When a region of cortex becomes active, local blood flow increases to deliver oxygen, and the balance between oxygenated haemoglobin (HbO) and deoxygenated haemoglobin (HbR) shifts. The two absorb near-infrared light differently, roughly in the 650–950 nm range where skin and bone are fairly transparent. An fNIRS system places light sources and detectors on the scalp, measures attenuation at two or more wavelengths, and converts the change into HbO and HbR concentration changes using the modified Beer-Lambert law.
Because it measures blood rather than electricity, fNIRS sees a slow haemodynamic response that peaks several seconds after the neural activity that caused it. That is its main limitation and also the reason it tolerates movement, talking and real-world settings better than EEG or fMRI.
fNIRS compared with EEG and fMRI
| fNIRS | EEG | fMRI | |
|---|---|---|---|
| Measures | Blood oxygenation (HbO, HbR) | Electrical activity | Blood oxygenation (BOLD) |
| Timing | Seconds (haemodynamic lag) | Milliseconds | Seconds |
| Spatial | About a centimetre, cortex only | Centimetres, blurred | Millimetres, whole brain |
| Movement tolerance | Good | Poor to moderate | Very poor |
| Portability | Wearable systems exist | Wearable | Scanner only |
| Typical sampling | Around 10 Hz | 250–1000+ Hz | About 1 volume per second |
HbO and HbR are two signals, not one
During activation, HbO typically rises and HbR falls slightly. They have different time courses, different sensitivity to systemic physiology, and opposite signs, so analysing them together as if they were two channels of one signal mixes two different measurements.
This is the most common mistake in software built for EEG and adapted to fNIRS. BCILattice puts a chromophore control (All, HbO or HbR) on every fNIRS-capable step for that reason. See fNIRS analysis.
What contaminates an fNIRS signal
Scalp blood flow
Light passes through the scalp twice, so superficial blood flow changes look like brain activity. Short-separation channels measure it so it can be regressed out.
Systemic physiology
Heartbeat (around 1 Hz), respiration (around 0.3 Hz) and Mayer waves (around 0.1 Hz) all modulate blood volume.
Motion
Optode slippage produces spikes and baseline shifts that need motion correction.
Hair and skin
Dark, dense hair and skin pigmentation reduce light coupling and signal quality.
fNIRS for brain-computer interfaces
The haemodynamic lag makes fNIRS a slow BCI. A decision takes seconds rather than fractions of a second. It suits tasks where speed matters less than robustness, such as mental arithmetic versus rest, workload monitoring, and hybrid systems that pair fNIRS with EEG. See multimodal BCI.
Cohorts are usually small and between-subject variability in optode placement is large, so evaluation design matters even more than for EEG. fNIRS machine learning covers what changes.
Questions
What does fNIRS stand for?
Functional near-infrared spectroscopy. It uses near-infrared light to measure changes in oxygenated and deoxygenated haemoglobin in the outer layer of the brain.
Is fNIRS better than EEG?
Neither is better overall. EEG has far better timing; fNIRS tolerates movement better and gives a more localised cortical signal. Many studies record both.
How deep can fNIRS measure?
Roughly the outer one to two centimetres of the cortex, depending on source-detector separation. It cannot reach deep brain structures.
What file format does fNIRS data use?
SNIRF is the community standard format for fNIRS data. Many systems also export proprietary or tabular formats.
What is the modified Beer-Lambert law?
It relates the change in light attenuation measured at each wavelength to changes in HbO and HbR concentration, with a correction for the longer, scattered path light takes through tissue.
Read the reference
Related
fNIRS Analysis Software
An fNIRS file holds a HbO and a HbR channel per source-detector pair. Every fNIRS-capable step here exposes a chromophore control, because decomposing or ranking both together mixes two different physiological signals.
fNIRS Machine Learning
The model canvas is the same one EEG uses. What changes for fNIRS is which features carry information at 10 Hz, how the chromophore enters the design matrix, and how little a small cohort forgives a bad split.
How EEG Works: What an EEG Measures
Electroencephalography (EEG) records the brain's electrical activity with electrodes on the scalp. It has millisecond timing and centimetre-scale blur, and knowing both is the key to using it well.
Multimodal and Hybrid BCI: EEG + fNIRS + EMG
A hybrid BCI combines two signals, such as fast EEG with slower but sturdier fNIRS, or adds stimuli as data. The hard part is not combining them. It is showing the combination actually helped.