The End Of Knowledge - Vault 3 - G.TEC BCI & Neurotechnology Spring School 2024 - Day 2

graph LR classDef present fill:#f9d4d4, font-weight:bold, font-size:14px; classDef hse fill:#d4f9d4, font-weight:bold, font-size:14px; classDef solutions fill:#d4d4f9, font-weight:bold, font-size:14px; classDef talk fill:#f9f9d4, font-weight:bold, font-size:14px; classDef brain fill:#f9d4f9, font-weight:bold, font-size:14px; classDef harpool fill:#d4f9f9, font-weight:bold, font-size:14px; classDef challenges fill:#f9d4d4, font-weight:bold, font-size:14px; A[Alex Ossadtchi] --> B[Alex: TMS talk at
Christoph's BCI school. 1] A --> C[Alex at Higher School
of Economics, Russia. 2] C --> D[Team: ML for compact
brain magnetic sensors. 3] A --> E[Track brain for processing
based on state. 4] E --> F[Brain rhythms: processing context,
excitatory/inhibitory states. 5] F --> G[Rhythms affect perception, attention,
memory, language, synchronization. 6] F --> H[Brain states transient: influence
motor, memory, simultaneity. 7] F --> I[Debate: rhythms impact
behavior/cognition. 8] F --> J[Tools needed: study complex,
hierarchical, rhythmic brain. 9] A --> K[Closed-loop: delays hinder
precise phase tracking. 10] K --> L[Filtering delays: real-time
algorithms needed. 11] A --> M[HARPOOL: ultra-low latency
EEG oscillation tracking. 12] M --> N[Runs on EEG, real-time,
triggers at phases. 13] M --> O[State space tracking outperforms
autoregressive modeling. 14] M --> P[HARPOOL: state space, 1/f
noise, real-time. 15] M --> Q[Phase and amplitude important:
state space computes both. 16] M --> R[Validated with simulations, data,
phantoms: modulates MEPs. 17] M --> S[TMS cortical muscle mapping:
phase affects MEPs. 18] S --> T[Excitatory phases: spatially disentangled
muscle motor representations. 19] M --> U[Enables TMS network exploration,
diagnostics, mapping, TACS, neurofeedback. 20] A --> V[Challenges: heart/brain tracking, TMS
coils, EEG resolution. 21] V --> W[Future: high-density EEG, inverse/
forward modeling for resolution. 22] V --> X[TMS phantom variability: use
sensorimotor rhythm subjects. 23] V --> Y[Modeling for optimal EEG/TMS:
refine phantom setup. 24] V --> Z[HARPOOL: add heart-brain condition
to reduce variability. 25] V --> AA[HARPOOL for fNIRS, fMRI-EEG
neurofeedback: hypotheses needed. 26] A --> AB[ECOG speech mapping: align
passive/active via ML. 27] AB --> AC[Decoded pronounced words: imagined
speech failed so far. 28] AB --> AD[Precise electrodes crucial: legal
limits hinder recruitment. 29] A --> AE[Alex, Christoph: collaborate further,
meet when possible. 30] class A,B,AE present; class C,D hse; class E,F,G,H,I,J brain; class K,L,M,N,O,P,Q,R,S,T,U,Z harpool; class V,W,X,Y,AA,AB,AC,AD challenges;

Resume:

1.-Alex presented on precisely triggering TMS experiments for medical applications at the BCI school hosted by Christoph.

2.-Alex works at the Higher School of Economics in Russia, which has grown into a large university with various departments.

3.-Alex's team develops machine learning solutions for analyzing data from compact, mobile sensors of magnetic fields around the brain.

4.-The talk focused on tracking brain readings accurately to determine how the brain processes external information based on its state.

5.-Brain rhythms (alpha, theta, delta, beta, gamma) determine the brain's context for processing information and reflect excitatory/inhibitory states.

6.-Studies show brain rhythms influence perception, attention, memory, language; synchronization of brain regions is key for communication.

7.-Brain states are transient; motor evoked potentials, memory performance, and simultaneity judgments are modulated by brain oscillation phases.

8.-Some studies question if rhythmic activity influences behavior/cognition; the brain is nonlinear and changes on small time scales.

9.-Proper tools are needed to study the brain's complex, hierarchical functional systems that switch rhythmically at different paces.

10.-In closed-loop systems, delays from processing EEG data to generating stimuli hinder precise phase tracking of brain oscillations.

11.-Filtering to extract rhythms causes output delays; new algorithms are needed to banish delays and work in real-time systems.

12.-Alex's team developed a hardware-powered ultra-low latency system (HARPOOL) to track oscillatory brain states on an EEG device.

13.-HARPOOL runs on the EEG device processor, avoids PC data transfers, works in real-time, and triggers stimuli at target phases.

14.-State space tracking using Kalman filters and modeling observation noise outperforms previous autoregressive forward modeling approaches.

15.-HARPOOL extends state space tracking, models 1/f brain noise, and is implemented on an EEG device for real-time phase tracking.

16.-Both oscillation phase and amplitude are important; state space approaches allow efficient computation of instantaneous amplitude and phase.

17.-HARPOOL was validated with simulations, real data, and phantom experiments; it showed expected modulation of motor evoked potentials.

18.-HARPOOL was used for cortical muscle representation mapping with TMS; inhibitory phases yielded weaker motor evoked potentials.

19.-Excitatory phases revealed spatially disentangled muscle representations in motor cortex compared to phase-agnostic or inhibitory stimulation.

20.-HARPOOL enables TMS brain network exploration, diagnostics, and efficient mapping; it can also guide TACS phase and neurofeedback.

21.-Challenges remain, including combining heart and brain activity tracking, positioning TMS coils, and EEG spatial resolution limitations.

22.-High-density EEG with up to 1024 electrodes, combined with inverse/forward modeling, may improve spatial resolution in the future.

23.-The team's TMS phantom experiments had some variability; they plan to use subjects with pronounced sensorimotor rhythms for refinement.

24.-Modeling can help determine optimal EEG electrode height for TMS; peeling watermelon skin may improve their phantom setup.

25.-Heart-brain interaction can be incorporated as an extra condition in HARPOOL to reduce variability in brain state tracking.

26.-The HARPOOL approach could work for fNIRS and fMRI-compatible EEG neurofeedback, but hypotheses are needed for the latter.

27.-Alex's team is working on ECOG-based speech mapping, aligning passive and active cortical mapping results using machine learning.

28.-They achieved reasonable accuracy decoding pronounced words from pre-speech ECOG data, but failed with imagined speech so far.

29.-Precise electrode placement is crucial for imagined speech decoding; legal limitations in Russia hinder optimal subject recruitment.

30.-Despite travel difficulties, Alex and Christoph hope to collaborate further and meet in person when circumstances allow.

Knowledge Vault built byDavid Vivancos 2024