fNIRS Flashcards
What is near-infrared spectroscopy?
Neuroimaging technique that uses light to determine brain activity
Device is physically similar to an EEG setup (cables attached to the head and sending the signal somewhere else)
• fNIRS is a neuroimaging technique that uses light
• fNIRS is sensitive to outer cortical region of the brain
• fNIRS is based on similar principles to MRI
• Most devices measure change in haemoglobin, from which neural activity is inferred.
what are the principles of fNIRS?
Shine near infrared light (not fully infrared, travels well through body), some passes through skin and bone and into brain and we can measure it
Oxy and deoxy blood absorb red light diff so we can infer [oxy vs deoxy blood]
Source optode emits light that goes through skull, some goes through brain and back out through skull and to a detector
Measures brain activity in a small region
Not all the light will go through the path from the light source to the light detector, some will be lost => the signal we’re looking for is actually quite small
• Light sources emit at two wavelengths.
• Based on light received at detector we can infer changes in oxy- and deoxy-haemoglobin
How can fNIRS go beyond the lab?
Trying to commercialise this technology (user friendly)
Whole head FNIRS device that wear all day and
measure brain activity
Moving towards whole head but modular design that is more comfy
What are different devices used for fNIRS?
Variety of devices that go from lab based, some that are aimed to scan specific parts of the brain or targeting very specific and small regions (the ear; everyday type devices)
What are cochlear implants and their importance in fNIRS?
Practicle considerations: cochlear implant = device that restores hearing by internally (under skin), electrodes in cochlea that prod electric current that excites the nerve to prod perception of sound, external part is what capts the sound
Cant use these with MRI or MEG, and the elect signals by the cchlear implants corrupt the EEG signals
So we use FNIRS
What are 6 benefits of fNIRS?
• Easy to set up and child friendly
• More accessible than MRI & MEG
• More robust than EEG to movement
• Only probes outer brain (feature and a bug)
• Can be used at the same time as other methods (EEG)
• No electrical artifacts from cochlear implants
What are the spatial and temporal resolutions of fNIRS?
Seconds range for temporal res (poor compared to
EEG and MEG)
Poor spatial res
Non invasive
What are the biological principles behind fNIRS?
Similar to MRI, theres a typical amount of oxy and deoxy blood at rest, when theres a stim, theres an aug of é demand to that region so more blood goes there (BOLD response) and thats what we measure
• Your brain needs blood in order to function. Blood carries oxygen and glucose to the brain as fuel for your neurons.
To fuel the neurons that activate, Hb is pumped to that region
• Haemoglobin is the specific protein that oxygen binds to on your red blood cells. As your brain uses up the oxygen, the haemoglobin on your red blood cells goes from being oxygenated to deoxygenated.
• By measuring the relative amounts of oxygenated and de-oxygenated blood we have a proxy measurement for brain activity.
2 wavelengths are absorbed dif by oxy and deoxy Hb and thats how we measure diffs in the 2 [c]
• Light source emits light at two specific wavelengths
• Photons bounce off in all directions and some photons make it back to the detector
• The two wavelengths are absorbed differently depending on the concentration of oxy- and deoxy-haemoglobin
• Can estimate the concentration of oxy- and deoxyhaemoglobin between source and detector
What are the 3 types of fNIRS?
- Continuous waves NIRS
- Frequency domain NIRS
- Time domain NIRS
What is continuous waves NIRS?
Emit light, detector measures how much light comes back
Most common
Put detectors at diff depths to measure diff distances
What is frequency domain NIRS?
Modulate light, light goes through tissue, is attenuated
and slowed down (shifted in the medium) so get f info
about the tissue that the light went through
What is time domain NIRS?
Send pulse of light (not continuous), some ligth will go a short distance and come back fast, other light will goo deeper and take longer to come back so can know the depth that we’re measureing but the deeper the light goes, the more signal loss there is
What is the Haemoglobin Absorption Spectra?
CONVERSION TO HAEMOGLOBIN CONCENTRATION ESTIMATES
Shine light at 2 diff wave lengths (measure at 2 wavelengths),
Absorption coefficients of oxy and deoxy Hb
Info we have is the amount of light at the 2 f and calculate from that the amount of oxy and deoxy blood at diff t points
What is the typical pipeline for fNIRS experiments?
1) Experimental design
2) fNIRS data acquisition
3) fNIRS data pre-processing: intensity to OD conversion, OD to concentration conversion, motion artefact correction, filtering
4) Statistical inference
What does a practical fNIRS set up look like?
Each optode (source that has 2 LEDs each with one wave length and detectors) Signal is converted to electrical signal or digital signal Up to 16 sources and 16 detectors, can be uncomfortable bc of all the cables (practical considerations to get a clean signal) If optodes press on head to much and make it red and inflamed, could get confused on what is brain signal and what is inflammation Optode is placed in grommets and have to make sure that theres good physical contact with head (vs EEG that has gel). if someone has a lot of hair its harder to get a good signal, have to separate hair to get good contact 1 source and detector 3 cm apart to make sure light has room to go in brain and come back
What are the experimental design options?
Unlimited
- Block design: Present same stim/same task for a period of t
then break then diff stim. If plot signals, get traditional haemodynamic responses, stim signal goes up and then back down
- Slow event related: Brief condition followed by long gap, its a long exp and is less natural. Response is smaller
- Rapid counterbalanced event related: Each condition in rapid succession, disintangle conditions in later analysis. Dont go back to baseline but know when stims where and signal that occurred so analyse to get response to each stim
What are 3 additional practical considerations for fNIRS experiments?
• Pressure from cap can cause pain increase over time
• Fatigue or other state changes of participant
• Large effect on non-neural signal contributions
• Behavioural response artifacts
• Button presses
• Verbal responses
What is the montage design?
Limited # of optodes so cant measure whole brain, have to target specific region of brain and have to know best spots to put optodes in order to get best signals
To decide where to put optodes:
Tools take map of head, initial distribution of sources and detectors and map out how light will propagate between them
Use atlas
Get Best source/detector pairing to target specific brain region
What is High density diffuse optical tomography?
a lot more sources and detectors (the future)
16 sources and detectors but more channels (pairings)
High density can have thousands of channels, much more overlap and knowledge of diff regions so have much more
info about brain activity, better res, whole head and multi distace,not just link between each optode but also across
them
What are 4 unique considerations to make when analyzing fNIRS data?
• The noise in fNIRS data is correlated
• Duetostrongcardiac, respiratory, and blood pressure changes
• Data is not independent across measurement channels
• Duetolowspatial frequency of superficial physiological signals
• Data is heteroscedastic
• Homoscedasticity = all the samples arise from the same noise distribution
• Due to properties of motion artifacts (> than physiological noise) constitutes a heavy-tail to the noise distribution
• Noise is non-stationary
• Systemic physiology appears to change of the length of the scan, reflects slow physiological changes
What are 2 approaches to fNIRS data analysis?
Waveform averaging
GLM
What is waveform averaging?
- Clean your data / metadata
- Conversion to oxy-deoxy haemolgobin
- Signal processing to enhance component of interest (filtering, artifact correction, etc) filter out frequencies of interest
- Epoch data about the stimulus onset
- Reject bad data segments
- Combine epochs
Present stim many times, response should be same each time but noise is random so when average, signal is strong but noise is reduced
What is GLM?
- Clean your data / metadata
- Conversion to oxy-deoxy haemolgobin
- Build a model of expected response
HRF shape, know when stims occurred and for how long and how brain should respond so make predicted model of how brain should respond - Add to model other factors you wish to model e.g., drifts in signal, systemic contributions
- Fit model to data
How do we model the expected response?
GLM: know timing and how brain should rép so build exected rép
Theres a lot of noise in real data but generally theres a bigger rép when the stim is presented
Theres a lot of noise so have to do a lot of trials and if signal is so small might have to average across diff ppl
How do we do signal processing?
When someone moves, the optode could disconnect and reconnect -> spike artefact
or all sensors move so get baseline shift
Filters wont remove these so have diff techniqques to apply the discontinuities and correct them but its not perfect
What are Mayer waves?
BP fluctuations occur around 0.1 Hz but these signals are in the same f range as neural rép so cant filter it out like heart rate so randomize inter stim interval of conditions or randomise
stim conditions, use exp design to help get around these instead of filtering them out
What are the signal components?
We’re not just measuring evoked responses, we have systemic components (heart rate signal, Mayer waves, changes in extracerebral activity) and these need to be removed, can filter out a section of f but some of these are the same as neural fs so use short channels
What are short channels?
SHORT CHANNELS CAN SUPRESS UNDESIRED COMPONENTS
Use different distances between source and detectors, at short distance, light doesnt go through brain so subtract it from light that went through brain
How can we do multimodal imaging?
How to combine diff neuroimaging modalities: EEG and fNIRS
EEG is good for fast signal and fNIRS for slow, they measure diff parts of the activity so they complement each other
Hard to integrate info of the 2 signals
What is multi-depth analysis?
• Modify spacing of source detectors to probe different depths
• Challenging as the number of photons returning to detector decreases exponentially with depth
• With many sensors (24x2 at Macquarie Uni) you can probe multiple locations and depths
Use fNIRS to probe at diff depths, if move sources aand detectors farther away, can get info from deeper
Played 2 diff sounds, one elicited info more at surface and one deeper
Without multiepth fNIRS wouldnt be able to disintangle these 2 signals
