Summary of "Day 2 (Part 2) The BCI & Neurotechnology Spring School 2026"

Scientific concepts, discoveries, and nature/medical phenomena

1) Intraoperative neurophysiology for brain tumor surgery (motor pathway protection)

Core phenomena

Ischemia affecting the cortical/corticospinal tract can manifest as MEP (motor evoked potential) deterioration during surgery.
MEP changes may be reversible warnings for impending injury.
The severity of MEP deterioration correlates with postoperative deficits.

Clinical evidence described

A debated study claimed no significant long-term neuro outcome effect, but confounding was discussed (e.g., higher rates of temporary clipping in one group).
Reply studies suggested that intraoperative monitoring/mapping reduces overall deficits.

Methodological workflow described (in surgery)

Place a strip electrode for motor mapping.
Use:
- Transcranial electric stimulation
- Direct cortical stimulation (including subcortical stimulation)
Monitor with sEMPs/SAPs (described as sensory evoked potentials) and MEPs.
During resection:
- Stimulate within the resection cavity to elicit responses in contralateral hand/face areas.
- Apply a double stimulation paradigm:
  - first provide stimulation artifacts,
  - then use time-lagged responses to delineate functional boundaries.
- Use results to decide where to stop resection.

Stimulation parameter concepts

Goal: selective activation near the axon hillock / gray–white matter border to target the corticospinal tract while minimizing current spread.
Monopolar vs bipolar stimulation:
- Bipolar: highest current tends to be in the middle.
- Monopolar: effects skew more toward layer 5.

MEP monitoring interpretation concepts

Distinguish between:
- Reversible vs irreversible MEP deterioration
Transient MEP reductions may indicate impending ischemia in the corticospinal tract.

NAPS warning optimization (predictive thresholds)

A large-cohort analysis (Tom from Germany mentioned) proposed a criterion combining:
- Affected-side amplitude decrement > 80%
- with increased stimulation intensity
Reported outcome: higher sensitivity / positive predictive value for impending dysfunction.

Direct cortical stimulation (DCS) vs transcranial stimulation (tES) tradeoffs

DCS is favorable when strip electrodes can be placed well on motor cortex.
tES can produce false positives, e.g.:
- parietal tumors and large resection cavities leading to “zaging”/subdural air.
Irrigation / reestablishment of signals may help identify false positives.

2) Subcortical distance estimation via stimulation thresholds

Scientific idea

Subcortical stimulation intensity correlates approximately linearly with distance to the corticospinal tract within a usable range.

Quantitative relation described

A near-linear region reported:
- intensities < ~15 mA
- distances to ~50 mm

Clinical use

Use the minimum intensity that elicits contralateral MEPs as a surrogate for proximity to corticospinal fibers.

Bipolar vs monopolar in different tumor contexts

Bipolar may be safer for “low-risk motor pathway” tumors.
Monopolar may work better in more infiltrative or procedural-history scenarios.

3) Awake/eloquent tumor surgery and maximization of gross total resection

Key principle

More extensive gross total resection is linked to improved overall survival.

Neurophysiology as a guiding tool

Mapping + monitoring add guidance beyond imaging alone:
- navigation/visualization plus physiology-based boundaries (MEPs/SAPs/DCS/subcortical stimulation).

4) Future outlook in neuro-oncology electrophysiology: cortical potentials (CPs)

New modality

Cortical potentials (CPs) are discussed as a next step, requiring:
- technical maturation
- better understanding of mechanisms and connectivity
- clarity on how CPs relate to tumor-relevant pathways (and ASEPs/AEPs)

Brain–spine interfaces after spinal cord injury (epidurally driven locomotion + cortical decoding)

5) Implantable brain–spine interface for severe spinal cord injury

Target problem/phenomenon

After spinal cord injury, brain commands to lumbar spinal locomotor circuits are disrupted, leading to paralysis.

Core technology

Epidural electrical stimulation (EES/ES) placed just below the lesion:
- engages spinal circuits transinaptically via dorsal root afferents
- activates motor neurons → muscle activation

Clinical transition described

From earlier predefined stepping patterns (lower patient control)
to patient intention decoding that drives stimulation.

Implant described

Partner-developed implant:
- ~64 epidural electrodes in some versions
- later versions use a single medially placed unit for severe cases
Wireless communication:
- an external antenna/cap on the head
- implant placed under the skull region

6) Pilot/clinical trial structure and participant outcomes (high-level)

Trial logistics

Enrollment → single surgery implantation → 2-week calibration → 14-week rehabilitation
Afterward: patient can take the system home.

Stimulation parameter space

Electrode configuration, frequency, passes per burst, amplitude
Adapted in real time via wireless control.

Calibration/mapping methodology (explicit steps)

Iterative algorithmic exploration:
1. algorithm proposes parameter sets
2. patient is stimulated
3. record kinematics + EMG
4. compute a reward
5. iterate until achieving “primitive” movements for walking/standing
Achieves multiple joint primitives within 1–2 days per phase (as stated).

7) Decoding model: intention classification from brain signals

Model concept

Recursive exponentially weighted switching multilinear model
Updates every ~100 ms
Uses prior ~400 ms to 1 s of data
Classifies intention per joint (step left/right, hip flexion/extension, knee/ankle movements).

Online closed-loop calibration

Additional labeling/cues every ~15 seconds
Decoder updates during therapy if signal characteristics shift.

8) “Brain GPT” self-supervised learning for robustness

Concept

A self-supervised model learns latent structure of brain signals.

Method outline

Convert signals into spectro-temporal-spatial feature tensors
Convert features into tokens
Mask tokens and train a transformer/encoder to reconstruct masked parts
Use learned latent features to improve decoder stability.

9) Results themes reported

Complete SCI participants

No voluntary movement without stimulation (expected for complete lesions).
With stimulation:
- standing/walking enabled (prosthetic-like effect)
Improvements maintained/enhanced with rehabilitation and home training.

Incomplete SCI participant

More residual motor function on one side → asymmetric improvement:
- incomplete side improves during rehab
- more severe side less responsive
With stimulation:
- standing longer
- walking with less need for body-weight support (as described)

10) Neurofeedback and closed-loop motivation tools

Neurofeedback provides real-time visualization (cursor/map) of brain feature states associated with intended joints.

11) Practical use and decoder stability over time

Decoder recalibration frequency

Early rehab: model updates about every ~4 days
Later: decoder becomes more stable with fewer updates.

Home use

Systems embedded in a walker enable independent training.

Glial tumors, epilepsy, and neuronal activity fueling tumor growth (mechanistic and clinical links)

12) Clinical epidemiology: seizures co-evolve with tumor progression

Phenomenon

Tumor growth and seizure control/frequency show parallel trajectories.

Key observations stated

High fraction of patients with low-grade and glioblastoma develop seizures over time.
Increasing seizure prevalence tracks increased pharmacoresistance.
Epilepsy correlates with prognosis:
- seizures at diagnosis/tumor evolution associate with different survival patterns (noted as shorter survival in some glioma series).

13) Neuronal activity can biologically fuel glioblastoma growth (preclinical evidence)

Animal model described

Xenograft human glioblastoma cells into mouse brain.

Manipulation #1: activate tumor-associated/parenchymal neuronal activity

Via optogenetics/control of neuronal components
Result: tumor growth ~doubled.

Manipulation #2: activate host neurons via optogenetics

Result: tumor growth reduced (as described).

Manipulation #3: visual cortex light exposure

Tumor implanted in visual cortex and exposed to light
Result: altered tumor growth.

Interpretation

Neuronal activity can modulate tumor growth (“fuel” concept).

14) Spatial relationship: brain activity predicts tumor location probability (MEG-based)

High baseline MEG activity correlates with likely glioblastoma locations.
Stronger correlation for glioblastoma than for astrocytomas.
Weaker for low-grade oligomas.

15) Epilepsy generation occurs in peritumoral cortex rather than the tumor core

Historical evidence

Jackson (1882) proposed epilepsy is the initial symptom and relates to cortical involvement.

Surgical recordings described

Intraoperative recordings:
- tumor sites may show flat activity
- epileptiform sharp activity appears in cortex surrounding the tumor

Modern tissue experiments described

Postoperative maintained tissue slices:
- epileptiform activity in peritumoral cortices
- absent in tumor tissue and remote cortex slices

16) Surgical implication: supramarginal resection and mapping needs

Concept

“Supramarginal surgery” aims to remove tissue beyond MRI-visible boundaries due to:
- infiltration
- epileptogenic cortex

Constraint

Lack of reliable biomarkers for:
- tumor cell infiltration vs
- epileptogenic zones
Spike hunting alone may not reliably map:
- irritative vs
- seizure-generating zones.

17) High-frequency oscillations (HFOs) as biomarkers for epilepsy and infiltration

Phenomenon

Ripples and fast ripples correlate with epileptogenic tissue and can align with tumor infiltration.

Tissue/in vitro mechanistic description

In living human tumor tissue on MEAs:
- epileptogenic progression includes:
  - low-frequency events (interictal → preictal discharges)
  - increasing HFO activity
- HFOs emerge in microdomains that accumulate until seizure onset.

Microdomain hypothesis

Macro electrodes may miss localized HFO microdomains due to spatial averaging.

18) New electrode technology to detect HFOs (conductive polymer electrodes)

Problem with classical macro electrodes

Example: ~5 mm diameter
Limited spatial sampling and high noise.

Conductive polymer electrode approach (POT / Panaxium)

Noise reduced by ~10× (as stated)
Detect high-frequency components embedded in noise
Very small contact diameters: ~10–30 µm
Thin electrodes: ~4–10 µm, with tunable spacing

Planned trial concept

Record intraoperatively, analyze electrophysiology, and decide infiltration/high vs low/no infiltration after resection.

19) Synaptogenic signaling link: TSP1 and gamma power association

Reported chain

Infiltrated regions show increased gamma power and altered functional connectivity.
Synaptogenic factor thrombospondin-1 (TSP1) upregulated in highly connected tissue.
Animal models:
- faster tumor growth and shorter survival when grafts originate from high-connectivity regions.

Drug repurposing

Gabapentin reduces proliferation (via reduced TSP1/proliferation mechanisms as described).
Clinical cohorts:
- glioblastoma patients treated with gabapentin show improved survival (Boston and San Francisco cohorts, as reported).

20) Shared biological pathways: glutamate, synapses onto glia, calcium oscillations

Mechanisms described

Tumor cells release high glutamate → increases neuronal activity → epileptic activity.
Tumor cells express glutamate receptors; glutamate supports growth/migration.
Neurons synapse onto glioma cells (fraction reported ~10–15%).
Synaptic events generate calcium oscillations in tumor cells; calcium activity propagates through tumor networks.
Neural activity influences tumor via growth factors (e.g., NLGN3 mentioned).
Tumor connectivity via microtubes supports survival/extension.

Chemogenetics evidence concept

Activating specific neurons accelerates tumor growth and drives spread toward more active regions → suggests activity-dependent tumor invasion.

Adaptive/Network-level deep brain stimulation (DBS) and translational neurotechnology

21) DBS targets and mechanistic levels of analysis

Targets discussed

Thalamic targets (e.g., VIM)
GPi (Tourette syndrome / Parkinson’s)
STN for Parkinson’s

DBS parameters (typical clinical examples described)

Pulse width: ~60 µs
Frequency: ~130 Hz
Amplitude: ~1–5 V

Mechanistic framing

Move from:
- single-neuron spike changes
- to oscillatory rhythms (LFP)
- to whole-brain connectivity patterns
Goal: identify oscillatory/network biomarkers for symptom states.

22) Multimodal human neuroscience for network biomarkers

Data sources

ECoG + DBS LFP + MEG/EEG + imaging

Examples stated

STN/basal ganglia rhythms and cortical rhythms:
- beta/alpha/theta spectral roles discussed
Relationship between dopamine receptors and connectivity/tracer uptake.
Dopamine medication vs STN DBS produce spectral differences (site- and frequency-specific effects).

23) Adaptive DBS as a “clinical BCI”

Core concept

Real-time detection of biomarkers (e.g., beta power) drives stimulation on/off or amplitude changes.

Translational status described

Fully implantable adaptive DBS reported as FDA/CE-approved (as stated).

Technical/clinical issues

Some patients fail due to insufficient signal quality (noise/distance to target).
Cardiac dipole proximity can create ECG noise, impacting eligibility.

Beyond beta power

Sleep disturbance and state-dependent offsets may break fixed-threshold adaptive schemes.
Proposal: stimulation driven by behavioral/context triggers (e.g., movement speed).

24) Movement-speed responsive stimulation idea

Stimulation effects depend on patient behavior at millisecond timescales.
Translational experiment:
- trigger stimulation during fast vs slow voluntary movements
- reported finding: fast-movement stimulation counteracts pathological slowing.

25) “Therapeutic BCI” definition and AI-driven adaptive circuits

Adaptive DBS + machine learning decoding can be framed as a therapeutic BCI.
Aims toward:
- scalable cross-patient models
- connectomic circuit targeting
- real-time symptom-specific adaptive stimulation

26) Real-time AI steering and “plug-and-play” adaptive stimulation

Pre-trained models may interpret signals without long individual training.
Validation described as contrasting against approaches requiring extended patient-specific calibration.

Researchers / sources mentioned (as featured)

(Only names explicitly present in the subtitles are included.)

Mark McDonald
Tom (mentioned: “Tom from Ging here in Germany”)
Lorenzo (Lorenzo group; surname unclear)
Emanuel Mand
Kathleen Zaidder (from Bon/Bochum region; subtitles unclear)
Catlin (Kathleen Catlin? name appears as “Catlin summarized her data”)
Manuel (mentioned repeatedly)
Ricky Matsumoto
Fran Don BL (surname unclear)
Michelle Leon
Navarete (associated with “ripple lab”)
John Pal
Jackson (historical paper, 1882)
(Paris) Pitié-Salpêtrière team (1966 study; individual names not given)
Akim Purakbari
Mera Chikaban
Sao Yin
Simon Little
Philip Star
Tim Denson
Tibokan Bianca Yu
Ari (supervisor name partially unclear)
Josine (supervisor name partially unclear)
Kate Schneider
Katalina P.
Eduardo Mug’s lab (adaptive DBS + gate signatures preprint mentioned; surname timing unclear)
Allesia Cavalo
David Martin
Adam (questioner; neurosurgeon in Denver; surname unclear)
Oscar
Clara / Clara wants… (questioner; not clearly tied to a researcher)

Go-to collaborators/partners

Onward (company mentioned; adaptive DBS system)
Clean tech / GTech / GTech amplify (company mentioned; multiple instances)
Panaxium (company mentioned; POT electrodes)
Avo (software company; “Alysia” mentioned)
Gilles Huberfeld / Uberfeld (pronunciation discussion; exact spelling uncertain)

Share this summary

Is the summary off?

If you think the summary is inaccurate, you can reprocess it with the latest model.

Summarize another video