The End Of Knowledge - Vault 3 - G.TEC BCI & Neurotechnology Spring School 2024 - Day 3 - Leo Schreiner, Matteo La Rosa, Pauline Schomaker

graph LR classDef pangolin fill:#f9d4d4, font-weight:bold, font-size:14px; classDef density fill:#d4f9d4, font-weight:bold, font-size:14px; classDef resolution fill:#d4d4f9, font-weight:bold, font-size:14px; classDef experiments fill:#f9f9d4, font-weight:bold, font-size:14px; classDef applications fill:#f9d4f9, font-weight:bold, font-size:14px; A[>g.Pangolin 1024] --> B[ ultra-high density
EEG, flexible PCBs 1] B --> C[1024 channels, pre-amplifiers,
better signal-to-noise 2] B --> D[8.6mm inter-electrode distance,
5.9mm sensor diameter 3] B --> E[Significant improvement over
standard 10-10 systems 4] B --> F[Setup process: prepare, montage,
apply, experiment, co-register 5] A --> G[Higher density EEG
benefits 6] G --> H[Epileptic source imaging,
visual processing studies 6] G --> I[Separated activation for
individual finger movements 7] I --> J[Significant channels per finger,
standard EEG broad activation 8] G --> K[2.5x signal correlation increase,
improved information content 9] A --> L[Hand gesture decoding
experiments 10] L --> M[Rock, paper, scissors
classification, execution/imagery 10] L --> N[78.9% accuracy 1s post-onset,
2 subjects tested 11] A --> O[Source reconstruction
with G.pangolin 12] O --> P[256 channels, median
nerve SEP example 12] O --> Q[FieldTrip toolbox, dipole fit
in left sensory cortex 13] O --> R[Whole scalp coverage
recommended, more methods 14] A --> S[Individual finger movement
study, 5 subjects 15] S --> T[Focal activation, 70.6%
pairwise classification accuracy 15] A --> U[Mapping central sulcus,
median nerve SEP 16] U --> V[95.2% accuracy classifying
post-/pre-central channels 16] A --> W[Grasping movement
representation study 17] W --> X[Power, precision, pinch grasps,
54 repetitions each 18] W --> Y[236 EEG, 96 EMG electrodes
per participant on average 19] W --> Z[Motion capture, MVC recorded,
ERD patterns analyzed 20] W --> AA[EMG 95%, EEG 75% peak
3-class grasp accuracy 21-22] W --> AB[Advanced analysis may
improve results further 23] A --> AC[Motor & visual tasks,
wider scalp coverage 24] AC --> AD[Temporal activations found
for thumb vs pinky 25] AC --> AE[Visual paradigms evoked
clear region activations 26] AE --> AF[Similar accuracies, informative
maps with higher density 27] A --> AG[512-channel visual
experiment demo 28] AG --> AH[Setup, skin prep, gel
application, grid placement shown 29] A --> AI[Reusability, visualization, triggers,
wireless discussed 30] class A,B pangolin; class C,D,E density; class F,G,H,I,J,K resolution; class L,M,N,O,P,Q,R,S,T,U,V,W,X,Y,Z,AA,AB,AC,AD,AE,AF,AG,AH experiments; class AI applications;

Resume:

1.-Leo presented the G.pangolin system, an ultra-high density EEG system with flexible PCBs attached to the scalp using medical adhesives.

2.-The system uses pre-amplifiers for better signal-to-noise ratio and can use up to 1024 channels by merging 4 G.HIamps.

3.-G.pangolin grids have 8.6mm inter-electrode distance and 5.9mm exposed sensor diameter. Conductive gel is used to fill the sensors.

4.-Compared to standard 10-10 or extended 10-10 systems, G.pangolin provides a significant improvement in spatial density of sensors.

5.-The process involves preparing grids, creating montage, applying grids and pre-amplifiers, running experiments, and co-registering electrode positions with MRI.

6.-Higher density EEG is important for source imaging in epileptic patients and studying spatio-temporal aspects of visual processing.

7.-Research shows increased spatial resolution with G.pangolin reveals separated EEG activation for individual finger movements compared to standard EEG.

8.-In finger tapping experiments, G.pangolin found significant channels for each finger, while standard EEG showed broad activation.

9.-Despite increased inter-electrode distance, the correlation between signals only increased by 2.5 times, indicating improved information content.

10.-Hand gesture decoding experiments classified rock, paper, scissors hand movements using G.pangolin in both motor execution and imagery.

11.-Classification accuracy reached 78.9% one second after movement onset. Subject 1 did motor execution, Subject 2 did motor imagery.

12.-G.pangolin data can feed source reconstruction algorithms. An example used 256 channels and median nerve stimulation SEP.

13.-The inverse problem was solved using the FieldTrip toolbox. Dipole fit localized right median nerve stimulation in left sensory cortex.

14.-Covering the whole scalp is recommended for best source reconstruction results. More methods can be applied.

15.-Five subjects performed individual finger movements. Significant channels showed focal activation. 70.6% average pairwise classification accuracy was achieved.

16.-Mapping the central sulcus used median nerve stimulation SEP. 95.2% accuracy in classifying post- and pre-central channels was achieved.

17.-Paulina studied representation of different grasping movements (power grasp, precision grasp, pinch) on objects using G.pangolin for BCIs/neuroprosthetics.

18.-The paradigm included preparation, execution until hold, and release phases. Each grasp-object combination was executed 54 times.

19.-On average, 236 EEG electrodes mainly on motor areas and 96 EMG electrodes on arm/hand muscles were used per participant.

20.-Motion capture and maximum voluntary contraction were also recorded. ERD patterns were analyzed over time in EEG.

21.-EMG, motion capture and EEG/EMG decoding results were compared over time. Reaction times and accuracy patterns were observed.

22.-EEG had 75% peak accuracy, EMG had 95% sustained accuracy for 3-class grasp decoding within an object. Some participant differences noted.

23.-Further improving results is anticipated with more advanced analysis. High-density coverage contributes to grasping movement research for BCIs/prosthetics.

24.-Matteo investigated benefits of wider scalp coverage with G.pangolin for motor execution/imagery and visual tasks.

25.-Covering temporal regions beyond sensorimotor cortex revealed some relevant activations for thumb vs pinky movement discrimination.

26.-Visual paradigms with different color/contrast stimuli evoked clear visual region activations, with differences observable using varying electrode densities.

27.-Accuracies for visual decoding were similar but topographic maps were more informative with higher density.

28.-Live demonstration of 512-channel visual experiment conducted. Photo sensor detected stimulus onsets for precise ERP timing.

29.-G.pangolin setup, skin preparation, conductive gel application, and grid placement were shown. Impedance reduction techniques were discussed.

30.-Reusability of grids, visualization of high-density EEG signals, trigger mechanisms, and wireless transmission constraints were addressed in Q&A.

Knowledge Vault built byDavid Vivancos 2024