Brain Imaging Methods: Structural and Functional

Brain imaging methods are ways of converting some physical property of neural tissue — hydrogen density, blood oxygenation, glucose consumption, radiotracer binding, electrical potential, magnetic field, or water diffusion — into a spatial map that a researcher can interpret. Broadly, methods divide into structural imaging, which reveals anatomy (shape, size, tissue integrity, connectivity), and functional imaging, which reveals dynamic processes (activity, metabolism, neurotransmitter binding) that change from moment to moment or task to task. This distinction is not absolute — a structural scan of a shrunken hippocampus in Neurodegeneration: Alzheimer's Disease carries functional meaning, and functional maps require structural scans as an anatomical reference — but it organizes how the tools are built and used. A second key distinction is between methods that measure hemodynamic/metabolic signals (fMRI, PET) and methods that measure electromagnetic signals directly generated by neurons (EEG, MEG). Hemodynamic methods rely on the tight coupling between local neural activity and local blood supply, called neurovascular coupling: active neurons demand more oxygen and glucose, triggering a local increase in blood flow that overshoots actual oxygen consumption. This gives excellent spatial resolution (down to ~1–2 mm for fMRI) but poor temporal resolution, because the vascular response lags neural firing by several seconds. Electromagnetic methods detect the actual ionic currents flowing during synaptic transmission and action potentials, offering millisecond temporal resolution, but because the skull and scalp blur and attenuate these tiny fields, spatial localization is inherently ambiguous — a problem formally known as the inverse problem (the same surface signal can arise from many different internal source configurations). A third axis is invasiveness and molecular specificity. PET requires injecting radioactive tracers and involves ionizing radiation, but it is the only common technique that can directly quantify specific molecular targets — dopamine D2 receptors, amyloid-β plaques, glucose metabolism — with a precision that is essential for connecting the neurochemical theories in psychiatric and neurodegenerative disease to actual living human tissue. MRI-based methods (structural MRI, fMRI, DTI) are noninvasive and radiation-free, making them repeatable in the same subject over years, which is precisely why they dominate developmental and longitudinal disease-tracking studies.

Structural MRI: hydrogen, magnets, and relaxation times. MRI exploits the fact that hydrogen nuclei (single protons) behave like tiny spinning magnets. Placed in a strong external magnetic field (clinical scanners typically use 1.5–3 Tesla; research scanners reach 7 Tesla or higher — for comparison, Earth's magnetic field is about 0.00005 Tesla), a majority of these proton "spins" align with the field, producing a net magnetization vector. A radiofrequency (RF) pulse tuned to the proton's resonant (Larmor) frequency tips this magnetization away from alignment. When the pulse stops, the protons relax back toward equilibrium, emitting a detectable RF signal as they do so. Two relaxation processes matter:

• T1 relaxation (longitudinal relaxation): how quickly the net magnetization realigns with the main field. Fat has short T1 (bright on T1-weighted images); cerebrospinal fluid (CSF) has long T1 (dark).
• T2 relaxation (transverse relaxation): how quickly spinning protons lose phase coherence with each other. CSF has long T2 (bright on T2-weighted images); this contrast is exquisitely sensitive to edema, demyelination, and stroke.

Because gray matter, white matter, and CSF have distinct hydrogen densities and relaxation properties, differential weighting of the pulse sequence produces high-contrast anatomical images — the basis of every structural MRI scan used to measure hippocampal volume in Alzheimer's disease or localize a tumor before surgery.

Functional MRI: the BOLD signal. fMRI does not detect neurons directly. It detects the blood-oxygen-level-dependent (BOLD) signal, which depends on the ratio of oxygenated hemoglobin (diamagnetic, does not distort the local magnetic field) to deoxygenated hemoglobin (paramagnetic, distorts the local field and speeds T2 decay). The causal chain is: 1. A neural population becomes active (increased synaptic and spiking activity, driven largely by the energy demands of restoring ionic gradients after synaptic transmission). 2. Local energy demand rises, and within roughly 1–2 seconds a delayed, disproportionately large increase in local blood flow occurs (neurovascular coupling), regulated by astrocytes, interneurons, and vasoactive signaling molecules such as nitric oxide. 3. This flow increase overcompensates for the actual rise in oxygen extraction, so the local ratio of oxy- to deoxy-hemoglobin actually increases — counterintuitively, active brain tissue becomes relatively more oxygenated. 4. Because deoxyhemoglobin is paramagnetic and dephases the MR signal, less deoxyhemoglobin means a locally brighter signal on T2-weighted images. 5. The full time course — the hemodynamic response function (HRF) — rises over about 4–6 seconds to a peak, then falls back over 10–20 seconds, sometimes dipping below baseline (the "post-stimulus undershoot"). This means fMRI has excellent spatial resolution (~1–3 mm in standard research scanners) but temporal resolution limited to seconds, not milliseconds — a fundamental mismatch with the millisecond timescale of actual neural computation. fMRI is also inherently relative: BOLD signal change is compared statistically between task and control conditions, not measured as an absolute quantity.

PET: direct molecular measurement via radioactive decay. Positron emission tomography involves injecting a molecule labeled with a positron-emitting radioisotope (commonly fluorine-18, carbon-11, or oxygen-15, with half-lives ranging from about 2 minutes to 110 minutes). As the isotope decays, it emits a positron that travels a short distance (~1 mm) before annihilating with an electron, producing two 511 keV gamma photons that fly off in almost exactly opposite directions. Detector rings surrounding the head register these coincident photon pairs; because the two photons arrive at nearly the same instant along a straight line, the scanner can reconstruct the three-dimensional location of the decay event. Different tracers answer different questions:

• [¹⁸F]fluorodeoxyglucose (FDG-PET): a glucose analog, quantifies regional glucose metabolism — reduced in specific temporo-parietal regions in Alzheimer's disease, sometimes years before symptoms.
• Receptor ligands (e.g., [¹¹C]raclopride for dopamine D2 receptors): quantify receptor density and occupancy, directly testing neurochemical hypotheses of schizophrenia and addiction discussed under Neurochemistry of Psychiatric Disorders.
• Amyloid and tau ligands (e.g., Pittsburgh Compound B, [¹⁸F]flortaucipir): bind selectively to pathological protein aggregates, allowing in vivo staging of Alzheimer's pathology that was previously possible only at autopsy.

PET's temporal resolution is limited (seconds to minutes per scan, since sufficient photon counts must accumulate), its spatial resolution is modest (~4–6 mm, worse than MRI, partly due to positron range and photon non-collinearity), and it necessarily involves ionizing radiation, restricting how often a given subject can be scanned. Its unique strength is specificity: no other common technique can say, in a living person, "this many dopamine receptors are occupied by this drug."

EEG and MEG: watching neurons think, in real time. When many pyramidal neurons in cortex are simultaneously depolarized by synchronous synaptic input (chiefly excitatory postsynaptic potentials at apical dendrites), their aligned dendritic currents create an "open field" dipole large enough to be detected at the scalp.

• EEG records the resulting voltage differences directly with scalp electrodes (typically 32–256 channels). Temporal resolution is on the order of milliseconds. Spatial resolution is poor (centimeters) because the skull, an electrically resistive layer, smears the current before it reaches the scalp.
• MEG records the tiny magnetic fields produced by the same intracellular currents, using arrays of superconducting quantum interference devices (SQUIDs) cooled in liquid helium. Magnetic fields pass through the skull largely undistorted, so MEG offers somewhat better spatial localization than EEG for the same millisecond temporal resolution, though it is far more expensive and preferentially detects currents tangential to the scalp surface (from sulcal cortex) rather than radial ones.

Both methods face the inverse problem: infinitely many combinations of internal current sources could produce an identical pattern of surface signals, so source localization always requires additional assumptions (e.g., a small number of discrete dipoles, or a distributed cortical current model) and is inherently less certain than the "forward problem" of predicting scalp signal from a known source. EEG's millisecond resolution makes it the tool of choice for tracking event-related potentials (ERPs) and for detecting the hypersynchronous spike-and-wave discharges central to Epilepsy and Neuronal Hyperexcitability, often localizing a seizure focus that later guides invasive electrode placement or surgery.

Diffusion Tensor Imaging: mapping the wiring. DTI is a variant of MRI sensitive to the random thermal (Brownian) motion of water molecules. In an unrestricted medium like CSF, water diffuses roughly equally in all directions (isotropic diffusion). Inside a myelinated axon, diffusion is heavily restricted perpendicular to the fiber (by the axonal membrane and myelin sheath) but relatively free along its length — anisotropic diffusion. By applying diffusion-sensitizing magnetic gradients along at least six independent directions and measuring signal loss in each, a diffusion tensor can be computed for every voxel, summarizing the three-dimensional shape of local diffusion as an ellipsoid. The tensor's principal axis indicates the dominant local fiber direction, and a scalar summary called fractional anisotropy (FA), ranging from 0 (fully isotropic) to 1 (fully directional), indexes how strongly diffusion is constrained along one axis — commonly interpreted (with caveats) as a proxy for white-matter tract integrity, myelination, and axonal coherence. Tractography algorithms then chain together the local fiber directions voxel by voxel to reconstruct entire white-matter pathways, such as the arcuate fasciculus connecting language areas, producing the closest available in vivo approximation of the human structural connectome.

Establishing the BOLD principle (Ogawa et al., 1990–1992). Seiji Ogawa and colleagues at AT&T Bell Laboratories first demonstrated, in anesthetized rats scanned at high magnetic field strength, that blood vessels became visibly darker on T2*-weighted MR images when the animals breathed low-oxygen air, because deoxyhemoglobin's paramagnetism distorted the local magnetic field and hastened signal decay. Reversing the manipulation (breathing pure oxygen) made the vessels less visible. This established, in a controlled physiological manipulation, that MRI signal intensity tracks blood oxygenation — the physical basis of BOLD contrast. Shortly after, in 1992, three independent groups (Ogawa's team, Kenneth Kwong and colleagues at Massachusetts General Hospital, and Peter Bandettini and colleagues at the Medical College of Wisconsin) showed BOLD signal changes in human visual and motor cortex during simple visual stimulation or finger-tapping tasks, without any injected contrast agent. This was the founding demonstration of fMRI as a noninvasive, repeatable technique for imaging human brain function — a landmark comparable in importance to the identification of the action potential itself for the study of systems-level brain organization.

Validating BOLD against ground-truth neural activity (Logothetis et al., 2001). A critical open question was exactly what BOLD reflects: spiking output, or local synaptic input and processing? Nikos Logothetis and colleagues addressed this directly by simultaneously recording intracortical electrodes (measuring both multi-unit spiking activity and local field potentials, which largely reflect summed synaptic currents) and BOLD fMRI signal in the visual cortex of anesthetized monkeys during visual stimulation. They found that BOLD signal amplitude correlated more strongly and more reliably with the local field potential than with spiking activity, and that BOLD could remain elevated even when spiking activity declined, as long as synaptic input persisted. This showed that fMRI is best interpreted as an index of the net synaptic input and local processing a brain region receives, not simply of how many action potentials its neurons fire — a distinction with real interpretive consequences whenever fMRI is used to claim a region "is activated."

PET and glucose metabolism in disease (Reivich, Sokoloff, and colleagues, 1970s–1980s). The FDG-PET method rests on the 2-deoxyglucose autoradiographic technique developed by Louis Sokoloff and colleagues in animals, later adapted for human PET imaging by Martin Reivich, Alfred Wolf, and colleagues using fluorine-18-labeled deoxyglucose. Because 2-deoxyglucose is transported into cells and phosphorylated like normal glucose but cannot proceed further in glycolysis, it becomes metabolically "trapped," and its local accumulation (measured by radioactivity) is proportional to local glucose consumption — a direct, quantifiable index of regional neural energy use. Applying this to Alzheimer's disease, multiple groups through the 1980s–1990s showed a characteristic pattern of reduced glucose metabolism in temporo-parietal association cortex and posterior cingulate, detectable before overt structural atrophy — establishing PET hypometabolism as one of the earliest in vivo biomarkers of neurodegenerative disease, well before autopsy-confirmed amyloid pathology could be imaged directly.

Diffusion anisotropy and white matter (Moseley et al., 1990; Basser et al., 1994). Michael Moseley and colleagues first showed, in cat brain, that water diffusion measured by MRI is markedly anisotropic in white matter and that this anisotropy is dramatically reduced within minutes of experimentally induced ischemic stroke — one of the earliest demonstrations that diffusion MRI could detect stroke far earlier than conventional T1/T2 imaging, a finding that transformed emergency stroke diagnosis. Peter Basser, James Mattiello, and Denis Le Bihan then formalized the mathematics of the diffusion tensor (1994), providing the framework — the full 3D tensor rather than a single diffusion coefficient — that converts directional diffusion measurements into fiber orientation estimates, the foundation of modern tractography.

Example 1 — Choosing a modality for a clinical question. A 58-year-old presents with sudden right-arm weakness and slurred speech, symptom onset 40 minutes ago; the team must decide whether to give clot-dissolving therapy, which is dangerous if there is instead a hemorrhage. Reasoning through modality trade-offs: PET is too slow (minutes to hours from tracer synthesis to image, plus it does not visualize hemorrhage well) and inappropriate for an emergency. EEG can detect global dysfunction but cannot localize a discrete lesion or distinguish stroke type. Structural CT, not covered in detail here, is fast and detects hemorrhage well but is poor at detecting very early ischemia. Diffusion-weighted MRI (a DTI-related sequence) is the correct choice: cytotoxic edema in acutely ischemic tissue restricts water diffusion within minutes (reduced apparent diffusion coefficient, appearing bright on diffusion-weighted images) long before conventional T1/T2 changes appear, exactly as Moseley's 1990 experiment demonstrated — allowing the team to both confirm ischemic stroke and localize it within the roughly one-hour treatment window.

Example 2 — Quantitative PET receptor occupancy. A new antipsychotic drug is tested with [¹¹C]raclopride PET, which competes with endogenous dopamine for striatal D2 receptors. Before drug administration, the binding potential (a PET-derived index proportional to available receptor density) in the striatum is measured at 3.0. After a therapeutic dose, binding potential falls to 0.6. Receptor occupancy is calculated as: Occupancy = (BP_baseline − BP_drug) / BP_baseline × 100% = (3.0 − 0.6) / 3.0 × 100% = 80% An occupancy of roughly 65–80% at striatal D2 receptors is the threshold empirically associated with clinical antipsychotic efficacy in schizophrenia, while occupancy above ~80% sharply increases the risk of extrapyramidal (parkinsonian) side effects, since it encroaches on the nigrostriatal motor dopamine pathway alongside the mesolimbic pathway implicated in psychosis (see Neurochemistry of Psychiatric Disorders). This single number — 80% — tells the clinical team the drug is likely effective but is dosed close to the side-effect threshold, motivating a dose reduction, a conclusion impossible to reach without direct molecular imaging of a living human brain.

Seiji Ogawa trained as a physical chemist, working for decades at Bell Laboratories on nuclear magnetic resonance long before it was applied to brains. His insight linking blood oxygenation to MR signal did not arise from neuroscience at all — it arose from basic physical chemistry: he understood that deoxyhemoglobin's unpaired electrons make it paramagnetic, a fact known since Linus Pauling's studies of hemoglobin magnetism in the 1930s, and he recognized this could be exploited as an endogenous, injection-free contrast mechanism for imaging brain physiology. This is a recurring pattern in imaging science: transformative neuroscience tools have repeatedly originated from physicists and chemists solving a physical problem first, with the neuroscience application following.

Presurgical fMRI mapping of language and motor cortex is now routine practice, allowing neurosurgeons to plan tumor resections that spare critical function — directly answering the opening question of this article. Amyloid- and tau-PET imaging are now used both in Alzheimer's research trials, to select patients whose pathology matches a drug's target and to track whether an anti-amyloid antibody actually reduces plaque burden, and increasingly in clinical diagnosis. FDG-PET remains a key tool for localizing the seizure focus in patients with epilepsy being considered for surgery, complementing EEG's electrical localization discussed under Epilepsy and Neuronal Hyperexcitability. Diffusion tractography guides neurosurgeons around critical white-matter tracts and is a primary emergency tool for early stroke diagnosis worldwide. In psychiatry, PET receptor-occupancy studies have directly shaped rational drug dosing for antipsychotics and antidepressants, replacing pure trial-and-error titration. EEG remains the frontline, low-cost, portable tool for diagnosing seizure disorders, sleep disorders, and monitoring depth of anesthesia. Looking forward, high-density EEG and implanted electrode recordings that localize function with millisecond precision are foundational to the signal-decoding problem at the heart of Neural Prosthetics and Brain-Computer Interfaces, where the very same questions this article raises — what does this signal actually reflect, and how precisely can its source be localized — become the engineering constraints on a device that must read a person's intended movement or speech directly from brain activity.

Ask the tutor: How would you design a combined imaging study to determine whether a candidate antidepressant changes both receptor occupancy and functional connectivity in the same patients?