Optical Neuroimaging: Diffuse Optics, fNIRS & Bedside Hemodynamics

01. The Near-Infrared Diagnostic Window

Biological tissue is largely opaque to ultraviolet and visible light due to strong electronic absorption by hemoglobin, melanin, and proteins ($\lambda < 650\,\text{nm}$). Above $\lambda > 950\,\text{nm}$, rotational-vibrational modes of water ($H_2O$) and lipids dominate attenuation.

Between these absorption regimes lies the optical diagnostic window ($\lambda \approx 650 - 900\,\text{nm}$), where absorption coefficients drop by orders of magnitude ($\mu_a \sim 0.01 - 0.1\,\text{mm}^{-1}$), permitting photons to penetrate several centimeters into biological tissue before being absorbed.

02. Photon Migration in Turbid Media: Diffusion Approximation

Biological tissue is not merely absorbing; it is an intensely turbid, multiple-scattering medium. The reduced scattering coefficient $\mu_s' = \mu_s(1 - g)$ ranges from $0.8 - 1.5\,\text{mm}^{-1}$, where the scattering anisotropy parameter $g \approx 0.9$ reflects highly forward-peaked Mie scattering by cell membranes and mitochondria.

Because $\mu_s' \gg \mu_a$, photon trajectories undergo rapid spatial randomization within a transport mean free path $l_{\text{tr}} = 1/\mu_s' \approx 1\,\text{mm}$. Beyond this distance, directional photon radiance collapses into an isotropic photon fluence rate $\Phi(\vec{r}, t)$ governed by the Diffusion Equation:

where $v = c/n$ is the speed of light in tissue ($n \approx 1.37 - 1.40$), $S(\vec{r}, t)$ is the isotropic source term, and the optical diffusion coefficient is: $$D = \frac{1}{3(\mu_a + \mu_s')} \approx \frac{1}{3\mu_s'}$$

The Banana-Shaped Sensitivity Kernel

For a point source and detector separated by distance $d_{\text{SD}}$ on a semi-infinite planar boundary, the photon adjoint probability distribution follows an arcing, banana-shaped profile. The mean physical penetration depth $\bar{z}_{\text{max}}$ scales roughly as:

For an adult with $d_{\text{SD}} = 30\,\text{mm}$, photons sample approximately $10 - 15\,\text{mm}$ deep—sufficient to reach the superficial gyral crests of the cerebral cortex, though heavily contaminated by extracerebral scalp and skull blood flow.

03. The Modified Beer-Lambert Law (MBLL)

Classical Beer-Lambert transmission $A = \varepsilon \cdot C \cdot d$ assumes a collimated beam in a pure, non-scattering absorber. In scattering tissue, multiple scattering lengthens the physical geometric distance $d_{\text{SD}}$ into an effective optical pathlength: $$L_{\text{opt}} = d_{\text{SD}} \cdot \text{DPF}$$ where $\text{DPF}$ is the dimensionless Differential Pathlength Factor ($\approx 5.5 - 6.5$ in adult human head).

Taking temporal derivatives to cancel out static tissue scattering and constant background absorption, the Modified Beer-Lambert Law computes optical density changes ($\Delta \text{OD}$) as a linear combination of chromophore concentration changes:

Dual-Wavelength Spectroscopic Inversion Matrix

Evaluating at two distinct wavelengths ($\lambda_1 = 760\,\text{nm}$ and $\lambda_2 = 850\,\text{nm}$) yields a solvable $2 \times 2$ linear system:

04. Optical System Topologies: CW, FD & TD

Clinical and research fNIRS systems fall into three primary engineering paradigms with varying hardware complexity, cost, and quantitative information content:

05. Grand Challenge Focus: Neonatal Bedside Brain Monitoring

06. Signal Processing: Motion Artifacts & Superficial Filtering

fNIRS signals are contaminated by three dominant noise sources: physiological oscillations (cardiac $\sim 1 - 2\,\text{Hz}$ neonatal, respiration $\sim 0.3 - 0.7\,\text{Hz}$, Mayer waves $\sim 0.1\,\text{Hz}$), sudden optode movement (baseline shifts and high-frequency spike spikes), and systemic extracerebral scalp hemodynamics.