technologies
In traditional BCIs, the quality of recorded neural signals tends to decrease over the duration of implantation, and the challenge of neural variability remains unresolved.109 Notable advancements in emerging BCI technologies were made in 2023, providing novel solutions for these issues.
The utilization of ultrasmall implants for signal recording has shown promise in enhancing implant compatibility. This approach minimizes brain damage and decreases immune responses.88 A groundbreaking human study demonstrated that transmitting electrodes through cerebral blood vessels to record sensorimotor cortex activity presents a viable solution to the challenges of long-term implantability, with no observed problems such as movement of the device or blockage of blood vessels for a year after implantation.89 FUS, a recently developed technology, requires only the placement of an implant on the surface of the brain, and has high sensitivity and data volume. However, fUS has a low temporal resolution and cranial opening is required, owing to the attenuation of sound waves by the skull.15, 90 Table 4 shows more research progress in emerging technologies.
| Researcher | BCI types | BCI signals | Country/Region | Subjects enrollment | Application | Clinical efficacy | Reference |
|---|---|---|---|---|---|---|---|
| Zhang et al. | Interventional BCI | Endovascular probe | USA | Rats (n = 3) | Developed an ultraflexible micrometer-scale neuroelectronic interface which uses a native delivery system. | Demonstrated the minimal invasiveness of the MEV probes and indicated major advantages of chronic electrophysiology recording. | [88] |
| Mitchell et al. | Interventional BCI | MRI | Australia/USA | ALS (n = 4), PLS (n = 1) | Recording neuronal signals from a blood vessel in the brain. | Safe and rapid implementation of BCIs in people with paralysis. | [89] |
| Griggs et al. | Interventional BCI | fUS | USA/France | Rhesus macaque (n = 2) | Demonstrated successful implementation of a closed-loop ultrasonic BMI by fUS. | Achieved up to eight movement directions using the BMI. | [15] |
| Rabut et al. | Interventional BCI | fUS | USA | Traumatic brain injury (male = 1), rats (n = 4) | Developed a polymeric skull replacement material to create an acoustic window, allowing ultrasound to monitor brain activity. | Demonstrated a fully non-invasive mapping and decoding of cortical responses to finger movement. | [90] |
| Cadoni et al. | Interventional BCI | Ultrasound | France/Switzerland/USA | Long–Evans male rats and WT male mice | The in vivo sonogenetic activation of the visual cortex could generate a behavior associated with light perception. | In vivo sonogenetic activation of the visual cortex generated a behavior associated with light perception. | [110] |
| Brannigan et al. | Interventional BCI | Endovascular arrays | UK/Australia | Cynomolgus monkeys (n = 2) | Tissue proliferation under electrodes can reduced the amplitude power of signals. | Subdural ECoG may provide chronic signal recordings. | [111] |
- Abbreviations: ALS, amyotrophic lateral sclerosis; BCIs, Brain–computer interfaces; ECoG, electrocorticography; fUS, functional ultrasound; MEV, micro-endovascular; PLS, primary lateral sclerosis; WT, wild-type.
From 2023 to 2024, BCIs also made progress in the field of visual reconstruction. Blindsight by Neuralink is designed to restore vision to individuals who have lost their eyes as well as their optic nerves by stimulating the visual cortex directly. This technology could potentially enable individuals blind from birth to perceive visual information, initially in low resolution but possibly extending to capabilities beyond normal vision, including seeing the infrared and ultraviolet portions of the spectrum. While this promises revolutionary enhancements, the technology is still in its nascent stages, with the gradual improvements expected. The Illinois Institute of Technology's Intracortical Visual Prosthesis (ICVP) focuses on creating artificial vision by bypassing damaged optical pathways to stimulate the visual cortex directly. Developed by a team led by Dr. Philip Troyk, ICVP has achieved significant milestones in clinical testing, enabling participants to perform basic, visually guided tasks. This system uses a series of fully implanted miniaturized wireless stimulators to convey visual information directly to the brain, helping improve spatial awareness and mobility in individuals with severe visual impairments.
Research in BCIs has achieved remarkable findings, yet substantial room for further enhancement remains. To improve the efficacy and performance of BCIs, several challenges must be addressed, including: minimizing the calibration time before use in order to achieve a plug-and-play design; reducing time delays and achieving smoother user control of the prosthetic device; ensuring long-term reliability, as implanted interfaces should be viable for extended durations, and potentially a lifetime1; advancing neuroscience research, which is crucial for BCI development; and emphasizing the generalization of BCI systems. System performance would benefit from a combination of decoder design and neural adaptation.112 Future research trajectories for BCIs are likely to focus on developments in wireless and fully implantable technologies, aiming for more flexible control mechanisms.
4 PROGRESS IN CLINICAL MEDICINE
Modern technologies for monitoring, evaluating, and treating diseases (and resulting dysfunction) are eagerly awaited in clinical settings. The rapid development of BCI technologies presents new applications in clinical medicine. Conditions such as stroke, spinal cord injury (SCI), depression, ADHD, autism, ALS, and more, may increasingly benefit from either invasive or non-invasive BCIs. Table 5 shows the non-invasive BCI application in stroke diseases, Table 6 shows BCI application in other diseases and dysfunctions.
| Researcher | Year | BCI types | BCI signals | Country/Region | Subjects enrollment | Application | Clinical efficacy | Reference |
|---|---|---|---|---|---|---|---|---|
| Alder et al. | 2023 | Non-invasive BCI | EEG | New Zealand | N = 6 | To develop a wearable technology to augment locomotor stroke rehabilitation. | Participants perceived and experienced the excite BCI intervention positively. | [113] |
| Al-Qazzaz et al. | 2023 | Non-invasive BCI | EEG | Iraq | NA | EEG signal complexity measurements to enhance BCI-based stroke rehabilitation. | Laplacian eigenmap (LE) with random forest (RF) and k-nearest neighbors (KNN) obtained 74.48% and 73.20% accuracy, respectively. | [114] |
| Cantillo-Negrete et al. | 2023 | Non-invasive BCI | NA | Mexico | NA | A practical guide to explain BCI intervention parameters and technical challenges in clinical settings for stroke rehabilitation. | Presented a comprehensive methodology for implementing BCI-based upper extremity therapy in stroke patients. | [115] |
| Cipriani et al. | 2023 | Non-invasive BCI | EEG | Italy | NA | To investigate the promotoer, a BCI-assisted intervention to promote upper limb motor recovery. | Presented a statistical analysis plan for a randomized controlled trial on the short- and long-term efficacy in post-stroke motor recovery. | [116] |
| Fateeva et al. | 2023 | Non-invasive BCI | P300 | Russian | N = 30 | To improve cognitive function in post-stroke patients. | Reported an increase in the mean score of the MoCA « attention » domain in the BCI group. | [117] |
| Fu et al. | 2023 | Non-invasive BCI | EEG | China | N = 66 | To explore the efficacy of functional-oriented, portable BCI training for hand motor recovery. | The progress of FMA–UE between the BCI group and the control group was significantly different. | [118] |
| Jadavji et al. | 2023 | Non-invasive BCI | EEG | Canada | N = 13 | To test the feasibility of BCI-based functional electrical stimulation in children with hemiparesis. | Mean classification accuracies were for training and for rehabilitation. | [119] |
| Jia et al. | 2023 | Non-invasive BCI | EEG | China | N = 13 | To put forward tailored BCI training based on neural reorganization to improve rehabilitation efficacy. | Personalized BCI training tends to connect the potentially reorganized brain areas with event-contingent proprioceptive feedback. | [120] |
| Jiang et al. | 2023 | Non-invasive BCI | EEG | China | N = 1 | To explore the efficacy of non-invasive BCIs in patients with moderate and severe upper limb dysfunction. | The results before and after treatment were FMA–UE = 12/14 in a 36-year-old male stroke patient. | [121] |
| Li et al. | 2023 | Non-invasive BCI | EEG | China | NA | To establish an approach to developing a hybrid EEG–EMG for stroke patients. | A classification accuracy of 88.89% was achieved on an EEG–EMG synchronized dataset for push and pull movements. | [122] |
| Li et al. | 2023 | Non-invasive BCI | NA | China | NA | A bibliometric visualization analysis to investigate research hotspots and trends in BCI technology in stroke. | China and the United States were high-productivity countries in BCI. | [123] |
| Liao et al. | 2023 | Non-invasive BCI | EEG | China | N = 40 | To explore the efficacy of BCI in promoting upper limb function. | Better functional outcome was observed after MI–BCI rehabilitation in FMA scores. | [124] |
| Lima et al. | 2023 | Non-invasive BCI | EEG | Brazil | N = 1 | To explore the effect of transcranial direct current stimulation combined with BCI in chronic stroke. | Positive trends in the motor function, coordination, and speed of the affected limb, as well as sensory improvements. | [125] |
| Liu et al. | 2023 | Non-invasive BCI | EEG | China | N = 60 | To explore the efficacy of MI–BCI in enhancing upper limb motor function and attention. | The FMA–UE score, alert network response time, orienting network response, and corrects number were increased in the BCI group. | [126] |
| Ma et al. | 2023 | Non-invasive BCI | EEG | China | N = 46 | To explore the efficacy of BCI intervention in upper limb recovery after stroke. | The FMA–UE score of the BCI group was significantly higher than that of the control group after treatment (p = 0.035). | [127] |
| Mang et al. | 2023 | Non-invasive BCI | NA | China | NA | A review to explain the cognitive-motor process in closed-loop BCI-mediated motor function recovery. | The recent progress in BCI-mediated post-stroke motor function recovery involved with cognitive aspect. | [128] |
| Pichiorri et al. | 2023 | Non-invasive BCI | EEG | Italy | N = 12 | To test the high-density corticomuscular coherence networks capturing motor abnormalities during simple hand movements. | Between-group analysis showed that the corticomuscular coherence (CMC) weight of the whole brain network was significantly reduced in patients during affected hand movements. | [129] |
| Ramirez-Nava et al. | 2023 | Non-invasive BCI | P300 | Mexico | N = 14 | To promote upper-limb motor function recovery. | Statistically significant differences between groups were reported for FMA (p = 0.012), ARAT (p < 0.001), and FIM (p = 0.025) scales. | [130] |
| Rustamov et al. | 2023 | Non-invasive BCI | EEG | USA | N = 30 | To investigate the effectiveness of the IpsiHand system, a contralesionally controlled BCI therapy for chronic stroke. | Chronic stroke patients achieved significant improvement in both proximal and distal upper extremity motor function. | [131] |
| Sebastián-Romagosa et al. | 2023 | Non-invasive BCI | EEG | Spain | N = 25 | To improve gait in chronic stroke patients. | A clinically significant increase in walking speed of 0.19 m/s. | [132] |
| Shou et al. | 2023 | Non-invasive BCI | NA | China | NA | A meta analysis to investigate the effectiveness of verum versus sham BCI on upper limb function recovery. | There were significant differences in the Fugl-meyer assessment for upper extremity (mean difference [MD] = 4.78, 95%; confidence interval [CI] [1.90, 7.65], I2 = 0%, p = 0.001); and modified barthel index (MD = 7.37, 95% CI [1.89, 12.84], I2 = 19%, p = 0.008). | [133] |
| Wang et al. | 2023 | Non-invasive BCI | EEG | China | N = 128 | To improve the comprehensive ability and quality of life. | At 6 months, the BCI group showed statistically significant improvements in limb motor function, mindful attention awareness, etc. | [134] |
| Zanona et al. | 2023 | Non-invasive BCI | EEG | Brazil | N = 40 | To explore the effect of BCI combined with mental practice and occupational therapy in promoting upper limb recovery and functional independence in subacute stroke patients. | Clinical effects were found in FIM (p = 0.001, d = 0.56); MAL-AOM (p = 0.001, d = 0.83); MAL-QOM (p = 0.006, d = 0.84); BBT (p = 0.004, d = 0.40); and JHFT (p = 0.001, d = 0.45). (Abbr: Functional independence measure [FIM]; motor activity log [MAL]; amount of use [MAL-AOM]; quality of movement [MAL-QOM]; the box and blocks test [BBT]; and the Jebsen hand functional test [JHFT]). | [135] |
| Zhang et al. | 2023 | Non-invasive BCI | EEG | China | N = 33 | To propose an adaptive BCI system for enhancing motor recovery. | The increase in FMA–UE scores in the BCI group was significantly greater (BCI: 9.8 ± 5.3; control: 4.3 ± 3.3; p = 0.003), and more likely to reach a MCID (BCI: 11/17 vs. control: 4/16; odds ratio: 5.5, 95% confidence interval = 1.2–24.8, p = 0.03). | [136] |
| Isaev et al. | 2024 | Non-invasive BCI | NIRS | Russia | N = 15 | To develop a multiple-session dataset of NIRS recordings in controlling BCI. | The BCI was controlled by imagined hand movements; visual feedback was presented based on the real-time data classification results. | [137] |
| Seta et al. | 2024 | Non-invasive BCI | EEG/EMG | Switzerland | N = 13 | To utilize brain- and muscle-derived features in discriminating simple hand motor tasks for a rehabilitative BCI. | The ERD/ERS and the CMC-based classification showed similar temporal evolutions of performance, with a significant increase in accuracy during the execution phase. | [138] |
| Ren et al. | 2024 | Non-invasive BCI | EEG | China | NA | A meta analysis to explore the effect of BCI-controlled functional electrical stimulation training on upper limb rehabilitation. | BCI–FES had significant immediate effects on upper limb function in subacute and chronic stroke patients, but evidence for its long-term impact remains limited. | [139] |
| Rustamov et al. | 2024 | Non-invasive BCI | EEG | USA | N = 26 | To investigate the effectiveness of contralesionally controlled BCI therapy in chronic stroke patients. | Chronic stroke patients achieved significant motor improvement in both proximal and distal upper extremity with BCI therapy. | [140] |
| Miladinović et al. | 2024 | Non-invasive BCI | EEG | Italy | N = 6 | To assess how time window duration affects performance, specifically classification accuracy and the false positive rate. | The results suggest an optimal time window of 1–2 s. | [141] |
| Luo et al. | 2024 | Non-invasive BCI | EEG | China | N = 64 | To assess the impact of BCI rehabilitation on lower limb motor dysfunction in individuals with acute ischemic stroke. | After 20 sessions of treatment, both groups improved in motor function, walking function, and activities of daily living. | [142] |
| Savić et al. | 2024 | Non-invasive BCI | Somatosensory event-related potentials (sERPs) | Serbia | N = 10 | To investigate the feasibility of a novel BCI device designed for sensory training. | Using a single electroencephalographic channel, attention classification accuracy ranged from 70% to 100% across all patients. | [143] |
| Kim et al. | 2024 | Non-invasive BCI | EEG | Korea | NA | A protocol to compare the effects of real and sham BCI on motor function and brain activity among patients with subacute stroke with weak wrist extensor strength. | Results have not been reported. | [144] |
| Krueger et al. | 2024 | Non-invasive BCI | EEG | Germany | N = 20 | To test the hypothesis that heightened neural plasticity earlier in the post-stroke period would further enhance corticomuscular functional connectivity and motor recovery. | The BCI group showed greater: FMA–UE improvement; motor-evoked potential amplitude; etc. | [145] |
| Qiu et al. | 2024 | Non-invasive BCI | EEG | China | NA | To discuss the impact of personalized BCI–VR rehabilitation programs on the recovery of motor functions in patients with stroke-induced hemiplegia. | The potential for this technology to transform rehabilitation, enhance patient outcomes, and contribute to the broader medical field remains vast and promising. | [146] |
| Su et al. | 2024 | Non-invasive BCI | EEG | China | N = 6 | To put forward an adaptive hybrid BCI for hand function rehabilitation. | The four-class gesture recognition accuracies of healthy individuals and patients improved to 94.37 ± 4.77% and 79.38 ± 6.26%, respectively. | [147] |
| Kleih et al. | 2024 | Non-invasive BCI | EEG | Germany | N = 7 | To evaluate the efficacy of visual P300 BCI use in supporting rehabilitation of chronic language production deficits. | All participants showed an improvement in spontaneous speech production. | [148] |
| Biswas et al. | 2024 | Non-invasive BCI | EEG | USA | NA | A protocol to describe the BCI–FES clinical trial examining the safety and efficacy of this system compared with conventional physical therapy. | Results have not been reported. | [149] |
| Gangadharan et al. | 2024 | Non-invasive BCI | EEG | India | N = 4 | To investigate the feasibility of MI–BCI in upper limb rehabilitation of chronic stroke survivors. | All participants successfully controlled the position of the virtual marble using sensorimotor rhythm. | [150] |
| Ma et al. | 2024 | Non-invasive BCI | EEG | China | N = 46 | The effects of MI-based BCI rehabilitation programs on upper extremity hand function in patients with chronic hemiplegia. | The increase in the FMA–UE score showed a positive correlation with the mean zALFF of the contralateral precentral gyrus (r = 0.425, p < 0.05), and the mean zReHo of the right cuneus (r = 0.399, p < 0.05). | [151] |
| Zhang et al. | 2024 | Non-invasive BCI | EEG | China | N = 9 | To utilize MI and high-frequency steady-state visual evoked potential (SSVEP) in constructing a hybrid BCI-controlled soft robotic glove. | 12 healthy subjects and 9 stroke patients achieved accuracy rates of 95.83 ± 6.83% and 63.33 ± 10.38, respectively. | [152] |
| Wang et al. | 2024 | Non-invasive BCI | EEG | China | N = 296 | A multicenter study to investigate the efficacy and safety of BCI in rehabilitation training on upper limb motor function among ischemic stroke. | The primary efficacy outcomes of the FMA–UE score change from baseline to 1 month were 13.17 (95% confidence interval [CI], 11.56, 14.79) in the BCI group and 9.83 (95% CI, 8.19, 11.47) in the control group (p = 0.0045). | [153] |
| Zhang et al. | 2024 | Non-invasive BCI | EEG | China | N = 20 | Performance of the action observation (AO)-based BCI and gaze metrics analysis. | The designed AO-based BCI could simultaneously induce steady-state motion visual evoked potential (SSMVEP) from the occipital region and sensory motor rhythm from the sensorimotor region. | [154] |
| Liu et al. | 2024 | Non-invasive BCI | EEG | China | N = 50 | An EEG motor imagery dataset for BCI in acute stroke patients. | The first open dataset addressing left- and right-handed motor imagery in acute stroke patients. | [155] |
| Brunner et al. | 2024 | Non-invasive BCI | EEG | Denmark | N = 40 | To investigate whether BCI-based training, combining motor imagery with FES targeting finger/wrist extensors, is more effective in improving severely impaired UL motor function than conventional therapy in the subacute phase after stroke. | Few patients (10/35) improved above the minimally clinically important difference of 6 points on ARAT, 5/15 in the BCI group, and 5/20 in the control group. In the logistic regression, only CST integrity was a significant predictor for improving UL motor function, p = 0.007. | [156] |
| Nagarajan et al. | 2024 | Non-invasive BCI | EEG | Singapore | N = 71 | Transferring a deep learning model from healthy subjects to stroke patients in motor imagery BCI. | An average MI detection accuracy of 71.15% ( ± 12.46%) was achieved across 71 stroke patients. | [157] |
| Qu et al. | 2024 | Non-invasive BCI | EEG | China | NA | A meta analysis to explore the clinical effects of BCI with robot on upper-limb function. | The BCI–robot systems had a significant improvement on motor function recovery. | [158] |
- Abbreviations: AO, action observation; ARAT, Action Research Arm Test; BBT, blocks test; BCIs, Brain–computer interfaces; BCI-VR, Brain-computer interface-Virtual reality; CI, confidence interval; CMC, coherence; CST, corticospinal tract; EEG-EMG, electroencephalogram-electromyography; EMG, electromyography; ERD, event-related desynchronization; ERS, event-related synchronization; FIM, Functional independence measure; FMA-UE, Fugl-Meyer assessment of upper extremity; JHFT, Jebsen hand functional test; KNN, k-nearest neighbors; LE, Laplacian eigenmap; MAL, motor activity log; MAL-AOM, MAL-amount of use; MAL-QOM, MAL-quality of movement; MCID, minimal clinically important difference; MI, motor imagery; RF, random forest; SSMVEP, steady-state motion visual evoked potential; EEG, electroencephalography; MD, mean difference; SSVEP, steady-state visual evoked potential.
| Diseases and dysfunction | Researcher | Year | BCI types | BCI signals | Country/Region | Subjects enrollment | Application | Clinical efficacy | Reference |
|---|---|---|---|---|---|---|---|---|---|
| SCI | Lorach et al. | 2023 | Invasive BCI | ECoG | Switzerland | N = 1 | A brain–spine interface (BSI) to help a patient with SCI regain the ability to walk. | The BSI enabled natural control over the movements of his legs while standing, walking, climbing stairs, and traversing complex terrains. | [84] |
| SCI | Zou et al. | 2023 | Invasive BCI | ECoG | China | NA | A review to highlight the application of the BSI for SCI patients. | The BSI can help SCI patients regain walking ability. | [159] |
| SCI | Quiles et al. | 2023 | Non-invasive BCI | EEG | Spain | N = 2 | BCI based on transfer learning to detect the appearance of obstacles during exoskeleton-assisted walking in SCI patients. | The false positives per minute (FP/min) decreased from 31.8 to 3.9 FP/min, and the number of repetitions improved from 34.9% to 60.3% NOFP/TP. | [160] |
| SCI | Colamarino et al. | 2023 | Non-invasive BCI | EEG | Italy | N = 30 | To test the efficacy of MI–BCI in improving hand sensorimotor function outcomes in patients with traumatic cervical SCI. | N/A | [161] |
| SCI | De miguel-Rubio et al. | 2023 | Non-invasive BCI | EEG | Spain | NA | A systematic review to evaluate the effectiveness of the combined use of virtual reality and the brain–machine interface in the treatment of SCI. | Statistically significant changes were found in the upper limb, involving improvements in shoulder and upper arm mobility, and weaker muscles were strengthened. | [162] |
| Depression | Lai et al. | 2023 | Invasive BCI | DBS/electric fields | China | N = 10 | To investigate the functional and structural connectivities related to and predictive of the clinical effectiveness of DBS at the ventral capsule/ventral striatum region for treatment-resistant depression (TRD). | Successful prediction of antidepressant effectiveness in out-of-sample patients was achieved through the optimal connectivity profiles constructed with both the functional connectivity (R = 0.49 at p < 10−4) and structural connectivity (R = 0.51 at p < 10−5). | [163] |
| Depression | Fang et al. | 2023 | Non-invasive BCI | DBS/electric fields | USA/China | NA | A closed-loop BCI system of predictive neuromodulation for managing treatment-resistant major depression. | The dynamic model accurately predicted nonlinear and multiband neural activity. | [164] |
| Tumor | Prinsloo et al. | 2023 | Non-invasive BCI | EEG | USA | N = 91 | Using BCI to relieve chronic chemotherapy-induced peripheral neuropathy. | The BCI and placebo groups reported significant symptom reduction; at 1 month, symptoms continued to improve in the BCI group only. | [165] |
| Tumor | Fink et al. | 2023 | Non-invasive BCI | EEG | Germany | N = 56 | To test the effectiveness of an alpha and theta neurofeedback training protocol as an established psycho-oncological treatment. | Affective symptoms of distress (p ≤ 0.01), depression (p ≤ 0.05), and generalized anxiety (p ≤ 0.05) decreased significantly over time. | [166] |
| ALS | Luo et al. | 2023 | Invasive BCI | ECoG | USA | N = 1 | To test the reliability of a chronically implanted BCI over long time periods with only initial model training and calibration. | Speech commands were accurately detected and decoded (median accuracy: 90.59%) throughout the 3-month study period. | [82] |
| ADHD | Cervantes et al. | 2023 | Non-invasive BCI | NA | Mexico | NA | A systematic review on the social robots and BCI video games for managing ADHD. | 32 systems focused on supporting cognitive or behavioral rehabilitation therapies, and 2 systems focused on supporting the study of brain areas of people living with ADHD. | [167] |
| Autism | LaMarca et al. | 2023 | Non-invasive BCI | EEG | USA | N = 7 | To test the feasibility of BCI training of mu EEG rhythms in children with autism and intellectual impairments. | Learners demonstrated behavioral improvements and show evidence of a short-term increase in mu suppression relative to non-learners. | [168] |
| Sleep disorder | Chen et al. | 2023 | Non-invasive BCI | EEG | China | NA | To report a paradigm shift in the realm of sleep assessment, offering a host of advantages and distinctions when compared to traditional polysomnography (PSG). | Ongoing advancements in BCIs involve the integration of artificial intelligence algorithms to enhance the accuracy of sleep stage classification and the identification of sleep disorders. | [169] |
| Walking disabilities | Siribunyaphat et al. | 2023 | Non-invasive BCI | SSVEP | Thailand | NA | This study utilized a quick-response code visual stimulus pattern for wheelchair control. | The proposed SSVEP method and algorithm yielded an average classification accuracy of 92%. | [170] |
| SCI | Feng et al. | 2024 | NA | NA | China | NA | A bibliometric study to investigate the current status and emerging trends in BCI implementation for treating SCI. | Existing research focuses on the application of BCI for improving rehabilitation and quality of life of patients with SCI. | [171] |
| SCI | Blanco-Diaz et al. | 2024 | Non-invasive BCI | EEG | Brazil | N = 7 | To propose an EEG gait imagery-based BCI for promoting motor recovery on the lokomat platform. | The developed BCI reached an average classification accuracy of 74.4%. | [172] |
| SCI | Pais-Vieira et al. | 2024 | Non-invasive BCI | EEG | Portugal | N = 1 | A case study exploring the activation of a rhythmic lower limb movement pattern during the use of a multimodal BCI. | Changes in pain were encoded in the theta frequency band by the left frontal electrode F3. | [173] |
| SCI | Levett et al. | 2024 | Invasive BCI | NA | Canada | NA | A systematic review on invasive closed-loop BCI technologies for the treatment of SCI in humans. | Ten patients (47.6%) underwent non-invasive FES with a cuff; one (4.8%) had invasive FES with percutaneous stimulation, and ten (47.6%) used an external device. | [174] |
| ALS | Leinders et al. | 2024 | Invasive BCI | ECoG | Netherlands | N = 1 | To explore the nocturnal function of implanted BCI in patients with late-stage ALS. | The developed nightmode decoder functioned error free 79% of nights over a period of ± 1.5 years. | [175] |
| ALS | Angrick et al. | 2024 | Invasive BCI | ECoG | USA | N = 1 | To report online synthesis of intelligible words using a chronically implanted BCI in a man with impaired articulation due to ALS. | Evaluation of the intelligibility of the synthesized speech indicated that 80% of the words could be correctly recognized by human listeners. | [176] |
| ALS | Pirasteh et al. | 2024 | Non-invasive BCI | EEG | Iran | NA | A review to introduce and review FFT, WPD, CSP, CSSP, CSP, and GC feature extraction methods in EEG-based BCI for advanced stage ALS. | Using BCI, disabled patients could communicate with their caregivers and control their environment using various devices. | [177] |
| ALS | Wyse-Sookoo et al. | 2024 | Invasive BCI | ECoG | USA | N = 1 | A year-long longitudinal study to test the stability of ECoG high gamma signals during speech. | Speech-related ECoG signal responses were stable over a range of syllables activating different articulators for the first year after implantation. | [178] |
| Tumor | Krawutschke et al. | 2024 | Non-invasive BCI | EEG | Germany | N = 18 | To investigate the emotional arousal and valence effects on the event-related P300 in a visual oddball paradigm. | P300 amplitudes decreased significantly (p < 0.05) from pre to post therapy. Patients achieved significant relief from depressive symptoms (p < 0.05). | [179] |
| Tumor | Fink et al. | 2024 | Non-invasive BCI | EEG | Germany | N = 42 | To investigate the efficacy of mindfulness and neurofeedback (NF) on early posterior negativity (EPN) in oncology patients. | Descriptive statistics showed increased EPN for negative stimuli after NF intervention, while EPN for positive stimuli only slightly increased. | [180] |
| Tumor | Rolbiecki et al. | 2024 | Non-invasive BCI | EEG | USA | N = 20 | A feasibility trial exploring virtual reality and neurofeedback as a supportive approach to managing cancer symptoms in patients receiving treatment. | Patients showed improvements in anxiety and pain. | [181] |
- Abbreviations: ADHD, attention deficit hyperactivity disorder; ALS, amyotrophic lateral sclerosis; BCIs, Brain–computer interfaces; BSI, brain–spine interface; CSP, Common Spatial Pattern; CSSP, common spatio-spectral pattern; ECoG, electrocorticography; EEG, electroencephalography; EPN, early posterior negativity; FFT, Fast Fourier Transform; GC, causality of Granger; MI, motor imagery; NF, neurofeedback; NOFP, no false positives and true positives; PSG, polysomnography; SCI, spinal cord injury; SSVEP, steady-state visual evoked potential; TP, true positives; TRD, treatment-resistant depression; WPD, Wavelet Packet Decomposition.
4.1 The clinical application of invasive BCIs
The clinical management of several conditions, including SCI, ALS, depression, stroke and tumors, may benefit from invasive BCIs.
In 2023−2024, invasive BCI made a breakthrough in promoting lower limb motor function in SCI patients. A study84 published in Nature reported that a SCI patient with chronic tetraplegia because of an accident 10 years prior, was able to walk independently with a mobility aid. This is a significant advancement in SCI management, enhancing motor function and facilitating the performance of daily life activities. The success of this cutting-edge application is deeply rooted in the ongoing collaboration and extensive research conducted by neurosurgeons and neuroengineers. Another study82 reported that an ALS patient with severe dysarthria could operate computer applications with six intuitive speech commands through an ECoG implant over the ventral sensorimotor cortex. The 3-month study duration demonstrates that chronically implanted ECoG-based speech BCIs can reliably control devices for assistive use over long periods of time. This improvement was achieved with initial model training and calibration only, suggesting that it can be used unassisted at home. Another breakthrough in invasive BCI technology is a novel device implemented in a patient with treatment-refractory depression. The prediction value of structural and functional correlates of the response at the ventral capsule/ventral striatum region from DBS was explored.163 Based on this, by simply switching to the implanted device, the patient was able to control his negative emotions. This brings promise for managing patients with treatment-refractory depression.
In addition to SCI and depression, progress79 in stroke disease was reported in a study published on August 23, 2023, in Nature. A patient who suffered a stroke nearly 18 years prior could communicate using a BCI system decoding brain signals into text at a speed of about 80 WPM. This was a substantial improvement on a previous communication device with a speed of just 14 WPM.
BCIs may also be applied as interventions in dysfunction resulting from tumors. Compared with mindfulness-based therapy, EEG biofeedback therapy can improve affective symptoms, quality of life, and self-efficacy.166 Additionally, chemotherapy-induced peripheral neuropathy in breast cancer survivors could be relieved through closed-loop BCI by modifying the electroencephalographic signals.165
4.2 The clinical application of non-invasive BCI
In addition to the utility of invasive BCI in clinical medicine, various diseases and conditions benefit from the progress in non-invasive BCI, including motor dysfunction in stroke, SCI, ADHD, autism, sleep disorders, wheelchair control in people with physical disabilities, and more.
Non-invasive BCIs in stroke, such as EEG–BCI and fNIRS–BCI, are constantly being explored in scientific research and clinical practice. Experimental paradigms and motor tasks are important areas of focus for BCI systems in stroke rehabilitation. In 2023, a new system118 combining functional-oriented tasks with EEG–BCI training was proposed and applied during a BCI training in stroke hand function rehabilitation. This team, consisting of engineers, rehabilitation physicians, and therapists, is well-suited to support the development and optimization of BCI systems, contribute to product research, and promote their clinical application, demonstrating a practical collaboration between medicine and engineering. FNIRS–BCI was also increasingly explored. High-density fNIRS–BCI for motor-task classification,182 correlation-filter-based channel and feature selection framework hybrid BCI applications,183 and fully data-driven hybrid deep learning to improve the classification accuracy of volitional control fNIRS-BCI184 were proposed by different scholars. In 2024, substantial progress in BCI stroke rehabilitation was made by a multicenter study (sample size: 296)153 which investigated the efficacy and safety of BCI training for the rehabilitation of upper limb motor function in ischemic stroke patients.
In the same year, noninvasive spinal cord electrical stimulation185 was applied for arm and hand function improvement in chronic cervical SCI, demonstrating its safety and efficacy.
ADHD is a neural disorder with symptoms of inattention, hyperactivity, and impulsivity. Many young people are affected by this disorder. BCI-based serious video games can be used as cognitive and neurofeedback training (NFT) for ADHD children.167
In children with autism spectrum disorder, the NFT of mu rhythms is used to improve behavior and EEG mu rhythm suppression during action observation. However, children with intellectual impairments were excluded. The mu EEG rhythms–BCI was applied in children with autism and intellectual impairments.168 Future studies on this type of BCI should focus on improving the identification of ideal mu-NFT candidates, improving the BCI task accuracy and outlining the rehabilitation mechanisms of the protocols.
Sleep disorders present formidable challenges to the overall well-being of individuals. To address these issues, a groundbreaking solution has emerged in the form of flexible and lightweight BCI devices designed to monitor and regulate sleep.169 This innovative technology signifies a paradigm shift in the realm of sleep assessment, offering a host of advantages and distinctions when compared with traditional polysomnography (PSG). Unlike conventional PSG, that necessitates the attachment of numerous electrodes and sensors to the body, flexible BCIs prioritize user comfort; their lightweight and unobtrusive design promotes a more natural sleep experience, mitigating the impact of external devices on sleep quality. Another study186 demonstrated reliable monitoring capabilities of a wearable flexible device comparable to PSG. Ongoing advancements in BCIs involve the integration of artificial intelligence algorithms to enhance the accuracy of sleep stage classification and the identification of sleep disorders. Future iterations may see flexible BCIs incorporating wireless connectivity, facilitating seamless data transmission and remote monitoring by healthcare professionals.
Brain-controlled wheelchairs for persons with severe physical disabilities have been in development for several years. A recent study170 set up a robust BCI system using a quick-response code visual stimulus pattern. This proposed steady-state visual evoked potential (SSVEP) BCI using a quick-response code pattern can be used for wheelchair control. However, the problem of visual fatigue due to long-time continuous control still awaits a solution.
4.3 Feedback and signals in clinical application
A meta-analysis187 outlined several kinds of BCI types which can be applied in neurorehabilitation. From the perspective of application forms and feedback, in addition to classic BCI types (which include exoskeleton-based BCI, functional electrical stimulation-based BCI, and virtual reality technology-based BCI), there are two kinds of new BCI types—smell-based BCI and electrohaptics-based BCI. Olfactory dysfunctions such as anosmia and hyposmia, and mild cognitive impairment are common in some clinics. The smell-based BCI may be used to improve olfactory dysfunction.188 Electrohaptics-based BCI can be used as a restorative to improve sensory function after a stroke or brain injury. It can also be applied in communication as an assistive BCI in severely disabled users.189 In the clinical application of BCI signals, SSVEP, EEG, ECoG, P300 and other signals are increasingly more mature. The combined application of EEG and fMRI,190 and EEG and fNIRS191 promotes the greater use of neurofeedback BCI.
5 SUMMARY AND FUTURE RESEARCH
Numerous advances were made in BCI technology in 2023–2024. With the development of BCI hardware and software, and the in-depth investigations into the basic medicine, other brilliant BCI inventions will be implemented in clinical settings, particularly as both invasive and non-invasive BCIs widen the scope for clinical application. Future advancements in BCI implementation require addressing clinical challenges such as regulatory hurdles, patient acceptance, and long-term efficacy. Ethical considerations, particularly regarding invasive BCIs, must also be carefully examined to ensure responsible development and adoption. Through multidisciplinary collaboration among engineers, neuroscientists, and healthcare professionals, BCI technologies hold immense promise for advancing disease management, enhancing human bodily functions, and improving the overall quality of life.
AUTHOR CONTRIBUTIONS
Shugeng Chen: Writing—original draft and editing. Mingyi Chen: Writing—original draft and editing. Xu Wang: Writing—review and editing. Xiuyun Liu: Writing—review and editing. Bing Liu: Writing—original draft and editing. Dong Ming: Writing—review and editing. All authors have read and approved the final article.
ACKNOWLEDGMENTS
This work was funded by Outstanding Young Program of the Ministry of Education of the People's Republic of China (0401260003), National Science Fund for Excellent Overseas Scholars (0401260011), National Natural Science Foundation of China (82202798, 62174109, 82472098, and 32300704), Shanghai Sailing Program (22YF1404200), Tianjin Natural Science Foundation Outstanding Youth Project (24JCJQJC00250) and Major Science and Technology Special Projects and Engineering - Major Project of National Key Laboratories (24ZXZSSS00510).
CONFLICT OF INTEREST STATEMENT
The authors declare no conflicts of interest.
ETHICS STATEMENT
Ethics approval was not needed in this study.
Open Research
DATA AVAILABILITY STATEMENT
Data sharing is not applicable to this article as no new data was created or analyzed in this study.
