Applications over the horizon – Advancements and challenges in brain-computer interfaces

The article introduces brain-computer interfaces (BCIs) as technologies that establish direct communication between the human brain and external devices, enabling control of robots, cursors, and interaction with virtual environments. By decoding neural signals into commands, BCIs can restore mobility and communication for paralyzed individuals and enhance human-computer interaction across medicine, gaming, and communication. It highlights notable recent progress, particularly two January 29, 2024 announcements: Neuralink’s Telepathy (United States) and the semi-invasive NEO device (China, Tsinghua University and Xuanwu Hospital), both reporting first wireless implantable BCI clinical human trials that have garnered significant attention.

The piece situates BCI within a spectrum of neural recording modalities, noting trade-offs among invasive (neuron spike detection, electrocorticography) and noninvasive (EEG, fMRI, fNIRS) approaches. It summarizes constraints of MEG, PET, and fMRI (cost, portability), and the metabolic basis of PET/fMRI/fNIRS (high spatial but low temporal resolution), contrasting them with EEG’s noninvasiveness but lower spatial resolution and SNR compared with ECoG and neuron spike detection. It references prior foundational and review works on BCIs for communication and rehabilitation, general BCI reviews, and Neuralink’s platform description (cited in the references).

This news article does not report original experimental methods but outlines BCI system components and approaches. A BCI system comprises the CNS user, brain signal acquisition, neural feedback, signal processing/decoding, a control interface, and peripheral devices. The pipeline: signal acquisition and processing, feature extraction, machine learning/pattern classification, decoding to control signals, and control of robots or cursors for assistive/rehabilitative applications. It details recording modalities: invasive (neuron spike detection—extracellular/intracellular, ECoG, endovascular) versus noninvasive (EEG, fMRI, NIRS). Neuralink’s invasive Telepathy targets extracellular neuron spike detection via implanted polymer threads/electrodes in cortex. The NEO system adopts a semi-invasive design with ECoG electrodes placed under the dura mater, an intracorporeal module embedded in the skull, and wireless power/communication to external devices; decoding is performed on connected computers/phones.

- Neuralink (Telepathy): First device demonstrated rapid implantation of 96 polymer threads with 32 electrodes each (total 3,072 electrodes), miniaturized custom electronics streaming full broadband electrophysiology from all electrodes, long-term implantable packaging, and low-latency spike-detection software, forming a prototype of a wireless human brain-machine interface. Following FDA approval for human trials (May 2022) and recruitment (September 2022), Musk announced the first human transplantation on January 29, 2023; 20 days post-implantation the volunteer had no adverse neurological effects and could control mouse movements by thought, and control phones/computers via neuron spike signals. Advantages include sensitive readout of neural representations at single-neuron resolution; concerns persist about user safety and long-term usability of invasive schemes. - Tsinghua University/Xuanwu Hospital (NEO): Longstanding work on BCI principles, algorithms, and translation. In 2022, achieved an equivalent information transfer rate of 62 bits/min in a BCI speller with epilepsy patients. On October 24, 2023, implanted a NEO device in a patient with 14-year quadriplegia from complete spinal cord injury; discharged 10 days post-surgery. After 3 months of home rehabilitation, the patient could drive a pneumatic glove via ECoG signals to perform brain-controlled tasks like drinking water; grasping decoding accuracy exceeded 90%. ASIA scores for spinal cord injury and sensory evoked potentials improved. The NEO utilizes a minimally invasive, wireless design with the intracorporeal module buried in the skull, dura-covered electrodes, external near-field wireless power and communication, and a processor about the size of two coins, enabling long-term use without internal batteries.

The article contrasts invasive and semi-invasive BCI strategies: invasive cortical implants (Telepathy) enable high-fidelity, single-neuron spike recordings for sensitive decoding but raise safety and long-term durability concerns; semi-invasive ECoG (NEO) offers improved safety and suitability for long-term use by placing electrodes under the dura and using wireless power/communication. The findings illustrate that both approaches can enable functional control (cursor/mouse movement, device operation, assistive grasping) and meaningful rehabilitation indicators in real-world/home settings. The discussion frames key obstacles that must be overcome to translate BCIs broadly: achieving long-term stability and biocompatibility of neural interfaces, advancing decoding algorithms for complex neural patterns to enable precise and intuitive control, and addressing ethical issues around privacy, data security, and informed consent. These considerations are central to making BCIs practical, safe, and trustworthy for patients and broader applications.

Recent clinical milestones from Neuralink’s invasive Telepathy and Tsinghua/Xuanwu’s semi-invasive NEO underscore rapid progress in neural interfaces, decoding, and real-time brain control of devices. While demonstrating tangible functional capabilities and early signs of rehabilitative benefit, continued research and collaboration are needed to improve interface longevity, decoding accuracy, and ethical frameworks. The field shows strong potential to transform healthcare, communication, gaming, and other industries, ushering in a new era of human-machine interaction.

As a news and perspective piece, the article does not present controlled experimental protocols or statistical analyses. It notes broader technical limitations and risks: degradation and potential tissue damage affecting long-term stability/reliability of implants; challenges in accurately decoding complex neural signals for precise, intuitive control; modality-specific constraints (e.g., noninvasive methods’ lower spatial resolution/SNR, metabolic imaging’s limited temporal resolution and lack of portability); and ethical issues of privacy, data security, and informed consent. Invasive approaches also present surgical risk and concerns over long-term usability.