State the classical Beer-Lambert law and explain why it cannot be applied directly to biological tissue
Describe the modified Beer-Lambert law and explain the role of the Differential Pathlength Factor (DPF)
Explain why the MBLL measures changes in concentration rather than absolute concentrations
List the key assumptions underlying the MBLL and recognise where each could be violated

We now learned that light in tissue is described by a classical particle transport model due to the predominance of scattering over absorption by roughly 100:1, resulting in an effective optical path length (the statistical average distance light travels between source and detector) much longer than the straight-line source-detector distance, and unknown. But how do we get from a change in light intensity at the detector to a number in micromoles of haemoglobin? The answer involves a law with a 300-year history, a problem that resisted solution for decades, and a modification introduced to enable all of modern functional fNIRS.

1. The Beer-Lambert Law

The quantitative relationship between light attenuation and absorber concentration was established over three centuries by three people working independently.

Pierre Bouguer (1729) was the first to observe that light intensity decreases exponentially as it travels through a medium. He was studying starlight passing through the atmosphere, and his insight was that each successive layer of the medium removes a constant fraction of the remaining light, regardless of how much light remains. Johann Heinrich Lambert (1760) formalized this mathematically, establishing that attenuation is proportional to the path length of light through the medium. And, as we saw in the previous lesson, August Beer (1852) then showed that attenuation is also proportional to the concentration of the absorbing substance. Together, their contributions give us what is now commonly called the Beer-Lambert Law (but, really, should be called the Bouguer-Beer-Lambert Law).

🎥 WATCH THIS

Beer-Lambert Law Explained

The Beer-Lambert Law is taught in virtually every chemistry and physics program, and there are many excellent video explanations. The video below is an example of an accessible, well-paced version that cover absorbance, transmittance, extinction coefficient, and concentration with clear diagrams of the classic cuvette setup.

In a non-scattering medium (e.g., a clear solution in a glass cuvette, a gas of known composition) the law is exact and powerful. It states that the absorbance of the medium is proportional to the concentration of the absorber and the distance light travels through it:

💡 The Beer-Lambert Law

$$A(\lambda) = log_{10}\left( \frac{I_0}{I}\right) = \epsilon(\lambda)\cdot c\cdot L$$

where:

A(λ) is the absorbance at wavelength λ, defined as the logarithm of the ratio of incident light intensity (I₀) to transmitted intensity (I). A higher A means more light was absorbed.
ε(λ) is the molar extinction coefficient of the absorber at wavelength λ, a fixed molecular property, tabulated from laboratory measurements in pure solutions.
c is the molar concentration of the absorber, i.e., what we want to determine.
L is the optical path length, the distance light actually travels through the absorbing medium. In a clear cuvette, this equals the physical width of the container.

The elegance of this law is that if you know ε (tabulated), measure OD (from your detector), and know L (the cuvette width), you can calculate c directly. In analytical chemistry, this is routine.

📜 ABSORBANCE (A) VS OPTICAL DENSITY (OD)

In the vast majority of contexts in chemistry, biology, and clinical spectroscopy, it is very common to use the term “absorbance,” as we did above, to denote the Beer-Lambert law. In these fields, the definition above uses base-10 for the log.

In physics, and optics in particular, the term “Optical Density” is employed more often, typically referring to the natural logarithm ($\ln$), because the physical decay of the light intensity is seen as a continuous exponential process ($I=I_0e^{-\alpha x}$).

This is typically a term of confusion, because most of the fNIRS literature uses the term “Optical Density” and the extinction coefficients tabulated from experiments measuring attenuation ($log_{10}$). To convert between them, we need to use change base of the logarithm, and since
$$\ln(x) = \ln(10) \times \log_{10}(x)$$
these two terms are related by a factor of 2.303. That’s why you see this factor quite often in guidelines to compute the absorption coefficient in fNIRS toolboxes. (Note, the relation above follows directly from the change-of-base formula of logarithms.)

2. Why the Classical Beer-Lambert Law Fails in Tissue

The Beer-Lambert law was derived for non-scattering media, and its derivation assumes that light travels in a straight line from source to detector, so that the path length L is simply the physical separation between the two. Although this is true in a cuvette sitting in a bench, this is certainly not the case in tissue.

In tissue, light undergoes thousands of scattering interactions per centimeter. The effective optical path length is typically three to six times the straight-line source-detector separation, and it depends on the scattering properties of the medium ($\mu_s’$), which a CW intensity measurement cannot determine independently. Without knowing L, you cannot convert an optical density measurement into a chromophore concentration using the classical law.

If you apply the classical law directly to tissue using the source-detector separation as L, you will systematically underestimate chromophore concentrations by a factor of three to six. Worse, the error will be inconsistent across participants, tissue types, wavelengths, and ages, because $\mu_s’$ varies with all of these.

💭 Pause and Think

Interestingly, Bouguer originally derived his law studying light attenuation in the atmosphere, a medium that scatters as well as absorbs. Yet his law is still described as valid for clear (non-scattering) media. Why might scattering in the atmosphere not cause the same problem as scattering in tissue?

Think about the geometry: in atmospheric measurements, light travels in a roughly straight line over very long distances with only occasional scattering. In tissue, by contrast, scattering is so frequent and so dominant that diffusion is reached within millimeters. This difference in regime is what determines whether the classical law applies.

3. The Modified Beer-Lambert Law

Quantitative fNIRS was made possible by recognizing that while the absolute effective path length is unknown in a CW measurement, it is possible to work around this by focusing on changes rather than absolute values. This idea came from David Delpy, Mark Cope, and colleagues working at University College London in 1988.

The reasoning is as follows. The total attenuation of light in tissue has two components: attenuation due to absorption (which changes when hemoglobin concentrations change) and a baseline term due to scattering (which is stable over short time windows). If you measure the change in optical density between a baseline state and a task state (written as ΔOD), the stable scattering term cancels out, and what remains is purely the contribution of the absorption change. You still need to know the effective path length, but you can estimate it from the known scattering properties of tissue using a multiplicative correction factor: the Differential Pathlength Factor (DPF).

💡 The Modified Beer-Lambert Law (MBLL)

$$\Delta OD(\lambda) = \sum_i \epsilon_i(\lambda)\cdot \Delta c_i\cdot d\cdot DPF(\lambda) + G(\lambda)$$

where:

ΔOD(λ) is the measured change in optical density at wavelength λ between a baseline and a task state (i.e., what the detector directly records).
εᵢ(λ) is the molar extinction coefficient of chromophore i at wavelength λ, known and tabulated (Prahl, 1999).
Δcᵢ is the change in molar concentration of chromophore i, which is what we want to determine for HbO₂ and HbR.
d is the source-detector separation (assumed as a straight-line distance between optodes, which is known from the probe geometry).
DPF(λ) is the Differential Pathlength Factor, i.e., a dimensionless factor that accounts for the lengthening of the effective path due to scattering. Together, (d · DPF) gives the effective optical path length.
G(λ) is a geometry-dependent scattering term that captures the baseline optical density contribution of scattering and any other non-absorption factors. When working with changes (ΔOD), G cancels out because it is stable over short time windows, which is why the MBLL works for tracking changes even though it cannot recover absolute concentrations.

The critical difference from the classical law is clear: where the original law used the geometric path length L (known exactly in a cuvette), the MBLL uses (d · DPF), i.e., the source-detector separation multiplied by an empirical correction factor that accounts for the additional path length due to scattering. And where the original law related absolute optical densities to absolute concentrations, the MBLL relates changes in optical density to changes in concentration.

Although this looks like a small change from the original Beer-Lambert law, the MBLL is a fundamental difference in what the measurement provides. It cannot tell you the absolute concentration of HbO₂ in the tissue at any moment. It can only tell you how much the concentration has changed relative to a baseline period. This is why fNIRS time series almost always show relative changes (denoted ΔHbO₂ and ΔHbR) rather than absolute concentrations.

🎍 must-read Paper — The Modified Beer-Lambert Law

Delpy, D.T., Cope, M., van der Zee, P., Arridge, S., Wray, S. & Wyatt, J. (1988). “Estimation of optical pathlength through tissue from direct time of flight measurement.” Physics in Medicine and Biology, 33(12), 1433–1442.

The paper that introduced the DPF concept and the Modified Beer-Lambert Law to NIRS. Delpy and colleagues used time-of-flight measurements of picosecond laser pulses to measure the actual mean optical path length in tissue for the first time, establishing that it was approximately 5.3 times the physical head diameter in rats. This gave the first empirical measurement of the DPF and the basis for the MBLL modification. The framework introduced here (i.e., measuring changes in optical density and dividing by the effective path length) represents virtually every CW fNIRS analysis performed since.

→ doi:10.1088/0031-9155/33/12/008

4. The Differential Pathlength Factor (DPF)

From the Section above, we can define DPF as the ratio of the effective optical path length to the source-detector separation. If the effective path length is 20 cm and the source-detector separation is 3 cm, the DPF is approximately 6.7. In practice, DPF values in the human head typically fall between 4 and 8, depending on wavelength, tissue type, and age.

Conceptually, the DPF is not a constant; it varies with several factors:

Wavelength: scattering decreases with wavelength (as described by $\mu_s’ \propto A\lambda^{-b}$), so the DPF is slightly larger at shorter wavelengths within the NIR window
Age: in one of the most practically important findings, Duncan et al. (1996) measured DPF values in 283 subjects from 1 day to 50 years old and showed that the DPF increases substantially from neonates to adults. Neonatal brain tissue scatters differently from adult brain tissue, and using an adult DPF for infant data (or vice versa) introduces systematic concentration errors
Tissue type: the DPF differs between brain, muscle, and other tissues because their structural compositions (and therefore $\mu_s’$) differ
Source-detector separation: the DPF is approximately constant for separations greater than about 2.5 cm, but varies at shorter separations

Again, CW fNIRS systems cannot measure the DPF. They can only measure the total light intensity reaching the detector, which is a combined function of absorption and scattering but cannot separate the two. The DPF must therefore be assumed in CW systems, estimated from published empirical values measured by TD or FD systems, or from Monte Carlo simulations of light transport in tissue models. This assumed DPF is then used in every application of the MBLL in that system. Using a wrong DPF introduces a systematic scaling error in all concentration estimates.

🎍 should-read Paper

Duncan, A., Meek, J.H., Clemence, M., Elwell, C.E., Fallon, P., Tyszczuk, L., Cope, M. & Delpy, D.T. (1996). “Measurement of cranial optical path length as a function of age using phase resolved near infrared spectroscopy.” Pediatric Research, 39(5), 889–894.

A study of 283 subjects from one day to 50 years of age, measuring cranial optical path length using frequency-domain (phase-resolved) NIRS at four wavelengths. Established the empirical age-DPF relationship that is still used in most fNIRS analysis software today. This paper is the reason that infant and paediatric fNIRS studies use different DPF values from adult studies.

→ doi:10.1203/00006450-199605000-00025

📜 DPF vs PPF vs Partial volume effect

Not rarely, you might also find a similar acronym in the literature: PPF. These two are related but distinct quantities that are sometimes confused, and they also have implications to a problem of any imaging technique called partial volume effect (PVE). Understanding the difference matters for interpreting brain-specific fNIRS measurements you may read in fNIRS literature (particularly in simulation studies, where PPF can be numerically estimated and is evaluated for sensitivity).

Term	What it is	What it tells you
DPF Differential Pathlength Factor	The ratio of the total effective optical path length to the source-detector separation, averaged over the full sensitivity volume (scalp + skull + CSF + cortex)	How far light travelled in total through all tissue layers. Used in the standard MBLL.
PPF Partial Pathlength Factor	The fraction of the total effective path length that passed through a specific tissue layer (for example, cortical grey matter alone.) It is always the case that PPF ≤ DPF.	How much of the light’s journey was through the brain specifically. Relevant for quantitative brain-specific measurements.
PVE Partial Volume Effect	Even within the cortex, the activated region typically occupies only a fraction of the total tissue volume sampled by the source-detector pair. The MBLL assumes concentration changes are uniform across the entire sensitivity volume.	A focal activation that fills, say, 20% of the sampled volume will produce a signal 80% smaller than it would if the whole volume were activated. The true local concentration change is underestimated unless the activated fraction is known.

These three effects compound one another and together explain why absolute quantification in fNIRS is so challenging. The DPF tells you that light travelled much further than the source-detector distance (a path length problem). The PPF tells you that only a fraction of that journey was through brain tissue (a layered tissue problem). The partial volume effect tells you that even within the brain, only a fraction of the sampled cortical volume may have been activated (a spatial localisation problem).

Each effect independently dilutes the measured signal relative to the true local concentration change. Using the DPF when you should use the PPF means your Δc estimates are already diluted by extracerebral contributions; failing to account for partial volume means they are further diluted by inactive cortex within the sampling volume. These are not merely theoretical concerns, they are the reason that fNIRS concentration changes are systematically smaller than the true local hemodynamic response, and why comparing absolute magnitudes between participants or studies requires great caution.

5. The Assumptions Behind the MBLL

The Modified Beer-Lambert Law is powerful, practical, and widely used. It is also an approximation and, as we will see, it is a specific solution of the photon diffusion equation under a particular set of simplifying conditions. Knowing these conditions is essential for judging when the MBLL can be trusted and when its outputs should be treated with additional caution.

⚠ Key Assumptions of the Modified Beer-Lambert Law

Assumption	What it means (and when it can break down)
Homogeneous semi-infinite medium	The tissue is assumed to be a uniform half-space with constant optical properties. In reality, the head is layered (scalp, skull, CSF, grey matter, white matter), and each layer has different $\mu_a$ and $\mu_s’$. This layered structure is the main reason why the PPF < DPF.
Scattering is stable and known	$\mu_s’$ is assumed constant over the measurement window. Valid for short functional paradigms in healthy tissue, but can break down over long sessions, in pathological tissue (edema, ischemia), or in tissue undergoing active (structural) changes.
Changes are small (linearisation)	The MBLL is a linearised approximation. It is accurate for small concentration changes, but becomes increasingly inaccurate for large hemodynamic changes where the non-linear terms of the full diffusion solution become significant.
Uniform sampling (single DPF)	A single DPF is used for all detected photons, implying they all sampled the tissue volume equivalently. In practice, photons sample different depths and layers. Using a single DPF averages over this heterogeneity, which is the same assumption that makes DPF different from PPF.
DPF is known accurately	In CW fNIRS, the DPF is assumed rather than measured. Using an incorrect DPF introduces systematic errors in the estimated concentration changes, and wavelength-dependent DPF errors produce crosstalk between HbO₂ and HbR estimates.

None of these assumptions renders the MBLL useless (far from it!). For detecting relative hemodynamic changes during functional paradigms in healthy adults, the MBLL performs well enough to support a very large and productive research literature. The important thing is to know what the assumptions are, which ones are likely to hold in your specific situation, and what the consequences are when they break down (particularly when you are comparing groups where these assumptions may vary).

Other models, including the full diffusion equation solution, exist and can be used in more specialised contexts. The MBLL is not the only tool; it is the simplest one that works well enough for most purposes.

💭 Pause and Think

A researcher is using CW fNIRS to study cerebral oxygenation in a group of children aged 4-8 years. They apply a fixed DPF of 6.0, taken from a table of adult values. What systematic error does this introduce, and in which direction?

From what you know about how DPF varies with age, would you expect the true DPF for this age group to be higher or lower than 6.0? And if you overestimate the DPF, do you over- or underestimate Δcᵢ? Trace through the MBLL formula to confirm your intuition.

📌 Key Takeaways

The Beer-Lambert Law (A = ε · c · L) relates optical density to absorber concentration and path length in clear, non-scattering media. It was developed across the 18th and 19th centuries by Bouguer (1729), Lambert (1760), and Beer (1852).
In tissue, the law fails because the effective optical path length L is unknown: it depends on scattering ($\mu_s’$), which a CW intensity measurement cannot determine independently. Applying the classical law directly to tissue produces systematic, uncorrectable errors.
The Modified Beer-Lambert Law (MBLL), introduced by Delpy et al. (1988), addresses this by working with changes in optical density (ΔOD) rather than absolute values, and replacing the unknown path length with (d · DPF), where the DPF accounts for scattering-induced path lengthening.
Because the MBLL uses changes, it recovers Δcᵢ (relative concentration changes), not absolute concentrations. This is why fNIRS data are always expressed as ΔHbO₂ and ΔHbR relative to a baseline.
The DPF is the total effective path length divided by the source-detector separation. It varies with wavelength, age, and tissue type. In CW fNIRS it must be assumed from empirical tables (e.g., Duncan et al., 1996). The PPF is the fraction of the path through a specific tissue layer (e.g., cortex); PPF ≤ DPF and is relevant for quantitative brain-specific measurements.
The MBLL rests on five key assumptions: homogeneous medium, stable scattering, small changes (linearisation), uniform sampling, and a known DPF. It works well for standard functional paradigms in healthy adults, but breaks down in layered tissue, pathological conditions, large hemodynamic changes, or when the wrong DPF is used.

📚 Further Reading & Landmark References

Delpy, D.T., Cope, M., van der Zee, P., Arridge, S., Wray, S. & Wyatt, J. (1988). “Estimation of optical pathlength through tissue from direct time of flight measurement.” Physics in Medicine and Biology, 33(12), 1433–1442. [doi] — The foundational paper introducing the DPF and the Modified Beer-Lambert Law.
Cope, M. & Delpy, D.T. (1988). “System for long-term measurement of cerebral blood and tissue oxygenation on newborn infants by near infrared transillumination.” Medical & Biological Engineering & Computing, 26(3), 289–294. [doi] — The companion paper applying the MBLL framework to newborn infants; the first practical use of the approach.
Duncan, A., Meek, J.H., Clemence, M., Elwell, C.E., Fallon, P., Tyszczuk, L., Cope, M. & Delpy, D.T. (1996). “Measurement of cranial optical path length as a function of age using phase resolved near infrared spectroscopy.” Pediatric Research, 39(5), 889–894. [doi] — The landmark study establishing age-dependent DPF values.
Hiraoka, M., Firbank, M., Essenpreis, M., Cope, M., Arridge, S.R., van der Zee, P. & Delpy, D.T. (1993). “A Monte Carlo investigation of optical pathlength in inhomogeneous tissue and its application to near-infrared spectroscopy.” Physics in Medicine and Biology, 38(12), 1859–1876. [doi] — Introduces the PPF concept and examines path length in layered tissue using Monte Carlo simulation; the formal basis for the DPF/PPF distinction.
Sassaroli, A. & Fantini, S. (2004). “Comment on the modified Beer-Lambert law for scattering media.” Physics in Medicine and Biology, 49(14), N255–N257. [doi] — An important clarification of a subtle error in the standard MBLL formulation, proposing a corrected notation.
Scholkmann, F. & Wolf, M. (2013). “General equation for the differential pathlength factor of the frontal human head depending on wavelength and age.” Journal of Biomedical Optics, 18(10), 105004. [doi] — Provides a practical general formula for DPF as a function of both wavelength and age, combining the Duncan et al. data with an analytical framework.

Society for Functional Near-Infrared Spectroscopy.
Content is for educational purposes.