Brain-Computer Interfaces: Medicine's Most Ambitious Frontier

• A brain-computer interface is a system that measures neural activity and converts it into digital commands for an external device.
• Non-invasive BCIs read signals through the scalp, while partially invasive and invasive BCIs place electrodes on or inside the brain to capture cleaner signals.
• In a 2024 study in the New England Journal of Medicine, an implanted speech BCI let a man with ALS communicate with 97.5% word accuracy.
• Implanted systems have allowed people with paralysis to control a robotic arm, turn attempted handwriting into text and walk again through a brain-spine interface.
• Surgical risks, scar tissue around electrodes and the privacy of neural data remain the main unresolved issues.
• Aegis Capital, a HealthTech & Longevity VC, backs early-stage startups in Central and Eastern Europe, including non-invasive technology that assesses the brain through eye movements.

What Is a Brain-Computer Interface?

A brain-computer interface (BCI) is a system that records brain signals, decodes the intention behind them and uses the result to control external devices without relying on muscles or peripheral nerves. It is a form of human-computer interaction in which the input comes directly from the nervous system. A BCI can let a person operate computers, move a cursor across a computer screen, use a mobile device or drive a prosthetic limb. For patients with severe disabilities caused by neurological disorders or injury, it offers a way to act on the environment when the pathways between the human brain and the body have been cut.

The US Food and Drug Administration, in its final guidance from May 2021, describes implanted BCI devices as neuroprostheses that interface with the central or peripheral nervous system to restore lost motor and sensory capabilities in patients with paralysis or amputation. That definition frames BCIs as medical devices with a therapeutic goal. Non-invasive headsets marketed for gaming or virtual reality fall outside it, and the distinction determines how each product is tested and regulated.

How a BCI System Turns Brain Activity Into Digital Commands

Every BCI system follows the same basic sequence, organized into three stages:

• Signal acquisition: electrodes or optical sensors record the brain's electrical activity or related changes in blood flow.
• Signal processing: software amplifies weak neural data, filters out noise and extracts the features linked to a specific intention.
• Decoding and output: an algorithm translates those features into digital commands that move a computer cursor, produce text or drive a robotic limb.

The user sees the result and adjusts, which closes a feedback loop between the brain and the device. The quality of the first stage sets the ceiling for the whole system, since a decoder can only work with the information the sensors capture.

Why Machine Learning Changed BCI Research

Decoding has become the fastest-moving part of the field. BCI research now draws on computational neuroscience, signal processing and artificial intelligence, and many recent systems use neural networks to map patterns of neural activity onto words or movements. In a 2021 study in Nature, a recurrent neural network decoded attempted handwriting from the motor cortex of a man whose hand was paralyzed by spinal cord injury and turned it into text at 90 characters per minute with 94.1% raw accuracy. With a general-purpose autocorrect applied offline, accuracy exceeded 99%.

Did you know: The handwriting BCI reached speeds comparable to typical smartphone typing in the participant's age group, which the authors put at 115 characters per minute. Earlier point-and-click BCIs, in which the user steers a cursor to select letters, were considerably slower.

Invasive vs. Non-Invasive BCIs

BCIs differ mainly in where the sensors sit, and that choice sets the balance between signal quality and surgical risks. Signals recorded through the skull are weaker and more distorted than signals recorded on the brain surface or inside brain tissue. The table below compares the three main approaches.

Non-Invasive BCIs: EEG and fNIRS

Electroencephalography (EEG) records electrical signals generated by large populations of neurons through electrodes placed on the scalp. EEG signals carry no surgical risk and are inexpensive to record, but the skull and scalp weaken and blur them, and muscle activity from blinking or jaw movement can contaminate the recording. Many non-invasive BCIs rely on motor imagery, in which the user imagines a movement and the system detects the related change in brain rhythms over the motor cortex.

Functional near-infrared spectroscopy (fNIRS) measures changes in blood oxygenation using light that passes through the skull, instead of measuring electrical activity. Blood flow follows neural activity with a delay of several seconds, so fNIRS responds more slowly than EEG. Both technologies fit into wearable devices, and that portability makes them a practical choice for rehabilitation and home use.

Partially Invasive and Invasive BCIs

Partially invasive BCIs place electrodes on the surface of the brain, beneath the skull but outside the cortex. A 2021 speech study in the New England Journal of Medicine used a high-density array of this kind, placed over the area of the sensorimotor cortex that controls speech. In April 2025, the FDA cleared a thin-film cortical array with 1,024 electrodes for temporary implantation of up to 30 days, a component of a wireless BCI system still in development.

Invasive BCIs insert microelectrodes into the cortex to record neuronal signals from individual cells or small groups of neurons. That precision has a biological cost: histological studies consistently show that a glial scar forms around implanted electrodes, and this scar tissue is associated with declining signal quality over time. The procedure also carries infection risk, so each implant has to justify its risk with a clear functional benefit for the patient.

Medical Applications: Restoring Communication and Movement

Most clinical BCI studies involve people with paralysis caused by spinal cord injury, brainstem stroke or ALS. The published results cover three kinds of lost function: speech, arm and hand control, and walking.

Speech and Text Communication

In the 2021 study in the New England Journal of Medicine, a man with anarthria after a brainstem stroke attempted to say words from a 50-word vocabulary while an implanted array recorded his cortical activity. The system decoded sentences in real time at a median rate of 15.2 words per minute, with a median word error rate of 25.6%, and the relevant signals stayed stable across the 81-week study. In 2024, the same research group reported in Nature Biomedical Engineering that the participant could switch between English and Spanish using a bilingual decoder.

Accuracy has since improved sharply. In the 2024 study in the New England Journal of Medicine, a man with ALS reached 99.6% word accuracy with a 50-word vocabulary on the first day of use, after 30 minutes of calibration. With a 125,000-word vocabulary, the system sustained 97.5% accuracy over 8.4 months and supported self-paced conversation at about 32 words per minute for more than 248 hours.

Robotic Arms, Prosthetic Devices and Sensory Feedback

Work on arm movement began in animals. In a 2008 study in Nature, two rhesus monkeys used signals from the motor cortex to control a multi-jointed prosthetic arm and feed themselves. In 2012, a study in the same journal showed that two people with long-standing tetraplegia could perform three-dimensional reaching movements and grasps with a robotic arm, using neural signals from a 96-channel microelectrode array, and one participant drank coffee from a bottle five years after implantation.

These neural prostheses lacked touch, which the body relies on for fine motor control. A 2021 study in Science added sensory feedback by stimulating the somatosensory cortex whenever the robotic hand made contact, supplying sensory signals the injured spinal cord no longer carried. With artificial touch added to vision, the participant's trial times on a standard upper-limb assessment fell by half, from a median of 20.9 to 10.2 seconds.

Spinal Cord Injury and Stroke Rehabilitation

BCIs can also help restore movement of the patient's own limbs. In a 2023 study in Nature, researchers connected cortical signals to epidural stimulation of the spinal cord, creating a digital bridge that let a man with chronic tetraplegia stand, walk and climb stairs. The system calibrated within minutes, stayed reliable for over a year including home use, and rehabilitation with it helped the participant walk with crutches even when it was switched off.

Non-invasive systems play a different role in stroke care, where they support rehabilitation more than direct device control. A 2026 systematic review and meta-analysis in the Journal of Medical Internet Research, covering 21 randomized controlled trials in chronic stroke, found better upper-limb motor function scores after BCI-based training than after control interventions, with the largest gains when BCIs were paired with functional electrical stimulation. The benefits did not persist significantly beyond the training period, so the field still has to show how to restore function for the long term.

Regulation, Safety and Ethics

Clinical results have raised practical questions that engineering alone cannot settle. Regulators need evidence of long-term safety, and lawmakers have started to define who controls the data a BCI records.

How Regulators Approach Implantable Devices

The FDA's 2021 guidance sets out recommendations for nonclinical testing and for the design of feasibility and pivotal studies of implanted BCIs. Long-term implanted BCIs for communication or movement remain investigational in the United States and reach patients only through approved clinical trials. The 2025 clearance of a cortical array for up to 30 days applies to temporary use, such as brain mapping and short BCI studies, not to permanent implantation.

Tip: Search ClinicalTrials.gov for recruiting BCI studies, and ask the study team how long the device stays implanted, what happens to it when the study ends, and who can access the recorded neural data.

Neural Data and Privacy

In April 2024, Colorado became the first US state to extend privacy protection explicitly to neural data, classifying it as sensitive data that companies may collect only with consent. The state legislature found that neural data can reveal intimate information about health, mental states, emotions and cognitive functioning. California followed with a similar amendment to its consumer privacy law, in force since January 1, 2025.

Informed consent is especially demanding in this field, since many BCI users have severe communication difficulties at the moment they enroll. Long-term effects of chronic implants on brain tissue are also still being studied, which makes ongoing consent and follow-up as important as the initial decision.

Reading the Brain Through the Eye

Not every neurotechnology needs to decode intention. Eye movements depend on several brain areas, from the visual cortex to the regions that plan and execute movement, and eye tracking is already used as an assistive input method for people with paralysis. The same measurement can serve diagnosis: changes in how the eyes move may reflect changes in the brain long before other symptoms appear, without any implant or surgery.

Example: Inoko Vision, a company backed by Aegis Capital, is developing NeuroFET, a non-invasive optical device that tracks retinal eye movements to assess the neurological state of the brain. By combining retinal imaging, eye tracking and proprietary data analysis, the technology targets early screening and monitoring of neurodegenerative diseases such as Alzheimer's and Parkinson's.

Why Neurotechnology Needs Specialized Early-Stage Capital

Implantable brain-computer interface technology requires years of preclinical testing, feasibility studies, long-term safety data and pivotal trials before any commercial use, and most of that cost comes before the first sale. Non-invasive tools face a shorter path, but they still need clinical validation, reliable hardware and a clear regulatory strategy. Early-stage investors with healthcare expertise help founders plan that path from the start and connect them with clinicians and research institutions.

Aegis Capital is a HealthTech & Longevity VC that invests in early-stage startups from Central and Eastern Europe, with AI and digital health and diagnostics among its core investment areas. The fund has a capitalization of PLN 80 million and offers an initial ticket of up to PLN 3 million, with total funding of up to PLN 8 million per company across follow-on rounds.

FAQ

What is a brain-computer interface used for?

In medicine, BCIs mainly help people with paralysis communicate and control external devices such as a computer cursor, a robotic arm or a spinal cord stimulator. Non-invasive systems are also used in stroke rehabilitation and in brain research.

Are brain-computer interfaces available to patients today?

Long-term implanted BCIs for communication or movement are available in the United States only through clinical trials. Non-invasive EEG systems are used more widely, mainly in rehabilitation and research settings.

How does a non-invasive BCI record brain activity?

A non-invasive BCI uses electrodes or optical sensors on the scalp to record brain activity without surgery. EEG measures electrical activity, while fNIRS measures changes in blood oxygenation linked to neural activity.

What are the main risks of invasive BCIs?

Invasive BCIs require brain surgery and carry a risk of infection. Over time, scar tissue can form around implanted electrodes and reduce signal quality.

Can a brain-computer interface read thoughts?

Current BCI devices decode specific, trained intentions such as attempted speech, handwriting or movement. Reading free-form thoughts remains science fiction, since decoders depend on training data from each user and work only within defined tasks.

Is neural data protected by law?

Protection is growing but uneven. Colorado and California classify neural data as sensitive personal data under their consumer privacy laws, while many other jurisdictions have no rules specific to it.

‍