Near-infrared spectroscopy uses light in the ~700-2500 nm window, where tissue and organic matter are relatively transparent, to read chemical composition and haemodynamics without contact or contrast, and it is already a working clinical and agricultural instrument rather than a laboratory prospect.
Summary
Near-infrared light penetrates several centimetres into biological material because water, haemoglobin and lipids absorb it comparatively weakly there. Two consequences follow, and they drive almost every application in the sources. First, molecular overtone and combination bands in the NIR give a chemical fingerprint, so reflected or transmitted spectra can be calibrated against protein, starch, fat, ash or lipid content. Second, oxygenated and deoxygenated haemoglobin have different NIR absorption spectra, so light shone through the scalp and returned to a nearby detector reports cortical blood oxygenation, which is the basis of functional NIRS (fNIRS).
The compositional branch appears here in two forms. In coronary intervention, NIRS-derived lipid core burden index (LCBI) is described as well validated and widely used for detecting lipid-rich, rupture-prone plaque, and is used as the reference against which a new depolarisation-sensitive optical frequency domain imaging metric is benchmarked in 29 co-registered pullbacks from 33 patients. In plant breeding, NIR is the routine assay for macronutrient traits: protein, starch, fat and ash were measured by NIR on mature grain across a USDA lima bean germplasm panel of 810 genotyped accessions with 141-308 accessions field-evaluated across three sites.
The functional branch is where the technology is least settled. Portable 36-channel fNIRS has been used to detect pharmacologically verified dental nerve block, with temporal haemodynamic response features outperforming raw GLM amplitude and leave-one-subject-out classification reaching AUC 0.68-0.76. Multichannel fNIRS separates anxiety-spectrum patient groups from controls on task-evoked HbO and HbT responses during verbal fluency, indexes prefrontal haemodynamics in impulsivity research, and can be run concurrently with single-pulse TMS across the whole head. A third, unrelated use is NIR as an energy delivery channel: remote, non-invasive NIR laser stimulation actuates an implanted antibacterial surface in a mouse model.
The parameters that decide any given NIR application are penetration depth versus the depth of the target, the strength and specificity of the chemical contrast at the wavelengths used, and whether a stable calibration transfers across samples or subjects. The last of these is the binding constraint for fNIRS: between-subject variability of the haemodynamic response is explicitly identified as the reason classifiers fail to generalise, and within-session cross-validation is identified as a source of inflated accuracy.
Evidence base
- 2026-05-28: Portable 36-channel fNIRS during verified dental nerve block gave leave-one-subject-out AUC 0.68-0.76 (best random forest 0.760, permutation p = 0.002) with per-patient calibration, in 13 block-success patients of 25 enrolled.
- 2026-05-28: Engineered temporal HRF features (late slope, mean first derivative, baseline-normalised amplitude) showed analgesia interaction effects d = 0.63-0.79 versus d = 0.56 for raw GLM amplitude, with a difference-in-differences interaction at p = 0.011.
- 2026-06-01: NIRS-derived LCBI is described as well validated and widely used for detecting lipid-rich coronary lesions; a depolarisation-sensitive OFDI metric showed high correlation with NIRS maxLCBI4mm across 29 co-registered pullbacks from 33 patients.
- 2026-06-06: NIR spectroscopy was used to measure protein, starch, fat and ash on mature grain in a USDA lima bean germplasm study covering 810 genotyped accessions and 141-308 field-evaluated accessions across three sites.
- 2026-06-12: Multichannel fNIRS during verbal fluency showed reduced task-evoked HbO and HbT, preserved HbR, increased temporal haemodynamic variability and reduced beta activation in anxiety-spectrum patients versus controls, concentrated in frontopolar and medial prefrontal cortex.
- 2026-07-24: Whole-head fNIRS was run concurrently with single-pulse TMS to the left DLPFC during a working memory task, showing load-dependent haemodynamic increases and TMS-driven modulation across DLPFC, superior medial gyrus, precuneus and parietal lobule.
- 2026-07-01: Remote, non-invasive near-infrared laser stimulation actuated an implanted bactericidal wrinkled MXene surface, which suppressed bacterial burden in a mouse subcutaneous implant infection model.
Open questions
- Does fNIRS cross-subject classification performance improve materially beyond AUC ~0.76 in larger, multi-site, prospectively validated cohorts, or is that near the ceiling imposed by between-subject haemodynamic variability?
- Why did only 13 of 25 patients achieve an analysable pharmacological block, and is the attrition a property of the anaesthesia protocol or of the fNIRS measurement?
- Does the high agreement between depolarisation-sensitive OFDI and NIRS LCBI hold in larger cohorts and at the lesion-level decision thresholds used clinically, which would make a separate NIRS catheter redundant?
- Do the group-level fNIRS differences reported across anxiety-spectrum and impulsivity cohorts survive out-of-sample subject-level testing, or do they follow the same inflation pattern the analgesia study warns about?
Assessment drafted 2026-08-31 from up to 14 KB sources using the technology-scorecard framework; scores are a draft read pending review.