From quantum spins and diffuse photon transport to electrodynamic volume conduction and microbubble acoustics: the first-principles mathematical and physical laws governing non-invasive neural signal detection.
Every neural recording or neuromodulation technology must confront a profound physical reality: the fundamental unit of computation in the mammalian nervous system occurs at the sub-micrometer, microvolt, and millisecond scale. An action potential propagates along an axon of diameter $0.2 - 2\,\mu\text{m}$, creating a transmembrane potential swing of $\approx 100\,\text{mV}$ across a $5\,\text{nm}$ lipid bilayer, driven by elementary ion channel currents on the order of $10^{-12}\,\text{A}$ ($1\,\text{pA}$).
Yet, to observe or manipulate this activity non-invasively, our sensors must sit outside the cranial protective capsule. We are separated from the cortex by the arachnoid mater, pia mater, cerebrospinal fluid (CSF), porous cranial bone, galea aponeurotica, and scalp skin—a barrier ranging from $8\,\text{mm}$ to over $20\,\text{mm}$ in thickness.
Non-invasive brain sensing is inherently an ill-posed physical inverse problem. We do not measure single spikes directly; instead, we measure macroscopic physical perturbations—magnetic field disturbances, phase shifts in precessing nuclear spins, photons traversing multiple scattering paths, or attenuated ionic potentials conductively smeared through resistive cranial bone. Engineering a high-performance neurodiagnostic instrument requires deriving the exact forward physical operator linking microscopic cellular pathology to external detector signals.
Perturbing the precession of water proton nuclear spins (\(^1\text{H}\)) in strong static magnetic fields. Sensing magnetic susceptibility shifts (\(\Delta\chi\)) induced by paramagnetic deoxyhemoglobin.
Injecting NIR light (\(650 - 950\,\text{nm}\)) into the tissue diagnostic window. Measuring differential absorption between oxy- and deoxyhemoglobin across highly scattering paths.
Detecting extracellular volume currents and associated magnetic flux generated by synchronous postsynaptic currents in aligned cortical pyramidal assemblies.
Transmitting high-frequency mechanical pressure waves through cranial windows or microbubbles to track red blood cell backscatter or deliver focused mechanical force.
Magnetic Resonance Imaging relies on the intrinsic magnetic dipole moment $\vec{\mu}$ of hydrogen nuclei (protons, $^1\text{H}$) omnipresent in tissue water ($\approx 70 - 80\%$ of brain mass). The proton possesses spin angular momentum $\vec{S}$ with spin quantum number $I = 1/2$.
In the absence of an external magnetic field, nuclear spin orientations are isotropically disordered, producing zero net macroscopic magnetization. When placed inside an external static magnetic field $\vec{B}_0 = B_0 \hat{k}$, the degeneracy of the magnetic energy levels splits into two eigenstates (Zeeman splitting):
where $\gamma$ is the gyromagnetic ratio ($\gamma/2\pi \approx 42.576\,\text{MHz/T}$ for $^1\text{H}$) and $\hbar$ is the reduced Planck constant. The population ratio between parallel ($N_\uparrow$, lower energy) and antiparallel ($N_\downarrow$, higher energy) states follows the Boltzmann distribution:
At physiological temperature ($T = 310\,\text{K}$) and $B_0 = 3.0\,\text{T}$, $\Delta E \ll k_B T$. A first-order Taylor expansion reveals an excess of only $\approx 10$ protons per million in the lower energy state. Summed over $N_s$ spins per unit volume, this quantum thermal bias produces the macroscopic net equilibrium magnetization vector $\vec{M}_0$:
Because $M_0 \propto B_0$ and the induced Faraday voltage in an RF receiver coil scales with precession frequency $\omega_0 = \gamma B_0$, the intrinsic MR Signal-to-Noise Ratio (SNR) scales superlinearly: $\text{SNR} \propto B_0^\alpha$, where $\alpha \approx 1.5 - 2.0$ depending on whether body noise or coil resistance dominates. Moving from low-field ($0.05\,\text{T}$) to clinical high-field ($3.0\,\text{T}$) yields a $\sim 60\times$ increase in raw magnetization.
Applying a resonant radiofrequency field $\vec{B}_1(t)$ at the Larmor frequency $\omega_0$ tips the magnetization vector into the transverse plane ($x\text{-}y$). Once perturbed, the evolution of $\vec{M}(t)$ is governed by the phenomenological Bloch Equations:
Decomposed into longitudinal ($M_z$) and transverse ($M_{xy} = M_x + i M_y$) components in a reference frame rotating at $\omega_0$:
Here, $T_1$ represents spin-lattice energy dissipation to thermal lattice vibrations, while $T_2$ represents irreversible spin-spin dephasing due to microscopic molecular dipolar interactions.
Functional MRI does not measure neural firing directly; it exploits the Blood Oxygenation Level Dependent (BOLD) effect discovered by Seiji Ogawa in 1990.
The molecular basis is iron electronic spin configuration in hemoglobin:
When red blood cells carrying HbR flow through microvessels (capillaries and venules), they generate microscopic spatial gradients in the local magnetic field $\Delta B(\vec{r}) \propto \Delta\chi \cdot B_0$. Water protons diffusing through these inhomogeneous field gradients experience different precession rates and rapidly dephase, shortening the effective transverse relaxation time $T_2^*$:
When a cortical neural circuit activates, oxygen consumption ($\text{CMRO}_2$) increases, initially producing a brief surge in HbR (the "initial dip"). However, within $1 - 2$ seconds, local neurovascular coupling mechanisms release vasoactive mediators (nitric oxide, adenosine, arachidonic acid metabolites) that dilate feeding pial arterioles. Cerebral blood flow ($\text{CBF}$) surges by $30 - 60\%$, vastly outstripping the modest $5 - 15\%$ increase in oxygen consumption.
This "hyperemic flush" dilutes venous deoxyhemoglobin, causing $[\text{HbR}]$ to drop below baseline. Consequently, $T_2^*$ lengthens, dephasing diminishes, and the $T_2^*$-weighted gradient echo MR signal increases by $1 - 5\%$ at $3\,\text{T}$.
Functional Near-Infrared Spectroscopy illuminates the head with low-power continuous wave (CW) or frequency-domain NIR light. Optical imaging of deep tissue is made possible by the diagnostic optical window ($650 - 950\,\text{nm}$), where water and lipid absorption reach their spectral minima, allowing photons to penetrate several centimeters into tissue before being absorbed.
Biological tissue is a turbid medium where scattering vastly exceeds absorption. Typical brain tissue parameters in the near-infrared are:
Because photons undergo hundreds of scattering events before absorption, their transport is governed by the reduced scattering coefficient:
Under the diffusion approximation ($\mu_s' \gg \mu_a$), the full Boltzmann Radiative Transfer Equation simplifies to the parabolic photon diffusion equation for photon fluence rate $\Phi(\vec{r}, t)$:
For an optical source and detector placed on the scalp surface separated by distance $d_{\text{SD}}$, the spatial sensitivity distribution of detected photons follows a curvilinear, banana-shaped photon migration path.
At standard adult optode spacing of $d_{\text{SD}} = 30\,\text{mm}$, photons penetrate to a mean depth of $\approx 12 - 15\,\text{mm}$, grazing the superficial gray matter of the gyral crowns. Widening the separation to $40\,\text{mm}$ increases cortical interrogation but penalizes detected optical power exponentially ($e^{-\mu_{\text{eff}} d_{\text{SD}}}$), demanding high-dynamic-range avalanche photodiodes (APDs) or silicon photomultipliers (SiPMs).
In a purely absorbing non-scattering medium, light attenuation follows the classic Beer-Lambert law $A = \log_{10}(I_0 / I) = \varepsilon \cdot C \cdot d$. In scattering tissue, the optical path length is elongated by a Differential Pathlength Factor (DPF) ($\approx 6.0$ in adult head), and an intensity loss factor $G$ arises from geometric scattering loss:
Taking temporal differential optical density relative to a baseline eliminates the unknown geometry factor $G$:
By measuring at two strategically chosen wavelengths—typically straddling the isosbestic point ($\approx 805\,\text{nm}$ where $\varepsilon_{\text{HbO}_2} = \varepsilon_{\text{HbR}}$), such as $760\,\text{nm}$ (sensitive to HbR) and $850\,\text{nm}$ (sensitive to $\text{HbO}_2$)—we solve the linear matrix system to unmix absolute molar concentration changes:
Because photons must pass twice through the scalp and skull (once going in, once coming out), up to $80\%$ of the detected optical attenuation change can arise from systemic scalp hemodynamics (cardiac pulsation, respiratory Mayer waves, autonomic skin vasoconstriction). Modern fNIRS instrumentation requires short-separation channels ($d_{\text{SD}} \approx 8\,\text{mm}$) that measure only superficial scalp blood flow, which is then subtracted from the long channel ($30\,\text{mm}$) via adaptive Kalman filtering.
Unlike fMRI and fNIRS, which measure slow metabolic and hemodynamic byproducts of neural activity ($1 - 5\,\text{s}$ lag), electroencephalography (EEG) and magnetoencephalography (MEG) measure the instantaneous electric and magnetic fields generated directly by neuronal transmembrane currents ($< 1\,\text{ms}$).
A common misconception is that EEG reflects action potentials. In reality, an action potential consists of two equal and opposite current loops propagating in opposite directions down an axon over $1\,\text{ms}$; the quadrupole field decays with distance as $1/r^3$, vanishing before reaching the scalp.
Instead, EEG and MEG are driven by excitatory postsynaptic potentials (EPSPs) on the apical dendrites of Layer V cortical pyramidal cells:
Because cortical pyramidal cells are arranged in perpendicular, columnar palisades (an "open-field" architecture), thousands of synchronized synaptic events summate constructively:
For neural signals of interest ($f < 1000\,\text{Hz}$), the electromagnetic wavelength in tissue is $\lambda = c / (f \sqrt{\varepsilon_r}) \approx 300\,\text{km}$, which dwarfs head dimensions ($0.2\,\text{m}$). The capacitive displacement current $\partial \vec{D}/\partial t$ is negligible compared to ohmic conduction current $\vec{J} = \sigma \vec{E}$ ($\omega \varepsilon / \sigma \sim 10^{-4}$). Therefore, Maxwell's equations reduce to the quasi-static formulation:
where $\sigma(\vec{r})$ is tissue electrical conductivity and $\vec{J}_p$ is the primary impressed biological current density.
The human head is modeled as concentric conductive shells: brain ($\sigma_b \approx 0.33\,\text{S/m}$), skull ($\sigma_s \approx 0.0042 - 0.01\,\text{S/m}$), and scalp ($\sigma_{\text{sc}} \approx 0.33\,\text{S/m}$). The skull acts as an electrical insulator with a conductivity ratio $\sigma_{\text{brain}} / \sigma_{\text{skull}} \approx 20:1 - 80:1$.
EEG measures electric potentials $\Phi$ that must conduct through the resistive skull. The skull acts as a spatial low-pass filter, spreading point currents outward into broad, blurred voltage topographies on the scalp ($20 - 30\,\text{mm}$ resolution).
MEG measures magnetic flux densities $\vec{B}$ via SQUIDs or optically pumped magnetometers (OPMs). Because biological tissue has magnetic permeability $\mu_r \approx 1.000000$ (identical to vacuum), magnetic fields pass through the skull and scalp completely unrefracted and undistorted. By the Biot-Savart law:
In a spherical conductor, radial currents produce zero external magnetic field ($\vec{B}_{\text{radial}} = 0$). MEG is exclusively sensitive to tangential dipoles located in the sulcal walls, providing sharper spatial localization ($3 - 5\,\text{mm}$) than scalp EEG.
Ultrasound introduces mechanical stress waves into neural tissue. In functional ultrasound (fUS), high-frame-rate plane-wave Doppler imaging tracks microvascular red blood cell velocity with sub-millimeter resolution. In focused ultrasound neuromodulation (FUS), concentrated acoustic radiation force opens mechanosensitive ion channels or transiently alters lipid bilayer capacitance.
Compressional acoustic waves traveling through viscoelastic brain tissue obey the linear wave equation with thermoviscous attenuation:
While soft brain tissue has acoustic impedance $Z = \rho c \approx 1.54 \times 10^6\,\text{Rayl}$ and modest attenuation ($\approx 0.5\,\text{dB/cm/MHz}$), cranial cortical bone is an extreme acoustic mismatch ($Z \approx 6.0 \times 10^6\,\text{Rayl}$, attenuation $\approx 10 - 20\,\text{dB/cm/MHz}$). Transcranial ultrasound encounters:
Conventional clinical ultrasound uses focused scan lines ($50\,\text{Hz}$ frame rate), which lack the sensitivity to detect slow microvascular blood flow ($< 1\,\text{mm/s}$) in small parenchymal capillaries.
fUS overcomes this barrier by emitting unfocused tilted plane waves at $5 - 20\,\text{kHz}$ pulse repetition frequency (PRF). Coherent compounding of multiple angles (e.g., $-10^\circ$ to $+10^\circ$) boosts SNR while sustaining frame rates above $1\,\text{kHz}$:
Applying advanced spatio-temporal Singular Value Decomposition (SVD) filters separates the slow-moving red blood cell backscatter from high-amplitude tissue pulsatile clutter, achieving a $50\times$ sensitivity gain over standard Doppler, resolving $100\,\mu\text{m}$ microvascular changes in awake mobile subjects.
No single neuroimaging modality can conquer all physical dimensions simultaneously. Every sensing architecture accepts an inescapable trade-off between penetration depth, spatial resolution, temporal fidelity, and patient invasiveness dictated by its underlying carrier physics.
| Modality | Biophysical Origin | Spatial Limit | Temporal Limit | Depth Limit | Fundamental SNR Constraint |
|---|---|---|---|---|---|
| fMRI (BOLD) | Magnetic susceptibility gradient ($\Delta\chi$) of HbR; $T_2^*$ spin dephasing | $\sim 0.5 - 1.5\,\text{mm}$ (capillary bed point spread) | $\sim 1 - 2\,\text{s}$ (hemodynamic impulse response) | Full human brain ($15 - 20\,\text{cm}$) | Thermal Johnson noise in tissue; $\text{SNR} \propto B_0 \cdot \Delta V \cdot \sqrt{t_{\text{acq}}}$ |
| fNIRS / DOT | Mie photon scattering ($\mu_s'$) & chromophore absorption ($\mu_a$) | $\sim 10 - 20\,\text{mm}$ (diffusive spatial spreading) | $\sim 10 - 100\,\text{ms}$ (detector sampling rate) | $\sim 15 - 25\,\text{mm}$ ($z \approx \frac{1}{2} d_{\text{SD}}$) | Photon shot noise; exponential attenuation ($e^{-\mu_{\text{eff}} d}$) |
| EEG | Pyramidal apical EPSPs; ionic volume currents through skull | $\sim 20 - 30\,\text{mm}$ (skull resistive blurring) | $< 1\,\text{ms}$ (instantaneous quasi-static field) | Surface cortex (deep sources severely attenuated) | Electrode-skin impedance, thermal electronic noise, environmental 60 Hz |
| MEG | Intracellular primary currents in sulcal gyri; magnetic flux $\vec{B}$ | $\sim 3 - 5\,\text{mm}$ (high sensor density + MRI prior) | $< 1\,\text{ms}$ (instantaneous electromagnetic) | Gyral cortex ($\sim 3 - 5\,\text{cm}$; falls as $1/r^2$) | Environmental ambient magnetic noise ($10^6 \times$ signal); SQUID noise |
| fUS | Acoustic backscatter from moving red blood cells (Power Doppler) | $\sim 100 - 200\,\mu\text{m}$ (acoustic diffraction limit $\lambda/2$) | $\sim 10 - 100\,\text{ms}$ (ultrafast plane-wave compounding) | $\sim 3 - 10\,\text{cm}$ (requires cranial window in adults) | Bone acoustic attenuation ($\approx 15\,\text{dB/cm/MHz}$) and thermal limits |
In Worksheet 4.1, your engineering team is required to build a computational forward model for your chosen clinical diagnosis. Follow this rigorous 4-step workflow: