Please enable JavaScript.

Coggle requires JavaScript to display documents.

FBI - MEG - Coggle Diagram

- - - - Requires lots of correction and a fancy room
      - 10 ^-13 T brain signal strength
        
        Comparable to a screwdriver 5m away, or a car 2km away
    - - Action potentials NOT measurable
        
        they diminish too quickly
      - Evoked responses work best for simple resonses
        
        eg visual stimulation
        
        Habituation is a factor
    - - Tangential
        
        Similar to EEG but rotated 90 degrees and smaller
        
        Magnetic field is perpendicular to neuron and current
        
        Right hand rule
      - Radial
        
        Magnetically silent. Not seen at all
    - - (most of) cortex easy to detect
      - Deep sources difficult to detect
        
        Does show up both using virtual electrodes and real ones. Just weak
      - Generally deeper sources require more source strength to be detected
  - - - Orthogonal in measurement signals
      - MEG not impacted by skull and other tissue in the same way
    - - values are absolute and have meaning
  - - - 306
        
        ~320,000 data points per sensor within 5 mins
      - Need to be cooled to superconducting levels
        
        via liquid heliun
      - Not typically worn/portable. Big ass machine
      - SQUID sensors
        
        developed by Cohen and Zimmerman
    - - First order gradiometer / reference gradiometer
        
        Further sources are more homogenous
        
        Can account for using coil pairs with opposite set ups (wound in opposite directions) to cancel out
        
        Brings noise level down enough to measure brain shift
    - - Made of mu metals and aluminium
        
        shields from external frequencies
  - - - Similar to EEG but more flat topped instead of spiky
    - - Similar to EEG
    - - Similar to EEG. Gamma range. Sucks to deal with
    - - slow hill shapes
    - - Looks like ECG idk
    - - Watches will be 1Hz clean signals
- - - - Better spatial resolution
  - - - On walls
        
        Reduces current and allows for movement with manageable artefacts
        
        Brings noise down by a factor of 10
      - 1.2km of wire
    - - can put sensors in further depending on subject head size
- - - - eg to Cz
    - - chained eg double banana shape
      - Helps narrow down specific location of negative maximum
        
        eg in epilepsy
    - - Weighted averaging. Average reference based on distance.
        
        Closer contributes more to average
        
        Aim to isolate electrodes from surrounding
  - - - Interictal
        
        Mixture of sharp spikes and slow waves
        
        Not a seizure but can spark one
      - Ictal
        
        Rhythmic >2.5Hz
        
        Evolves in amplitude/freq
        
        May spread
        
        During seizure itself
- - - - Loss of connections and shift towards random network
        
        Functional and structural connectivity loss
      - Hub vulnerability
        
        AD shows targetted hub-focused damage
        
        Not AD specific
      - Slowing
        
        greater theta power, lower alpha and beta
      - Vicious cycle of hyperactivation
        
        Activation -> amyloid deposition -> structural atrophy -> functional compensation -> overloading of hubs
        
        inhibitory to excitatory E/I balance
        
        Hyperexcitability as early mechanism for neurodegeneration
        
        Epilepiforms seen predominantl during sleep
      - Ex machina project
        
        model guided tDCS for AD
        
        Changes excitablity
        
        Basically info for personalised tDCS
    - - More of a shift towards regular networks
      - MEG is roughly normal
    - - Severe slowing
        
        Peak in theta ~4-8Hz
      - FIRDA: frontal intermittent rhythmic delta activities
  - - - Increased atrophy
      - Increased thalamus-cortex functional connectivity
        
        Associated with worse cognition and disability
      - Is a hub. relay station
    - - More differnence in conenctivity of nodes
    - - less efficient
    - - Cognitive impairment once critical threshold of structural damage is surpassed
      - Structural-functional coupling
        
        Higher coupling means less FC diversity
        
        In MS, only long range connections show higher s-f coupling
        
        Long range ones are more difficult to correct for
      - Neuronal dynamic modelling
        
        Shows that disconnection induces higher FC
        
        Increases clustering coefficient -> more modular isolation
        
        Consistently higher in men
    - - Theoretically universal theory
      - Generallyhigher hub FC in pwMS
        
        Capacity still limited -> death from overload -> more strain on neighbouring nodes/hubs
  - - - Helps avoid having to be awake during surgery
        
        Largely reproduces IAP results
    - - Affected by tumours, often spreading far from visible tumour
      - other frequency bands also affected
        
        higher frequencies more affected
    - - Increased clustering associated with higher seizure vulnerability
    - - ~20%
      - More initial lateralisation shows more shifts later
    - - More activity -> more tumour spreading
      - Indirect and direct communication between tumour and brain -> more tumour growth
    - - Lower MEG power -> longer progression-free survival, even when controlling for other covariates
      - MEG power used to monitor tumour growth
        
        But imminent growth not picked up cross-sectionally
  - - - Epileptic zone (EZ)
        
        theoretical area that needs to be removed
      - Resection area (RA)
        
        Area actually removed
        
        Has high hub status
        
        Maybe remove area connected to pathological hub instead of the hub itself?
        
        Removing hubs good for surgical outcome but not for patient long term
        
        Tough to distinguish between pathological and cognitive hubs
      - Irritative zone
        
        where epileptiforms are detected
        
        High frequency oscillation zone (HFO)
        
        w/in irritative.
        
        80-500Hz
        
        Seizure onset zone
        
        Where seizures begin
        
        Overlaps with HFO. Within irritative
    - - Helps distinguish between RA and non-RA regions
      - Does not help with seizure free vs non seizure free
- - - - Optimal balance of integration and specialisation
    - - Random
    - - Everything has equal connections
  - - - use of patient data in model
        
        used to predict prognosis
  - - - hub score
- - - - Coils and head shape digitised with Polhemus system
        
        Decreases coregistration error w/ MRI from ~1cm to ~4mm
  - - - Equivalent current dipole
    - - ~sphere
  - - - Radial not detected
      - Opposite orientations not detected
      - More sources than sensors (10,000 vs 300)
    - - Dipole fitting
        
        Assumption that there are only a few sources w/ limited spatial extent
        
        Valid in certain situations
        
        best when studying first x00ms of average ER
        
        Cost-function used to measure degree that model matches daa
        
        Choice of cost function is important
        
        Squared difference or % variance
        
        Chi-squared -> takes noise into account
        
        Reduced chi-squared --> also reduces number of parameters modelled
        
        Compute forward problem and then compare
        
        Different forward problems. Update and change dipoles' parameters
        
        (location, orientation and strength)
        
        Limitations
        
        User dependent
        
        Different starting points will impact results
        
        Searches for local minimum
        
        Different assumptions, guesses, # of sources, latencies etc
        
        Need large SNR
        
        Use of averaged signal to compensate
        
        Higher cognitive functions are poorly phase locked
        
        Monte Carlo simulation used to assess sensitivity to noise
        
        Add noise --> model --> repeat
        
        95 CI of dipoles based on solutions found.
        
        Choice of dipole model:
        
        Single latency
        
        Fixed parameters: location, orientation, and strength
        
        Multiple latencies
        
        Moving dipole model
        
        Location and orientation can vary
        
        Rotating dipole model
        
        Location fixed. Orientation variable
        
        Fixed dipole model
        
        Fixed location and orientation. Amplitude variable
        
        using this we can indeed tell our arse from our elbow
      - Minimum norm
        
        LORETA
        
        Assumes that neuronal activity is spatially smooth
        
        Limits spread
        
        Other constraints can be added with different solutions and assumptions
        
        FOCUSS, VARETA, MUSIC, DICS etc
        
        Chooses the solution with minimum power required from sources
        
        Assumption that brain is efficiently evolved to use minimal energy
        
        Thus biases towards closer locations (cortex)
        
        Compensate by rewarding depth (depth weighting)
        
        This has a bit of a fiddle factor
        
        Usable with both averaged and non-phase locked responses -> evoked and spontaneous
        
        Requires predefined source space (eg pyramidal layer)
        
        Use of automatic segmentation via freesurfer
        
        Lacks variation in location and orientation. Only varies in amplitude/strength
        
        Limitations
        
        Bias towards superficiality (compensated for)
        
        Solutions are spread out and difficult to interpret
        
        Add constraints ((LORETA, FOCUSS, Dynamical statistical parametric mapping etc)
        
        Source space choice bias
        
        eg hippocampal activation would be attributed to the temporal cortex
      - Beam forming
        
        SAM
        
        no two neuronal sources are perfectly correlated
        
        To study averaged responses and cortical oscillations
        
        Mostly user independent
        
        Gotta choose frequency band to analyse
        
        Decent correspondence to fMRI
        
        Lower resolution but can look at different frequency bands
        
        Also good correspondence with anatomical MRI -> follows sulci
        
        Weighting/virtual radar output
        
        Different weighting to prioritise x sensors for x regions
        
        repeat per voxel in the brain
        
        Sensitivity changes with different data
        
        Spatial filter
        
        Ideally lets through ROI and blocks others
        
        Ideal is a hard cutoff. Realistic, general filter is more of a gradient
        
        want to minimise AUC
        
        2 constraints:
        
        Unity gain
        
        Minimum beamformer output
        
        Minimise power of spatial filter
        
        Optimised:
        
        Lets in as much for ROI and as little for potential interference, and somewhere between for rest
        
        Need data covariance and forward solution per voxel of frequencyband of interese
        
        Limitations
        
        Cannot detect perfectly correlated sources
        
        ie with evoked responses
        
        Potential solution?
        
        Look for 2 sources at a time --> alter unity gain constraint
        
        Issue is that you don't know beforehand when 2 will be perfectly correlated
        
        Computationally intense
        
        Dipole source sensitivity
        
        Good with point source but poor with extended source
        
        Gets localised worse with stronger signals
        
        Head model sensitivity
        
        Better to use a more complex model than single sphere
        
        Multi-sphere model
        
        Sphere per sensor
        
        Complex but solveable
        
        Need enough recording for accurate data covariance