Accessing neural signals
Compare non-invasive, endovascular, cortical-surface and penetrating interfaces without the hype.
Getting closer to neurons usually buys signal fidelity and bandwidth, but costs surgical risk, narrows the area sampled and makes reversal less easy.
Non-invasive
Measures electrical, magnetic, blood-flow or oxygenation signals without entering the body.
- Placement
- Sensors remain outside the skin and skull.
- Real examples
- Clinical EEG caps, MEG systems, MRI scanners and wearable fNIRS.
- Spatial resolution
- Centimetres for EEG; millimetres for MRI-based imaging.
- Temporal resolution
- Milliseconds for EEG/MEG; seconds for fMRI/fNIRS.
- Bandwidth
- Usually tens to hundreds of sensors; signals mix large neural populations.
- Longevity
- Repeatable over years, but placement and physiology vary between sessions.
- Reversibility
- Fully removable after each session.
- Best for
- Broad research, clinical monitoring and low-risk interfaces.
- Surgical risk
- None
No surgical wound; modality-specific screening still applies.
Endovascular
Records local field activity through the vessel wall without a craniotomy.
- Placement
- An electrode-bearing stent is delivered through a vein and sits beside motor cortex inside a cerebral vessel.
- Real examples
- Synchron's investigational Stentrode brain–computer interface.
- Spatial resolution
- Local cortical populations near the implanted vessel.
- Temporal resolution
- Milliseconds.
- Bandwidth
- About 16 electrodes in current Stentrode arrays; lower throughput than surface or penetrating arrays.
- Longevity
- Designed for chronic use; long-term evidence remains limited.
- Reversibility
- Potentially removable early, but endothelial tissue can grow over the device.
- Best for
- Communication and device control when open brain surgery is undesirable.
- Surgical risk
- Moderate
Catheter procedure with vascular, clotting and long-term implant risks.
Subdural / cortical surface
ECoG records local field potentials with less skull blurring than scalp EEG.
- Placement
- Arrays sit on the cortical surface beneath the dura after a craniotomy.
- Real examples
- Clinical epilepsy-monitoring grids; investigational speech-decoding ECoG arrays.
- Spatial resolution
- Millimetres to about one centimetre, depending on contact spacing.
- Temporal resolution
- Milliseconds.
- Bandwidth
- Typically tens to hundreds of contacts over a limited cortical area.
- Longevity
- Clinical grids are often temporary; chronic arrays may scar or shift.
- Reversibility
- Can be removed surgically, requiring another operation.
- Best for
- Epilepsy mapping and high-bandwidth cortical decoding.
- Surgical risk
- High
Requires cranial surgery; infection, bleeding and seizures are material risks.
Intracortical
Records local field potentials and, in suitable conditions, individual-neuron spikes.
- Placement
- Microscopic electrodes penetrate cortical tissue and sit closest to neurons.
- Real examples
- Utah arrays, Neuropixels probes and Neuralink's investigational flexible threads.
- Spatial resolution
- Micrometres to sub-millimetres around each electrode.
- Temporal resolution
- Sub-millisecond to milliseconds.
- Bandwidth
- Tens to thousands of channels; potentially includes single-unit spikes.
- Longevity
- Signals can drift or fade as tissue responds; stability varies by device and person.
- Reversibility
- Removal needs surgery and cannot undo microscopic tissue injury.
- Best for
- Research requiring precise motor, sensory or speech decoding.
- Surgical risk
- High
Penetrating brain surgery carries bleeding, infection and tissue-response risks.