FNIRS Flashcards
- fNIRS: functional principle, optodes, light
absorption spectra, (de-)oxygenated
haemoglobin
Measures changes in electrochemical activity and blood levels through their effect on optical properties: specifically, the changes in oxyhaemoglobin and deoxyhaemoglobin ratio.
Most body tissues are relatively transparent to light in near-infrared range (700-1000nm)
● Oxy-Hb and deoxy-Hb have different optical properties and different NIR light absorption
spectrums
o Spectral band: the optical window
o Oxy-Hb= 760nm, deoxy-Hb= 850nm
Light source, optode ● Typically, LEDs Detectors ● Placed 2-7cm away from optode, can detect light after travelling through tissue. Varying the distance changes the measurement depth
- fNIRS: spatial resolution, temporal
resolution, limitations
Whats amazing about fNIRS?
Time resolves and frequency domain systems
Provides information on phase and amplitude.
+ Precise, more information on time and frequency
- relatively poor spatial resolution
+ portable and applicable in natural settings ( is not affected by movement)
- BCI: purpose and functional principle,
suitable brain measures (EEG, fNIRS, fMRI,
etc.),
Interface which connects brain to computer, allows control over external devices without relying on motor input
How it works
1. Measure brain activity with functional neuroimaging method
2. Extract specific features
Anything useful about signals controllable by BCI: activation level, connectivity measure, etc.
3. Translate features into commands that operate a device using rule-based algorithms
Specific temporal/ spatial characteristics = specific output
Machine learning algorithms improve accuracy of prediction/ classification
Requires repeated acquisition of functional data from a subject to optimise the algorithm
4. Command transferred to computer
5. Neurofeedback: overall task performance relayed to individual as sensory feedback
Allows user to regulate state of specific brain function to achieve better performance
Different BCI Systems
EEG based BCI :
Use P3300: computer detected when elicited by presentation of matric elements containing chosen alphabetical characters
● Slow cortical potentials and steady state visual-evoked potentials are also used, as well as motor-imagery related synchronisation
MEG based BCI:
Used in real time BCI
● Can classify over spatial attention around a fixation point through the
modulatory properties of posterior alpha rhythms
● 90% classification accuracy for motor and motor imagery tasks
● Better spatial resolution and better SNR than EEG
fmri based BCI:
Used in real time BCI
● Can classify over spatial attention around a fixation point through the
modulatory properties of posterior alpha rhythms
● 90% classification accuracy for motor and motor imagery tasks
● Better spatial resolution and better SNR than EEG
Nirs based BCI:
Left versus right motor imagery
● Hemodynamic response corresponding to P300 component
● Binary subjective preference evaluated on a single trial basis of NIRS
signal during decision making (80%)
● Direct encoding: think of what you want to say/ do: difficult
Indirect encoding: use of easily differentiable activity (motor imagery
versus inner speech)
● Can use spatial features: different kinds of mental activity
Can use temporal features: start responding after a certain delay Can use magnitude features: let subject reach a certain level of brain activation
- BCI: role of signal-to-noise ratio, data
processing steps (multivariate pattern
analysis / machine learning/ statistics),
relevant outcome measures (decoding
accuracy and speed)
Signal to noise ratio:
How much signal is measured compared to noise?
Why is this important in BCIs?
BCIs aim to replace the motor pathways or to enable communication with unresponsive patients
We need functional imaging data as an input
E.g., participant performs a mental imagery task or a motor task
Different methods (EEG, MEG, fMRI, fNIRS) each with its own strengths & weaknesses
The more noise in the data we less useful the BCI for later application
What is multivariate pattern analysis (MVPA)?
MVPA = information stored in a pattern of activity
Some areas decrease in activity others increase 🡪 analyzes the entire pattern of activation
Input obtained needs to be classified to generate the corresponding BCI response
What is Machine learning? How does it work?
Give feedback to the computer to increase the accuracy
Task repeated multiple time until performance of BCI is optimized
Neurofeedback
- BCI: practical value, example applications
(locked-in syndrome, control of external
equipment)
Study by Sorger:
Left versus right motor imagery
● Hemodynamic response corresponding to P300 component
● Binary subjective preference evaluated on a single trial basis of NIRS
signal during decision making (80%)
● Direct encoding: think of what you want to say/ do: difficult
Indirect encoding: use of easily differentiable activity (motor imagery
versus inner speech)
● Can use spatial features: different kinds of mental activity
Can use temporal features: start responding after a certain delay Can use magnitude features: let subject reach a certain level of brain activation
2 performed to form “channel” to localise optode pair. Activity patterns evoked can be best discriminated. Data analysis focused on this “best optode pair”/ channel
● Cut off between yes and no
o Mean of t-values of “yes” and “no” trials from first answer-encoding run
o Midpoint between average amplitude of “yes” and “no” runs used as classification
criterion in the other runs. FNIRS response amplitudes below the midpoint were classified as “no”
Results
● Half of patients reached >70% accuracy: sufficient performance for binary communication
BCIs
o Multi-trial accuracy not much greater than single trial accuracy
o Large individual differences: BCI effectiveness depends on the person
