China Approves Third BCI Medical Device Standard, World's First for AI Brainwave Processing

China's National Medical Products Administration (NMPA) has approved the country's third medical device standard for brain-computer interfaces (BCI), described as the world's first product standard for BCI devices that use artificial intelligence algorithms to process brainwave data. The standard takes effect on September 1, 2027.

The new framework sets unified requirements for the collection, processing, labeling, storage and access of brainwave data, aiming to bring consistency to a field where research and industry practices have so far diverged widely.

Li Shu, deputy director of the medical device institute under the National Institutes for Food and Drug Control (NIFDC) and a drafter of the standard, said it lays down clear technical specifications covering electrode placement, sampling frequency, and how data should be collected and recorded. With these in place, he said, the datasets used to train BCI devices will be of higher quality, decoding performance will be better assured, and diagnosis and treatment will become safer and more effective.

Li pointed to the fragmented state of electroencephalogram (EEG) data collection before the standard was issued. Research institutions and companies used differing methods — some with 64 channels, others with 128 — while sampling frequencies ranged from 2 Hz to 1,000 Hz, and annotation practices also varied. Data gathered this way, he said, was difficult to compare or reuse.

BCI medical devices work like intelligent machines that can read brain signals, learning from large volumes of EEG data to train AI algorithms to interpret a patient's intentions. That process is closely tied to both treatment outcomes and patient safety.

Zheng Jia, deputy director of the NIFDC's Institute for Medical Device Standardization, said such devices could help paralyzed patients control a robotic arm with their thoughts, or help patients with speech disorders express what they want to say. But if the quality of training data is uneven — comparable, he said, to using blurry photographs to train a facial recognition system — the device may fail to decode the patient's true intentions, affecting treatment and even posing safety risks.

According to Li, the new standard will allow data from different research departments to be benchmarked against one another and shared for reuse. It gives developers a unified quality threshold clarifying what data can be used, lets production departments trace and track EEG data generated at different times, and helps regulatory and evaluation bodies establish a common technical yardstick.

Yuan Peng, deputy director of the NMPA's Department of Medical Device Registration, said the standard moves the emerging BCI industry from a situation where "everyone speaks their own language" to one of unified dialogue, allowing these medical devices to reach the market faster and in a more standardized way.