What changed between revision 1 and revision 2, compared over the article's markdown source with its frontmatter — so a change to status, evidence or reversibility shows up as the edit it is.
Older · revision 1
Import· initial corpus import, not an edit
Initial import of content/ecog-interfaces.md — the filesystem corpus, unchanged. Not an edit.
12.8 kB
Revision 1
Revision 2
lines 7–14 → 7–146 unchanged lines not shown
trl: 6trl: 6categories: ["cybernetics"]categories: ["cybernetics"]tags: ["bci", "ecog", "speech decoding", "neural recording", "implants", "epilepsy"]tags: ["bci", "ecog", "speech decoding", "neural recording", "implants", "epilepsy"]summary: "Brain–computer interfaces that record from electrode arrays resting on the cortical surface rather than penetrating it, trading single-neuron resolution for stability and coverage."summary: "Brain-computer interfaces that record from electrode arrays resting on the cortical surface rather than penetrating it, trading single-neuron resolution for stability and coverage."updated: "2026-07-27"updated: "2026-10-29"humanEvidence: "Almost the entire evidence base is human: epilepsy patients temporarily implanted for seizure mapping, plus single participants with paralysis or ALS who have used chronic arrays for speech, cursor and walking control."humanEvidence: "Almost the entire evidence base is human: epilepsy patients temporarily implanted for seizure mapping, plus single participants with paralysis or ALS who have used chronic arrays for speech, cursor and walking control."access: "Clinical grids are cleared only for temporary use around epilepsy surgery; every chronic interface use is a small investigational study, and no chronic product is approved."access: "Clinical grids are cleared only for temporary use around epilepsy surgery; every chronic interface use is a small investigational study, and no chronic product is approved."reversibility: "difficult"reversibility: "difficult"lines 20–46 → 20–465 unchanged lines not shown
"caption": "Class of neural interface", "caption": "Class of neural interface", "rows": [ "rows": [ { "label": "Signal", "value": "Cortical surface potentials" }, { "label": "Signal", "value": "Cortical surface potentials" }, { "label": "Key frequency band", "value": "High gamma, ~70–200 Hz" }, { "label": "Key frequency band", "value": "High gamma, ~70-200 Hz" }, { "label": "Spatial resolution", "value": "Millimetres" }, { "label": "Spatial resolution", "value": "Millimetres" }, { "label": "Clinical origin", "value": "Epilepsy seizure mapping" }, { "label": "Clinical origin", "value": "Epilepsy seizure mapping" }, { "label": "Typical contacts", "value": "64–1,024" }, { "label": "Typical contacts", "value": "64-1,024" }, { "label": "Placement", "value": "Subdural or epidural" }, { "label": "Placement", "value": "Subdural or epidural" }, { "label": "Regulatory status", "value": "Cleared for temporary use; chronic BCI investigational" } { "label": "Regulatory status", "value": "Cleared for temporary use; chronic BCI investigational" } ] ]}}`````` **Electrocorticography interfaces** are brain–computer interfaces that record from electrode arrays**Electrocorticography interfaces** are brain-computer interfaces that record from electrode arrayslaid on the surface of the cortex, either beneath the dura (subdural) or on top of it (epidural),laid on the surface of the cortex, either beneath the dura (subdural) or on top of it (epidural),without penetrating brain tissue. They occupy the middle of the invasiveness spectrum: betterwithout penetrating brain tissue. They sit in the middle of the invasiveness range. Their spatialspatial and spectral resolution than scalp electroencephalography, worse than a penetratingand spectral resolution is better than scalp electroencephalography and worse than a penetrating[[utah-array]], with mechanical stability that neither of the alternatives matches.[[utah-array]], and their mechanical stability is better than either. The technique exists because neurosurgery needed it first. Grids of platinum contacts have beenThe technique exists because neurosurgery needed it first. Grids of platinum contacts have beenplaced on the cortex for decades to localize seizure onset and map eloquent cortex before epilepsyplaced on the cortex for decades to localize seizure onset and map eloquent cortex before epilepsysurgery, which means the field inherited a large clinical base of patients temporarily instrumentedsurgery. The field therefore inherited a large clinical base of patients temporarily instrumentedwith electrodes and willing to participate in research. Most of what is known about human corticalwith electrodes and willing to take part in research, and most of what is known about how speech isrepresentations of speech comes from that population.represented in human cortex comes from that population. ```figure```figure{"key": "bci-sensor-types", "caption": "Three placements and three trade-offs: signal quality rises with invasiveness, and so does surgical risk."}{"key": "bci-sensor-types", "caption": "Three placements and three trade-offs: signal quality rises with invasiveness, and so does surgical risk."}lines 49–118 → 49–1182 unchanged lines not shown
## How it works## How it works A standard clinical grid is a silicone sheet holding platinum discs a few millimetres across on aA standard clinical grid is a silicone sheet holding platinum discs a few millimetres across on aone-centimetre pitch. Research arrays go finer: high-density grids at two to four millimetres, andone-centimetre pitch. Research arrays go finer. High-density grids use a two to four millimetrethin-film micro-electrocorticography arrays with hundreds or thousands of contacts atpitch, and thin-film micro-electrocorticography arrays carry hundreds or thousands of contacts atsub-millimetre spacing on a substrate thin enough to conform to the cortical surface, includingsub-millimetre spacing on a substrate thin enough to conform to the cortical surface, including intointo sulci.sulci. The most useful feature for a [[brain-computer-interface]] is high-gamma power, roughly 70 to 200The most useful feature for a [[brain-computer-interface]] is high-gamma power, roughly 70 to 200hertz. Broadband high-frequency activity recorded at the surface correlates well with the firinghertz. Broadband high-frequency activity recorded at the surface correlates well with the firingrate of the population of neurons directly underneath, so it functions as a proxy for local spikingrate of the population of neurons directly underneath, so it works as a proxy for local spiking thatthat can be measured without pushing anything into the tissue. Lower-frequency rhythms carrycan be measured without pushing anything into the tissue. Lower-frequency rhythms carrycomplementary information about state and movement preparation. Together these features feedcomplementary information about state and movement preparation. Together these features feedstandard [[neural-decoding]] pipelines.standard [[neural-decoding]] pipelines. Because the array does not penetrate, it does not sever capillaries or provoke the encapsulationBecause the array does not penetrate, it does not sever capillaries or provoke the encapsulationresponse that displaces neurons from a penetrating tip. Fibrous tissue does form around a subduralresponse that displaces neurons from a penetrating tip. Fibrous tissue does form around a subduralgrid, but the recorded signal degrades far more slowly than intracortical spikes do. This is thegrid, but the recorded signal degrades far more slowly than intracortical spikes do. This is thecentral engineering argument for the approach.main engineering argument for the approach. > [!note] The resolution ceiling> [!note] The resolution ceiling> Surface recording cannot resolve individual neurons, and no improvement in electrode density> Surface recording cannot resolve individual neurons, and no increase in electrode density changes> changes that. Cerebrospinal fluid and the dura conduct and blur; the signal at any contact is a> that. Cerebrospinal fluid and the dura conduct and blur the signal, so what reaches any contact is> weighted sum over a cortical patch. Denser arrays sample that blurred field more finely, which> a weighted sum over a patch of cortex. Denser arrays sample that blurred field more finely, which> helps, but the information about individual cells is gone before it reaches the electrode.> helps, but the information about individual cells is lost before it reaches the electrode. ## Development history## Development history The clinical practice long predates the interface application. Patients admitted for invasiveThe clinical practice long predates the interface application. Patients admitted for invasiveseizure monitoring spend one to two weeks with grids or depth electrodes in place, awake andseizure monitoring spend one to two weeks with grids or depth electrodes in place, awake and able toavailable to perform tasks, and that window has supplied most of the human cortical physiologyperform tasks, and that window has supplied most of the human cortical physiology behind modernbehind modern decoders. The same population, instrumented with hippocampal depth electrodes ratherdecoders. The same population, instrumented with hippocampal depth electrodes rather than surfacethan surface grids, supplies the data behind [[memory-prosthesis]] research. The arrangement hasgrids, supplies the data behind [[memory-prosthesis]] research. The arrangement has its own ethicalits own ethical literature: participants are patients first, recruited during a stressful clinicalliterature. Participants are patients first, recruited during a stressful clinical episode, and theepisode, and the research question is never the reason the electrodes are there.research question is never the reason the electrodes are there. ```timeline```timeline[[ { "year": "1930s–1950s", "title": "Clinical electrocorticography", "text": "Wilder Penfield and Herbert Jasper establish intraoperative cortical recording and stimulation mapping at the Montreal Neurological Institute." }, { "year": "1930s-1950s", "title": "Clinical electrocorticography", "text": "Wilder Penfield and Herbert Jasper establish intraoperative cortical recording and stimulation mapping at the Montreal Neurological Institute." }, { "year": "2004–2010", "title": "ECoG as a BCI signal", "text": "Groups working with epilepsy patients show that high-gamma activity from surface electrodes supports rapid, accurate cursor control learned within minutes." }, { "year": "2004-2010", "title": "ECoG as a BCI signal", "text": "Groups working with epilepsy patients show that high-gamma activity from surface electrodes supports rapid, accurate cursor control learned within minutes." }, { "year": "2013", "title": "Multi-dimensional arm control", "text": "A participant with tetraplegia controls a robotic arm in three dimensions using a subdural array, demonstrating that surface signals support continuous control." }, { "year": "2013", "title": "Multi-dimensional arm control", "text": "A participant with tetraplegia controls a robotic arm in three dimensions using a subdural array, demonstrating that surface signals support continuous control." }, { "year": "2016", "title": "Home use in ALS", "text": "A fully implanted electrocorticography system in Utrecht lets a woman with late-stage ALS select letters at home, remaining functional over years." }, { "year": "2016", "title": "Home use in ALS", "text": "A fully implanted electrocorticography system in Utrecht lets a woman with late-stage ALS select letters at home, remaining functional over years." }, { "year": "2019", "title": "Exoskeleton control", "text": "A French team reports a tetraplegic participant operating a whole-body exoskeleton using bilateral wireless epidural implants." }, { "year": "2019", "title": "Exoskeleton control", "text": "A French team reports a tetraplegic participant operating a whole-body exoskeleton using bilateral wireless epidural implants." }, { "year": "2019–2025", "title": "Speech decoding", "text": "Successive reports from UCSF decode sentences, then conversational-rate text, synthesized voice, and near-real-time streaming speech from surface arrays over speech motor cortex." }, { "year": "2019-2025", "title": "Speech decoding", "text": "Successive reports from UCSF decode sentences, then conversational-rate text, synthesized voice, and near-real-time streaming speech from surface arrays over speech motor cortex." }, { "year": "2023", "title": "Brain–spine interface", "text": "Epidural cortical implants drive a spinal stimulator to restore volitional walking after spinal cord injury." } { "year": "2023", "title": "Brain-spine interface", "text": "Epidural cortical implants drive a spinal stimulator to restore volitional walking after spinal cord injury." }]]`````` ## Speech and communication## Speech and communication Surface arrays have dominated speech decoding for an anatomical reason. The representations of theSurface arrays have dominated speech decoding for an anatomical reason. The representations of thelips, jaw, tongue, and larynx are laid out across several centimetres of ventral sensorimotorlips, jaw, tongue and larynx are laid out across several centimetres of ventral sensorimotor cortexcortex on the lateral surface of the brain — a large, shallow, two-dimensional map, which ison the lateral surface of the brain. A large, shallow, two-dimensional map of this kind is what aexactly what a conformal grid is good at sampling. Penetrating arrays sample it densely in a fewconformal grid samples best. Penetrating arrays sample it densely within a few square millimetres,square millimetres; a grid samples it sparsely across the whole map.whereas a grid samples it sparsely across the whole map. The line of work from UCSF began with synthesizing intelligible speech from cortical activityThe line of work from UCSF began with synthesizing intelligible speech from cortical activityrecorded during spoken sentences,[^anumanchipalli2019] then decoded attempted words in a man withrecorded during spoken sentences.[^anumanchipalli2019] It then decoded attempted words in a man withanarthria after brainstem stroke,[^moses2021] then reached conversational rates with text,anarthria after brainstem stroke,[^moses2021] reached conversational rates with text, synthesizedsynthesized voice, and an animated avatar,[^metzger2023] and more recently reduced the latencyvoice and an animated avatar,[^metzger2023] and more recently cut the latency enough forenough for near-continuous streaming output. These results, discussed in detail undernear-continuous streaming output. These results, discussed in detail under[[speech-neuroprosthesis]], come from single participants using research hardware.[[speech-neuroprosthesis]], come from single participants using research hardware. Communication systems built on smaller chronic implants have a longer service record. A fullyCommunication systems built on smaller chronic implants have a longer service record. A participantimplanted system using a small number of surface electrode strips was used at home by a participantwith ALS used a fully implanted system with a small number of surface electrode strips at home towith ALS to select letters for years, without recalibration drift severe enough to stop itselect letters for years, and recalibration drift never became severe enough to stop itworking.[^vansteensel2016] The bit rate was low. The reliability was the point.working.[^vansteensel2016] The bit rate was low; the result that mattered was the reliability. ## Movement restoration## Movement restoration lines 123–166 → 123–1674 unchanged lines not shown
The most consequential application connects cortex to spinal cord rather than to a machine. In aThe most consequential application connects cortex to spinal cord rather than to a machine. In a2023 report, epidural cortical implants decoded walking intention and drove an implanted epidural2023 report, epidural cortical implants decoded walking intention and drove an implanted epiduralspinal stimulator, restoring volitional walking to a man with an incomplete spinal cord injury, whospinal stimulator, restoring volitional walking to a man with an incomplete spinal cord injury. Healso showed some neurological recovery that persisted with the system off.[^lorach2023] This "digital bridge" framing — using a BCI to reconnect the nervous system toalso showed some neurological recovery that persisted with the system off.[^lorach2023] Thisitself rather than to a robot — is now a major direction in [[neuroprosthetics]]."digital bridge" framing, in which a BCI reconnects the nervous system to itself rather than to arobot, is now a major direction in [[neuroprosthetics]]. ## Advantages and limits## Advantages and limits The advantages are stability, coverage, and channel scaling. Thin-film arrays can carry a thousandThe advantages are stability, coverage and channel scaling. Thin-film arrays can carry a thousand oror more contacts without a thousand penetrations, and one company's high-density cortical filmmore contacts without a thousand penetrations, and one company's high-density cortical film receivedreceived United States clearance in 2025 for temporary use during surgery, a step toward chronicUnited States clearance in 2025 for temporary use during surgery, a step toward chronic approval.approval. The durability argument has an existing precedent: [[deep-brain-stimulation]] leads sitThere is a precedent for durability. [[deep-brain-stimulation]] leads sit in human brains for ain human brains for a decade or more without failing, which suggests that a chronically implanteddecade or more without failing, which suggests that a chronically implanted neural device is notneural device is not intrinsically impossible, only that penetrating recording tips are aintrinsically impossible and that penetrating recording tips are a particularly demanding case.particularly demanding case. Flexible penetrating threads of the kind used by [[neuralink]] attemptFlexible penetrating threads of the kind used by [[neuralink]] try to split the difference,to split the difference, accepting penetration but minimizing stiffness.accepting penetration while minimizing stiffness. The limits are structural. Access still requires opening the skull, though narrow-slot andThe limits are structural. Access still requires opening the skull, though narrow-slot and burr-holeburr-hole insertion techniques reduce the exposure compared with a full craniotomy. Subduralinsertion techniques reduce the exposure compared with a full craniotomy. Subdural placement carriesplacement carries risks of haemorrhage, infection, and cerebrospinal fluid leak. Grids can shift.risks of haemorrhage, infection and cerebrospinal fluid leak, and grids can shift. The resolutionAnd the resolution ceiling caps the number of independent control dimensions available, which isceiling also caps the number of independent control dimensions available. That is why thewhy the highest-dimensional results still come from penetrating electrodes, and why the endovascularhighest-dimensional results still come from penetrating electrodes, and why the endovascular[[stentrode]] — with far fewer contacts and greater distance from cortex — sits lower still.[[stentrode]], with far fewer contacts and more distance from cortex, sits lower still. Any chronic recording of cortical activity, whatever the electrode, generates a data stream fromAny chronic recording of cortical activity, whatever the electrode, produces a data stream fromwhich more can be inferred than the user intends to communicate. That problem is examined underwhich more can be inferred than the user intends to communicate. That problem is examined under[[mental-privacy]], and it applies with particular force to speech-decoding systems.[[mental-privacy]], and it applies with particular force to speech-decoding systems. ## Outlook## Outlook The plausible near-term path runs through fully implanted, wireless, high-channel thin filmsThe plausible near-term path runs through fully implanted, wireless, high-channel thin films aimedtargeted at speech loss, where the clinical need is unambiguous and the decoding results areat speech loss, where the clinical need is clear and the decoding results are strongest. No currentstrongest. Whether such devices sustain performance for a decade, and whether the surgery becomestrial is old enough to say whether such devices sustain performance for a decade, or whether theroutine enough for a neurologist to recommend it to a newly diagnosed ALS patient, are opensurgery will become routine enough for a neurologist to recommend it to a newly diagnosed ALSquestions that no current trial is old enough to answer.patient. A quieter question is what surface arrays reveal about cortical organization itself. Dense gridsA quieter question is what surface arrays reveal about cortical organization itself. Dense gridsover speech and motor cortex have produced maps that were unavailable from animal work, and thatover speech and motor cortex have produced maps that animal work could not provide, and those mapsinformation feeds back into decoder design — a loop between measurement and application thatfeed back into decoder design. The loop between measurement and application resembles theresembles the relationship between [[connectomics]] and theories of neural computation.relationship between [[connectomics]] and theories of neural computation. ## See also## See also removed, struck through added, underlinedLine numbers count the serialised markdown of each revision, frontmatter included.
121 lines are the same in both revisions, and 42 of them are not shown. Neither revision here is necessarily the current article — read Electrocorticography interfaces for the text this wiki stands behind today.