NEO BCI lets a man quadriplegic for 14 years drink unaided

NEO (Neural Electronic Opportunity) wireless minimally invasive epidural brain-computer interface: eight epidural electrodes over the sensorimotor cortex, a battery-free coin-sized skull implant powered and read through the scalp, and a Riemannian-geometry grasp-intention decoder driving a pneumatic exoskeleton glove

A classifier based on Riemannian geometry, with a hidden Markov model for temporal smoothing.

Exact release not established.

Tsinghua University School of Medicine, Prof. Hong Bo's team, with Neuracle Medical Technology (Shanghai) Co., Ltd. (博睿康科技), the trial sponsor

The decoder detected intended grasping movements and triggered the pneumatic glove.

The deployment used the NEO brain-computer interface.

Not assessed

Not assessed

Eight chronic epidural electrodes over the sensorimotor cortex of a patient whose cortex still generated grasp-related signals after 14 years of complete C4 injury. A grasp-intention decoder calibrated in about ten minutes and robust to the signal drift and chewing artifacts of home use. A battery-free implant powered and read through the scalp, so the patient could train at home daily after discharge ten days post-surgery. A wearable pneumatic glove able to execute the decoded grasp on the patient's right hand, which retained limited forearm movement.

Not assessed

The AI contribution is the decoder, a Riemannian-geometry classifier with a hidden Markov model for temporal smoothing. Calibrated in about ten minutes on beta, low-gamma and high-gamma features of eight-channel epidural ECoG, it detects grasp intention and triggers the pneumatic glove. The preprint reports 94% calibration accuracy and a grasp-detection F1-score rising above 0.9 after three months (average 0.91 over nine months). It gives 100% object-transfer success with the BCI against 35% without. The electrodes, implant, surgery and glove are not AI. The authors attribute the later neurological recovery to repeated BCI training, not the decoder alone.

The AI contribution is the decoder, a Riemannian-geometry classifier with a hidden Markov model for temporal smoothing. Calibrated in about ten minutes on beta, low-gamma and high-gamma features of eight-channel epidural ECoG, it detects grasp intention and triggers the pneumatic glove. The preprint reports 94% calibration accuracy and a grasp-detection F1-score rising above 0.9 after three months (average 0.91 over nine months). It gives 100% object-transfer success with the BCI against 35% without. The electrodes, implant, surgery and glove are not AI. The authors attribute the later neurological recovery to repeated BCI training, not the decoder alone.

Not assessed

drink water unaided, with grasp-decoding accuracy above 90%

Improved Physical Health

Not assessed

People.

Not assessed

Not assessed

Not assessed

Not assessed

Yes

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

drink water unaided, with grasp-decoding accuracy above 90%

his own state before implantation, quadriplegic for 14 years after complete cervical spinal cord injury (ASIA A)

Reference: 处于四肢瘫痪状态已经14年 (no hand function for 14 years). Result: 实现自主喝水等脑控功能，抓握解码准确率超过90% (brain-controlled drinking, grasp decoding accuracy above 90%).

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Developer: Tsinghua University School of Medicine, Prof. Hong Bo's team, with Neuracle Medical Technology (Shanghai) Co., Ltd. (博睿康科技), the trial sponsor. Operator: Xuanwu Hospital, Capital Medical University, Beijing (teams of Zhao Guoguang and Shan Yongzhi).

Benefit: Beijing, China. Developer: China. Operator: China.

Not assessed

Not assessed

Not assessed

Not assessed

The patient's motor cortex must still produce decodable grasp intention despite years of paralysis. The decoder must stay accurate without frequent recalibration across months of home use. The patient or a caregiver must be able to attach the external coil and the glove for daily sessions. Neurological recovery beyond assisted grasping is a single-patient finding and must be reproduced in further patients.

Not assessed

Later work includes a September 2024 medRxiv preprint on nine months of home use and a Tsinghua Medicine page on a 2025 multicenter trial. Their findings remain unverified.

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

AIBD-0123-source-0001–0006

Numbered references below.

Not assessed

Developer, Independent

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

People.

People: 3.

People: After 14 years of complete quadriplegia the patient regained the ability to drink and eat by himself through the BCI-driven glove. The preprint reports neurological recovery: upper-limb motor score +5, ARAT +27, and the ability to hold objects without the BCI.

Not assessed

Not assessed

Not assessed

Not assessed

Independently reported

Not assessed

Not assessed

Healthcare. Neurotechnology and spinal cord injury rehabilitation.

NEO was built to let people with tetraplegia control external equipment, including a pneumatic exoskeleton glove, from epidural brain signals. The aim is assistance and hand-function rehabilitation at home. Restoring this patient's grasp so that he could drink was that purpose.

Date basis unassessed: 2024 (precision unassessed). Updated: 21 October 2026.

Not assessed

SHA-256: fbb504178db4…

oecd_aim: 2024-01-27-c761.

Not assessed

Not assessed

6 revisions. Latest: 21 October 2026.