附录1_脑电磁与脑机接口

Appendix 1: Overview of Brain Neural Electrical Activity Signal Acquisition Technologies

From Basic Principles to Current BCI Applications and Consciousness Decoding Challenges

Author: AI Rider Jueju + Grok
Date: 2026-06-30

Abstract

The core of neural activity is the electrophysiological process generated by neurons via ionic currents, accompanied by electric and magnetic field signals. These signals provide an observable window for brain-computer interfaces (BCI) and neuroscience. Non-invasive technologies (such as EEG and MEG) and invasive technologies (such as intracortical electrode arrays) are highly consistent in their signal acquisition principles, but they exhibit significant differences in signal fidelity, spatial resolution, and information bandwidth. This paper reviews the basic mechanisms of brain neural electrical activity, major acquisition technologies, current clinical and research applications, a comparison of the pros and cons between invasive and non-invasive methods, and the fundamental limitations of current technologies in decoding deep consciousness. Existing technologies can reliably extract motor intentions and speech planning signals, whereas direct access to subjective conscious experiences still faces major challenges. Future developments rely on signal quality improvement, AI decoding optimization, and ethical considerations.

1. Introduction

The human brain consists of approximately 86 billion neurons, processing information through electrochemical signals. Understanding and decoding these signals is not only a fundamental question in neuroscience but also crucial for developing technologies for assistive communication, motor recovery, and even cognitive enhancement. Since the commercialization of EEG in the mid-20th century, brain signal acquisition technology has advanced rapidly. In recent years, projects such as Meta's Brain2Qwerty have demonstrated the potential for non-invasive, real-time sentence decoding, while invasive systems such as Neuralink and Synchron have achieved mind-controlled devices in humans. This review integrates the electrophysiological basis, acquisition methods, current applications, and limitations, focusing on the reverse engineering "from neural electrical signals to intent/consciousness expression."

2. Fundamental Principles of Brain Neural Electrical Activity

Neural electrical signals originate from transmembrane ionic flow (mainly Na⁺, K⁺, Ca²⁺, Cl⁻). Key processes include:

Action Potential: An all-or-nothing rapid depolarization (~1 ms) responsible for long-distance conduction.
Postsynaptic Potentials (PSPs): Local graded potentials, which are the main contributors to population signals.
Synchrony and Volume Conduction: Individual neuronal signals are weak; what is detected is the current dipole formed by large-scale synchronized activity. These currents generate measurable electric fields (EEG) and magnetic fields (MEG).

Electromagnetic Concomitance: According to Maxwell's equations, the movement of electrical charges generates both electric and magnetic fields simultaneously. Non-invasive technologies measure these secondary fields outside the head, whereas invasive ones directly approach the primary currents.

Signal characteristics include multi-band oscillations (δ, θ, α, β, γ), reflecting different brain states. The inverse problem is the core challenge: reconstructing internal sources from external measurements, which requires anatomical modeling and constraints.

3. Classification and Principles of Brain Signal Acquisition Technologies

3.1 Non-invasive Technologies

EEG: Scalp electrodes measure voltage differences. Spatiotemporal resolution: temporal <1 ms, spatial ~7-10 mm. Advantages: portable, inexpensive; Disadvantages: distorted by volume conduction, susceptible to artifact interference.
MEG: Measures magnetic fields (SQUIDs sensors). Temporal resolution is the same as EEG, spatial resolution ~2-6 mm (especially for cortical tangential sources). Advantages: relatively "clean" signal, not distorted by the skull; Disadvantages: expensive equipment, requires a shielded room, low sensitivity to deep/radial sources. Emerging OPM-MEG is improving portability.

The two are complementary, with MEG performing better in decoding tasks such as Meta's Brain2Qwerty.

3.2 Invasive Technologies

ECoG (Cortical Surface): Subdural electrode grids with high spatial resolution (millimeter level).
Intracortical Microelectrode Arrays (Utah Array, Neuralink threads): Inserted into the cortex, providing single/multi-unit resolution. High channel count (hundreds to 1024+), supporting wireless transmission. Synchron's Stentrode uses an endovascular approach, reducing invasiveness.

Invasive methods directly record local field potentials (LFPs) and spikes, yielding an extremely high SNR.

4. Current Applications of the Technology

Non-invasive: Clinical epilepsy monitoring, BCI prototypes (e.g., Meta's Brain2Qwerty v2: MEG + AI achieving ~39% average WER sentence decoding). In research, used for cognitive neuroscience (perception, language production).
Invasive: Already entered human clinical trials. Neuralink has achieved mind-controlled cursors/devices; Synchron supports text entry; Blackrock platform has long been used for motor decoding. Applications focus on communication and motor recovery for patients with severe disabilities (ALS, tetraplegia, locked-in syndrome).

AI (especially Transformer and LLM fine-tuning) is a common accelerator, capable of mapping raw signals to semantic outputs.

5. Systematic Comparison between Invasive and Non-invasive Methods

Signal Quality: Invasive >> Non-invasive (SNR, bandwidth, resolution).
Risk and Accessibility: Non-invasive is safe, easy to use, and scalable; invasive involves surgical risk, inflammation, and long-term stability issues, but offers a higher performance ceiling.
Decoding Performance: Non-invasive is currently usable at the sentence level but with errors; invasive approaches natural speed and accuracy.
Development Trajectory: Non-invasive catches up relying on data scale and AI; invasive is constrained by biocompatibility.

6. Gaps and Challenges in Deep Consciousness Decoding

Current technologies generally lack direct discrimination of deep consciousness:

What is decoded is primarily observable motor/speech intentions (overt or imagined), rather than subjective experiences (qualia), abstract thought streams, or distributed whole-brain dynamics.
Reason: Signals primarily reflect local synchronized activity; consciousness involves large-scale, cross-network dynamics and internal states, which are difficult to fully capture with the current electrophysiological window.
Ethical issues: Privacy, boundaries of "mind reading", and informed consent. Although non-invasive methods carry low risks, large-scale deployment may still trigger concerns over data misuse.

Existing BCIs are far from achieving "mind uploading" or comprehensive thought reading, serving more as assistive tools.

7. Future Directions

Hybrid systems (non-invasive hardware + advanced AI).
Portable high-resolution MEG and flexible biocompatible electrodes.
Multimodal integration (electrophysiology + blood oxygenation + behavior).
Open datasets and ethical frameworks (such as Meta's Digital Brain Project).
Potential breakthroughs: If non-invasive performance continues to improve in a log-linear fashion, or if long-term stability of invasive methods is resolved, practical BCIs will benefit more people.

8. Conclusion

Originating from the same electrophysiological roots, brain signal acquisition technologies have formed a complementary ecosystem. Non-invasive methods provide a safe entry point, while invasive methods offer a high-performance pathway, with AI acting as the key amplifier. Despite substantial progress in fields like assistive communication, fundamental limitations remain in decoding deep consciousness. Future research must balance performance, risk, and ethics to drive the transition from "intent decoding" to a more comprehensive neuro-cognitive understanding.

References (Selected): Papers related to Meta Brain2Qwerty, public materials of Neuralink/Synchron, classic reviews on MEG/EEG, etc.

Disclaimer: This review is based on open scientific literature and the latest public progress (as of June 2026). It is intended for information sharing only and does not constitute medical advice. With the rapid development of technology, specific applications should refer to the latest clinical data.