The End Of Knowledge - Vault 3 - GTEC BCI & Neurotechnology Spring School 2024 - Day 1

graph LR classDef kapella fill:#f9d4d4, font-weight:bold, font-size:14px; classDef ecog fill:#d4f9d4, font-weight:bold, font-size:14px; classDef signalProcessing fill:#d4d4f9, font-weight:bold, font-size:14px; classDef experiments fill:#f9f9d4, font-weight:bold, font-size:14px; classDef localization fill:#f9d4f9, font-weight:bold, font-size:14px; A[Christoph Kapeller] --> B[Kapeller: G-Tech's invasive dept.
lead on neurotech. 1] A --> C[ECoG: 20-64 channel grids
record high frequencies. 2] C --> D[Stereo EEG: depth electrodes
for deep structures. 3] C --> E[ECoG: higher amplitude, frequency,
resolution than EEG. 4] C --> F[ECoG: subdural during craniotomy.
Stereo EEG: through screw holes. 5] C --> G[Electrodes: passive head box
connections to amplifier. 6] C --> H[Intracranial signals: stronger amplitude,
higher frequencies than EEG. 7] C --> I[High gamma: focused activation.
Low frequencies: widespread suppression. 8] C --> J[Past demos: ECoG cursor,
games, robotic control. 9] A --> K[Experiment considerations: opportunity, time,
protocol, channels, processing, analysis. 10] K --> L[Grounding, referencing, line noise
crucial for signal quality. 11] K --> M[Impedance checks identify bad
channels. ORs noisier than units. 12] K --> N[Sampling rates =1200Hz. 24-bit
ADCs for wide range. 13] K --> O[Experiment design: XML protocol,
raw viz, online processing. 14] A --> P[Offline review: signal quality,
re-referencing, task intervals. 15] P --> Q[Time-frequency analysis: 5-145Hz. Grid
mapping of task biomarkers. 16] P --> R[Classifier training: CSP filtering,
band-power, LDA. 17] P --> S[Online BCI: spatial/temporal filtering,
high gamma, classification. 18] P --> T[Real-time validation: fast decoding
of faces vs. symbols. 19] P --> U[Continuous decoding for robotic
arm control from motor cortex. 20] A --> V[Anatomical localization: pre-op MRI,
post-op CT, brain segmentation. 21] V --> W[Electrodes snapped to brain
surface. Stereo EEG along shafts. 22] V --> X[3D electrode coordinates and
brain region assignment. 23] V --> Y[Group-level localization on templates
for biomarker, symptom topology. 24] V --> Z[Enables stimulation mapping, passive
BCIs. Functional mapping use case. 25] V --> AA[Impedance, epileptic activity, connectivity
analysis visualization. 26] A --> AB[Core pipeline: recording, filtering,
normalization, classification, output. 27] A --> AC[Careful design allows effective
ECoG BCIs from OR to chronic. 28] A --> AD[Localization and multi-modal integration
help interpret mechanisms, localize function. 29] A --> AE[Stereo EEG generally safe
except bleeding risk. 30] class A,B kapella; class C,D,E,F,G,H,I,J ecog; class K,L,M,N,O,P,Q,R,S,T,U signalProcessing; class V,W,X,Y,Z,AA,AD localization; class AB,AC,AE experiments;

Resume:

1.-Christoph Kapeller leads G-Tech's invasive department, working on cortical technologies, intracranial electrical stimulation, and neuromodulation experiments using ECoG implants.

2.-ECoG grids with 20-64 channels are implanted on cortical regions by neurosurgeons to record high gamma and other oscillations up to 1kHz.

3.-Depth electrodes (stereo EEG) can be inserted into the brain to record from deep structures, in addition to ECoG.

4.-ECoG has higher amplitude, frequency range, and spatial resolution compared to EEG. Stereo EEG and ECoG have similar electrode properties.

5.-ECoG grids are placed subdurally during craniotomy. Stereo EEG electrodes are implanted through screw holes without removing the skull.

6.-Electrodes have passive connections to a head box which connects to the biosignal amplifier. Quick connectors can interface multiple electrodes.

7.-Intracranial signals have stronger amplitude and higher frequency range than EEG. ECoG and stereo EEG can record up to 1kHz.

8.-High gamma band shows focused cortical activation during tasks, while low frequencies show widespread suppression. High gamma relates to neural firing.

9.-ECoG cursor control, video game interfaces, and robotic control have been demonstrated using motor cortex signals in the past.

10.-Clinical opportunity, limited time, protocol design, channel assignment, control computer, processing platform, and analysis pipeline are key experiment considerations.

11.-Ground and reference electrode selection, connection to earth ground, and proper line noise management are crucial for good signal quality.

12.-Impedance checking helps identify bad channels. Operating rooms introduce more line noise interference than monitoring units. Proper referencing helps.

13.-Sampling rates of 1200Hz and higher are used. 24-bit ADCs allow wide input range. Passive cable connections go to the amplifier.

14.-Experiment design uses an XML-based task protocol loaded into GHYSIS/Simulink. Raw data visualization and online processing are set up.

15.-Offline data review in MATLAB with GBS Analyze checks signal quality, re-references data, and defines task-related analysis intervals.

16.-Time-frequency analysis from 5-145Hz and topographic mapping to grid layouts help visualize task-related high gamma biomarkers across the grid.

17.-Classifier training data is extracted, spatially filtered with CSP, band-power extracted, and used to train an LDA classifier.

18.-Online BCI output is generated by spatially and temporally filtering data, extracting normalized high gamma band-power, and classifying it.

19.-The real-time BCI is validated with the subject, showing fast decoding of presented faces vs symbols from untrained images.

20.-The same signal processing pipeline is applied for continuous decoding for robotic arm control based on motor cortex activity.

21.-Anatomical localization of electrodes on the brain is done by co-registering pre-op MRI, post-op CT and segmenting the brain.

22.-Electrodes are snapped to the brain surface segmentation to correct for brain shift. Stereo EEG electrodes are localized along shafts.

23.-3D coordinates of electrodes are determined and assigned to brain regions. fMRI co-registration can map function and symptoms.

24.-Group-level anatomical localization on template brains allows mapping biomarker and symptom topology across subjects with variable grid locations.

25.-Effective stimulation mapping and passive BCI experiments are enabled. Diagnostic functional mapping is an important use case.

26.-Impedance and epileptic activity can also be visualized spatially. Connectivity analysis can be applied using BCIs.

27.-The core pipeline is ECoG/stereo EEG recording, spatial and temporal filtering, normalization, classification, and output generation for real-time operation.

28.-Careful design of protocols, signal processing and validation allows for effective ECoG BCIs from the operating room to chronic implants.

29.-Electrode localization and integration with other brain mapping modalities helps interpret mechanisms and localize function for many use cases.

30.-Despite penetrating the brain, stereo EEG appears generally safe other than bleeding risks. Symptoms specifically from the shafts are not obvious.

Knowledge Vault built byDavid Vivancos 2024