fNIRS
How it works
1) Source optodes emit near-infrared light (usually at least two wavelengths in the ~700–900 nm range) onto the scalp. 2) Photons scatter through tissue along a banana-shaped path and some return to detector optodes placed about 30 mm away; penetration depth is roughly half that separation, reaching the cortex. 3) From the changes in detected light intensity, using the modified Beer-Lambert law, changes in HbO and HbR concentrations are derived (distinguishable by their different absorption spectra; isosbestic point ~810 nm). 4) The hemodynamic response (HbO up, HbR down) is a marker of neuronal activity. 5) Signal features are filtered, cleaned of physiological confounds (e.g. via ~8 mm short-separation channels) and classified, e.g. with a machine-learning model, for BCI or cognitive-state assessment.
Problem solved
Enables non-invasive, portable and relatively inexpensive readout of cortical activity in naturalistic settings (including during movement), bridging the gap between the bulky, expensive fMRI and EEG, which has poor spatial resolution and a different (electrical) signal nature.
Components
LEDs/lasers emitting near-infrared light and photodetectors capturing light returning from tissue, placed on the scalp (a source-detector pair forms a measurement channel).
Official
At least two wavelengths (one below, one above the isosbestic point ~810 nm) to separate the HbO and HbR contributions.
Conversion of light intensity into HbO/HbR concentrations via the modified Beer-Lambert law, filtering, removal of motion and physiological artifacts, and feature extraction.
Official
Implementation
Systemic circulation changes (scalp, blood pressure, heart rate) mix with the cortical signal and can mimic activation.
fNIRS reaches only the superficial cortex (most sensitive just under the skull) and cannot measure deep structures.
Optode motion and hair (dense/dark/curly) degrade contact and introduce artifacts.
The slow hemodynamic response (peak ~5–6 s) limits the speed of intention decoding.
Evolution
Frans Jöbsis publishes in Science that near-infrared light penetrates tissue and enables non-invasive in vivo oxygenation monitoring.
Several research groups independently report hemoglobin-concentration changes during cognitive tasks, establishing fNIRS as a functional neuroimaging modality.
Introduction of the term "optical topography" and one of the first practical multi-channel commercial systems.
fNIRS begins to be used as a signal modality for brain-computer interfaces, including device control.
Hyperparameters (configurable axes)
Number of source-detector pairs (channels). More channels = better coverage and spatial resolution.
At least two wavelengths in the ~700–900 nm range to separate HbO and HbR.
Distance between source and detector optodes; governs penetration depth (~half the separation).
fNIRS instrument mode.
Computational complexity
Time complexity: O(kanały × próbki) na okno. Space complexity: O(kanały × bufor).
Compute bottleneck
The main limitation is not compute but the slow dynamics of the hemodynamic response (several-second lag), shallow penetration (cortex only), and contamination from scalp blood flow and motion artifacts.
Execution paradigm
Continuous, parallel sampling of all source-detector channels.
Continuous optical signal acquisition, no routing.
Parallelism
Channels and features can be processed independently/in parallel.
Hardware requirements
fNIRS acquisition and analysis are lightweight and run on CPU/embedded devices; the key is dedicated optical hardware (optodes, NIR sources, detectors).