A Brain-Computer Interface (BCI) is defined as a system where, when a user actively performs specific mental tasks or receives specific external stimuli, sensor technology acquires signals generated by the central nervous system. Features of brain signals representing or encoding the user's intent are directly converted into communication and control commands for interacting with a computer-centric machine system. The results of the interaction are then fed back to the user to regulate their mental activity strategies.

Based on the invasiveness of electrode placement, BCIs can be categorized into invasive, partially invasive, and non-invasive types. Non-invasive BCIs detect weak signals from the brain using electrodes attached to the scalp, allowing for safer monitoring of whole-brain signals. Currently, 80% of BCI companies in China have adopted the non-invasive acquisition technology route.

From a policy perspective, in March 2025, China's National Healthcare Security Administration (NHSA) issued the "Neurological System Medical Service Price Project Guidelines," which for the first time created a separate pricing item for BCI technology. Recognizing the need for constant equipment adjustment in non-invasive BCI, the guidelines added a new pricing item: "Non-invasive BCI Adaptation Fee."

1.2.1. Non-invasive BCI: Signal Acquisition

On the "brain-reading" side, limited by signal resolution, the marginal effectiveness of product development based on EEG signals is expected to decrease. Consequently, multi-modal brain imaging that integrates different signals—such as brain structure, brain blood oxygen, and brain magnetism—is becoming a research hotspot in both scientific research and clinical settings. Based on the type of biological signal detected, non-invasive BCIs can be divided into Electrophysiological Signal Activity-Dependent (ESAD) sensors and Blood Oxygen Level-Dependent (BOLD) sensors.

(I) ESAD Sensors: EEG and MEG

1. EEG (Electroencephalography)
   EEG uses electrodes placed on the scalp surface to non-invasively capture the potential differences generated by the electrical activity of brain neurons...
   *(Detailed content on EEG, MEG including SQUID-MEG and OPM-MEG, fNIRS, PAI, and multi-modal fusion follows in the original, covering technical principles, advantages, limitations, and examples like the BrainFusion framework from South China University of Technology, which achieved 95.5% classification accuracy in motor imagery tasks by fusing EEG and fNIRS.)*

1.2.2. Signal Preprocessing, Feature Extraction, and Decoding

Raw EEG signals are very weak and often accompanied by significant non-target signals or noise, requiring further processing and decoding.
First, preprocessing uses techniques like signal amplification, filtering, noise reduction, and transformation to eliminate non-cerebral interference and enhance target signal features...
(Further details cover decoding methods including Deep Learning like CNN/RNN/LSTM, Transfer Learning, Manifold Classification, Adaptive Learning, and EEG source analysis.)

On the "brain-writing" side, neuromodulation technology uses physical (electrical, magnetic, optical, ultrasonic, etc.) or chemical means to excite, inhibit, or modulate signal transduction in neurons or neural networks adjacent to or remote from the central, peripheral, and autonomic nervous systems, thereby improving patient neurological function. In 2023, the global neuromodulation device market exceeded $8billionandisexpectedtosurpass$16 billion by 2030, with a CAGR of 10-12%.

Non-invasive neuromodulation includes Transcranial Magnetic Stimulation (TMS), Transcranial Electrical Stimulation (TES), and Transcranial Ultrasound Stimulation (TUS). Globally, over three-quarters (76.2%) of clinical trials for neurological disorders use non-invasive methods. Among specific modalities, TMS (50%) and TES (24.4%) are the most commonly used.

2.2.1. Transcranial Magnetic Stimulation (TMS)
TMS places an insulated coil over the scalp to generate a magnetic field, applying electromagnetic pulses that change the membrane potential of cortical neurons, inducing action potentials and modulating neural networks...
(Details include FDA and NMPA approvals for depression, migraine, OCD; key players like Neuronetics (NeuroStar) and BrainsWay (Deep TMS); and challenges such as limited deep brain access and equipment portability.)

2.2.2. Transcranial Electrical Stimulation (TES) & 2.2.3. Transcranial Focused Ultrasound (tFUS)
(Explanations of TES (including tDCS, tACS, tRNS) and tFUS, including their evolving forms like HD-tES, TI-TES, technical challenges (e.g., skull impedance for ultrasound), and current regulatory status.)

2.3.1. Technical Principle of Closed-Loop Systems
Originating from engineering, open-loop vs. closed-loop control concepts are now used in neuroscience. In open-loop neuromodulation, intervention (stimulation or therapy) is applied without real-time feedback. Closed-loop systems monitor behavioral or physiological signals continuously and adjust intervention parameters in real-time based on feedback, aligning more closely with the brain's natural operating logic.

A closed-loop neuromodulation system uses sensors (EEG, MEG, fMRI) to record biomarkers, monitoring brain state. Collected signals are processed by a decoder and algorithms for feature extraction and pattern recognition, converting them into understandable commands. These commands can control external devices. When precise neural activity regulation is needed, it enters a "neural stimulation closed-loop": error signals are processed, stimulation parameters (intensity, frequency) are dynamically adjusted by an optimizer, and an encoder drives stimulators (TMS, tDCS/tACS, TUS) to modulate the brain. Post-stimulation brain activity is re-acquired, creating a "stimulation-monitoring-decoding-re-stimulation" non-invasive neuromodulation loop for precise regulation.

2.3.2. Technical Bottlenecks
Non-invasive closed-loop neuromodulation faces challenges, primarily the conflict between the need for high precision and non-invasive constraints (skull attenuation, noise, stability). Processing delays, high computational resource demands for complex algorithms, and clinical translation hurdles (regulatory needs, high R&D costs, long payback periods) also constrain implementation.

(Subsequent sections of the original article detail applications and progress in specific clinical areas, including:)

• Disorders of Consciousness (DOC): Using BCI and TMS/tDCS for assessment and potential arousal effects.

• Alzheimer's Disease (AD): Example of Sinaptica's personalized closed-loop system (SinaptiStim-AD) which received FDA Breakthrough Device Designation, showing potential to slow AD progression by >80% over 6 months in a Phase 2 trial.

• Depression & Anxiety: Examples include combined tDCS+rTMS treatment, and Magnus Medical's SAINT system (MRI-guided TMS) achieving 90.5% clinical remission in severe treatment-resistant depression.

• Motor Disorders: Applications in stroke rehabilitation, brain connectivity research using TMS-EEG.

• Epilepsy: Use of EEG for automatic seizure detection (e.g., TU Delft's system predicting seizures 17 minutes in advance with 92% sensitivity), and research into Temporal Interference (TI) stimulation.

• ADHD & Autism (ASD): Recent clinical trials (published in JAMA Network Open) providing evidence for tDCS in pediatric ADHD (improving working memory, interference control) and transcranial pulsed current stimulation (tPCS) in children with autism (improving social functioning and sleep).

Conclusion
Non-invasive BCI and closed-loop neuromodulation are rapidly developing, showing clear potential in intelligent healthcare. Closed-loop neuromodulation, as a core advanced direction, effectively overcomes the limitations of open-loop systems. Despite challenges in signal processing, real-time performance, and clinical translation, it holds promise for unlocking further potential in exploring brain function and improving neurological health.

(The article concludes with a list of 13 references.)