血管内神经接口：当前平台与临床成熟度

Neurointerventional techniques are facilitating a new class of neural interfaces that record and stimulate brain activity from within the cerebral vasculature. Conventional scalp electroencephalography (EEG) is safe and widely scalable but is limited by skull attenuation and volume conduction, whereas electrocorticography and stereoelectroencephalography provide higher-amplitude signals at the cost of craniotomy or stereotactic depth implantation and procedure-related morbidity. Endovascular approaches offer a distinct access paradigm by leveraging familiar catheter-based workflows to reach cortical veins and dural sinuses. They occupy a practical middle ground that enhances signal quality relative to scalp EEG while mitigating some of the procedural risks associated with open or multi-trajectory intracranial implants. This narrative review summarizes the historical evolution and major device classes, including catheter-based electrodes, stent-electrode arrays, and emerging leadless or wireless systems, with emphasis on leading clinical platforms such as Stentrode (a stent-electrode recording array from Synchron, New York, USA), and EP-01 (an EEG device from Epsilon Medical, Japan). We synthesize evidence on implantation targets, deliverability, signal characteristics relevant to epilepsy evaluation and brain-computer interface applications, stimulation feasibility, and translational constraints governing clinical adoption, including antithrombotic management, vascular patency, imaging surveillance, complications, and device failure modes. We highlight decision-linked endpoints, particularly concordance with conventional intracranial EEG for seizure lateralization, and outline essential reporting elements needed to compare studies across anatomical locations, referencing strategies, and artifact environments. Finally, we provide pragmatic recommendations for neurointerventional adoption and identify priorities for next-generation device development, registries, and multicenter prospective trials.