Physiological Interpretation of Hemoglobin Changes

In the previous lesson, we discussed the physiological chain that makes fNIRS work for brain function studies: neural activity triggers a CBF increase that outpaces the CMRO₂ increase, OEF falls, and the local haemoglobin pool shifts toward HbO₂ and away from HbR. In that discussion, we treated the vascular bed beneath a channel as a single, well-mixed compartment, but in fact, a real vascular bed contains arterioles, capillaries, and venules, each with a different oxygenation state, and a single fNIRS channel sums over all of them. Here, we will ask what follows from that fact, and what it means for how HbO and HbR should be read.

1. The Arterial and Venous Compartments

Arterial blood carries oxygen into tissue; venous blood carries away what remains after the tissue has extracted what it needs. Arterial blood is close to fully saturated with oxygen, and is therefore mostly HbO₂. Venous blood has already given up a fraction of its oxygen (quantified through the OEF from the previous lesson), and therefore carries relatively more HbR. A natural hypothesis that follows directly from this is that, within the mixed vascular bed a channel samples, ΔHbO should be weighted toward the arterial and inflow side of the vascular tree, and ΔHbR toward the venous and outflow side. This logic is sound, but logic alone does not establish how strongly this plays out in real tissue, where all three compartments sit within the same sampled volume and their relative blood volumes and dynamics interact.

Support for this compartmental argument comes from methods that can resolve individual vessel types directly, rather than inferring their behaviour indirectly.

A related line of evidence comes from two-photon measurements of tissue oxygen tension itself. For instance, Devor and colleagues (2011) imaged oxygen partial pressure in rat cortex at high spatial resolution during sensory stimulation, and found that the oxygenation increase was largest close to blood vessels and fell off with distance from them, consistent with a vessel-anchored source of the signal rather than a spatially uniform tissue-wide change. In their work, they also provide a concrete mechanistic answer to the CBF over-delivery question from the previous lesson: the authors argue the “overshoot” of oxygen delivery exists specifically to keep tissue distal from vessels adequately oxygenated, since those locations would otherwise dip below baseline during the transient phase of the response.

2. Differences Between HbO and HbR

Regardless of their specific physiological origins, ΔHbO and ΔHbR may provide complementary information that should be analysed in conjunction, and never separately. In several studies, however, ΔHbR is completely or at least largely ignored, and only ΔHbO changes are considered. Unfortunately, this approach can lead to misinterpretation and undermines the impact of any fNIRS-based research.

The reason that ΔHbR is frequently ignored is because, in practice, ΔHbO has a better signal-to-noise ratio (SNR) than ΔHbR. At least two mechanisms contribute to this. The first is a matter of physiological amplitude. During activation, the dominant driver of ΔHbO is a substantial volume of highly oxygenated arterial blood moves into the sampled region as CBF and CBV increase, producing a large, directly-driven concentration change. ΔHbR’s net change is the result of two effects working against each other: the same inflow of oxygenated blood dilutes the existing pool of deoxygenated haemoglobin, while ongoing oxygen extraction in the capillary bed continues to generate new HbR. This difference between two partially opposing effects tends to be smaller and noisier than ΔHbO for a comparable stimulus.

The second mechanism is instrumental, and follows directly from the absorption spectra introduced in Unit 1. Recall that wavelengths below the isosbestic point (roughly 805 nm) are weighted more toward HbR, while wavelengths above it are weighted more toward HbO₂. Wavelengths in the lower part of the fNIRS range are also more strongly absorbed by tissue overall. This means fewer photons reach the detector at the wavelength doing most of the work of estimating HbR, which increases noise on exactly the channel already carrying the smaller physiological signal. The two mechanisms compound rather than offset each other.

3. The “Expected” Hemodynamic Response

Putting the previous lesson and this one together gives the pattern most fNIRS works recognise as the expected result of a well-behaved, task-related activation: HbO rises, HbR falls, with HbO showing the larger and cleaner signal of the two. This is what is meant by the canonical response, and it follows from the CBF/CMRO₂ mismatch, the compartmental weighting of the two signals, and the SNR asymmetry.

In practice, this pattern is often used as a rough data-quality check: a channel showing a clear HbO increase alongside an HbR decrease is behaving as expected. But “as expected” is doing real work in that sentence, and the rest of this lesson looks at what happens when the physiology itself departs from expectation.

Even in healthy tissue with fully intact neurovascular coupling, this “expected” response can have some variations, and two well-documented variations are worth knowing before you encounter them in real data.

The initial dip is a brief, small increase in HbR (or decrease in HbO) reported in the first one to two seconds after stimulus onset, before the canonical response takes over. It was first described in optical imaging by Malonek and Grinvald (1996), who proposed it reflects a short window in which local oxygen extraction outpaces the still-developing blood flow response playing out transiently before the flow increase catches up. The finding proved difficult to replicate consistently, however: it has been reported in some studies, across fMRI, fNIRS, and optical imaging, and species, but not others, and its existence remains genuinely contested rather than settled. Hu and Yacoub (2012) give a detailed account of this history for fMRI specifically, and the same inconsistency applies to fNIRS.

The negative response, a task-related decrease in HbO alongside an increase in HbR (i.e., the reverse of the canonical pattern), is also documented, and is not automatically evidence of an artifact or of poor data quality. For instance, Devor et al. (2007) showed in rat cortex that a negative BOLD-type response can correspond to genuinely suppressed local neural activity accompanied by arteriolar vasoconstriction, a real physiological event rather than noise. A reversed pattern in your own data could be therefore something to investigate, rather than something to discard by default.

4. The Fast Optical Signal

Everything in this lesson concerns a hemodynamic signal: a change in blood oxygenation that follows neural activity by roughly one to two seconds. A separate line of research has asked whether near-infrared light can also detect something closer to neural activity itself, on a millisecond timescale. Active neurons change shape slightly as they depolarise, and this changes how strongly nearby tissue scatters light. Gratton et al. proposed that this scattering change, the event-related optical signal (EROS), could be measured non-invasively, reporting spatial agreement with fMRI and temporal agreement with visual evoked potentials (Gratton et al., 1997). Franceschini and Boas (2004) independently reported a related fast signal during finger tapping, tactile, and median nerve stimulation, though explicit that the method still “required further optimization and validation,” and that hundreds of trials had to be averaged to see it.

Other groups working independently reached a more skeptical conclusion. For instace, Steinbrink et al. (2005), building on a failed replication attempt by another group, calculated a physical upper limit for how large a genuine scattering signal could plausibly be when measured non-invasively through the intact adult skull, and found it should be orders of magnitude smaller than what had been reported. Several years later, Radhakrishnan et al. (2009), testing under carefully optimised conditions in awake macaque monkeys with thousands of averaged trials, detected the hemodynamic response reliably on every block but found no measurable fast signal, despite this being, notably, the same group that had reported a positive result in humans five years earlier.

This is not a settled story, and it is not included here as one. But the EROS represents a different physical signal altogether, based on scattering rather than absorption, aimed at a different target, neuronal structure rather than haemoglobin concentration, and because the field’s own difficulty in reliably reproducing it across independent groups and species is a useful, concrete illustration of how demanding non-invasive optical measurement of the brain really is. The rest of this course, including everything from here on, concerns the hemodynamic signal only.

📚 Further Reading & Key References

1. Hillman, E.M.C., Devor, A., Bouchard, M.B., et al. (2007). “Depth-resolved optical imaging and microscopy of vascular compartment dynamics during somatosensory stimulation.” NeuroImage, 35(1), 89–104. [doi] — Vessel-resolved evidence for the arterial/venous compartment argument in Section 2.
2. Devor, A., Sakadžić, S., Saisan, P.A., et al. (2011). “‘Overshoot’ of O₂ is required to maintain baseline tissue oxygenation at locations distal to blood vessels.” Journal of Neuroscience, 31(38), 13676–13681. — Two-photon oxygen imaging supporting the compartmental argument, and a mechanistic account of the CBF over-delivery question from Lesson 1.
3. Malonek, D. & Grinvald, A. (1996). “Interactions between electrical activity and cortical microcirculation revealed by imaging spectroscopy: implications for functional brain mapping.” Science, 272(5261), 551–554. — Origin of the initial-dip observation.
4. Hu, X. & Yacoub, E. (2012). “The story of the initial dip in fMRI.” NeuroImage, 62(2), 1103–1108. — An honest account of how contested the initial dip remains.
5. Devor, A., Tian, P., Nishimura, N., et al. (2007). “Suppressed neuronal activity and concurrent arteriolar vasoconstriction may explain negative BOLD.” Journal of Neuroscience, 27(16), 4452–4459. — Evidence that a negative response can reflect real, suppressed neural activity.
6. Girouard, H. & Iadecola, C. (2006). “Neurovascular coupling in the normal brain and in hypertension, stroke, and Alzheimer disease.” Journal of Applied Physiology, 100(1), 328–335. — How disease alters the coupling this whole lesson assumes is intact.
7. Huppert, T.J., Hoge, R.D., Diamond, S.G., Franceschini, M.A. & Boas, D.A. (2006). “A temporal comparison of BOLD, ASL, and NIRS hemodynamic responses to motor stimuli in adult humans.” NeuroImage, 29(2), 368–382. — Named here as the test this lesson’s hypothesis motivates; full treatment in Lesson 3.4.
8. Gratton, G., Fabiani, M., Corballis, P.M., et al. (1997). “Fast and localized event-related optical signals (EROS) in the human occipital cortex: comparisons with the visual evoked potential and fMRI.” NeuroImage, 6(3), 168–180. — The origin of the EROS claim, referenced in Section 7.
9. Franceschini, M.A. & Boas, D.A. (2004). “Noninvasive measurement of neuronal activity with near-infrared optical imaging.” NeuroImage, 21(1), 372–386. — An independent positive report, referenced in Section 7, with its own caveats about optimisation and validation.
10. Steinbrink, J., Kempf, F.C.D., Villringer, A. & Obrig, H. (2005). “The fast optical signal — robust or elusive when non-invasively measured in the human adult?” NeuroImage, 26(4), 996–1008. — The skeptical counterpoint referenced in Section 7.
11. Radhakrishnan, H., Vanduffel, W., Deng, H.P., Ekstrom, L., Boas, D.A. & Franceschini, M.A. (2009). “Fast optical signal not detected in awake behaving monkeys.” NeuroImage, 45(2), 410–419. — The negative result referenced in Section 7, notably from the same group as the 2004 positive report.