Brain Signals as Data

Studying the brain computationally means working with neural recordings — data streams that reflect underlying neural activity. But the signals neuroscientists measure are not direct readouts of individual neuron activity; they are spatial and temporal averages, filtered through tissue, physics, and measurement technology. Understanding what each measurement captures, at what resolution, and with what artifacts is essential before applying any analysis.

This chapter surveys the major neural recording modalities, their signal origins, and the data properties that shape analysis choices.

The Signal Hierarchy: Scales of Neural Activity

Neural activity can be measured across a hierarchy of spatial and temporal scales:

No single modality captures everything. Invasive methods (single unit, LFP) have high resolution but require surgery. Non-invasive methods (EEG, fMRI) have low resolution but can be applied to humans safely.

EEG: Electrical Activity at the Scalp

Electroencephalography (EEG) places electrodes on the scalp and measures the voltage differences produced by synchronous neural activity. The measured signal reflects the summed postsynaptic currents of large populations of cortical pyramidal neurons with aligned geometry.

The Signal

Why can we measure neural signals at the scalp? A single neuron's current is unmeasurably small at distance. But when thousands of pyramidal neurons in a cortical patch are simultaneously active and geometrically aligned (perpendicular to the cortical surface), their current dipoles add constructively → a detectable electrical field at the scalp.

Signal amplitude: 10–100 μV. Frequency content: mostly 0.1–100 Hz (useful range).

What does EEG measure? Primarily the synchronous postsynaptic currents (not spikes) of large cortical ensembles. Action potentials are too brief and asynchronous to produce measurable scalp signals. Deep sources (hippocampus, brainstem) are hard to detect because the signal attenuates with distance and is blurred by the skull (which is a poor electrical conductor).

EEG Frequency Bands

The oscillatory structure of EEG carries interpretable information:

• Delta (0.5–4 Hz): slow oscillations during deep sleep. Delta power inversely correlates with sleep quality degradation. Abnormally high delta in wakefulness suggests encephalopathy or focal brain damage.

• Theta (4–8 Hz): prominent during drowsiness and in the hippocampus during spatial navigation and memory encoding. Frontal midline theta is associated with working memory load.

• Alpha (8–12 Hz): the dominant EEG rhythm in relaxed wakefulness with eyes closed. Strongest over posterior cortex. Alpha power is a proxy for cortical inhibition — high alpha = less processing in that region. Used in neurofeedback and BCI.

• Beta (12–30 Hz): prominent during motor activity and active thinking. Beta decreases before and during movement (event-related desynchronization, ERD) and returns after movement completion (event-related synchronization, ERS). Used in motor BCI.

• Gamma (30–100 Hz): associated with active sensory processing and attention. Hard to measure reliably at scalp EEG because it's weak, easily contaminated by muscle artifacts, and requires high-frequency sampling.

EEG Artifacts

EEG is notoriously artifact-prone. Common sources:

• Eye blinks and movements (EOG): large electrical deflections from eye muscles; dominate frontal channels
• Muscle activity (EMG): high-frequency broadband noise, especially from jaw, neck, and scalp muscles
• Heart signals (ECG): regular QRS complex artifacts, especially in neck electrodes
• Movement: electrode motion produces large transient artifacts
• Line noise: 50 or 60 Hz interference from electrical equipment

Preprocessing pipeline always includes: band-pass filtering → artifact removal (ICA, regression, or manual rejection) → re-referencing.

MEG: Magnetic Fields

Magnetoencephalography (MEG) measures the magnetic fields produced by neural currents using superconducting quantum interference devices (SQUIDs). It detects the same synchronous neural currents as EEG but from a different physical property:

• MEG is sensitive to tangential (horizontally oriented) sources; less sensitive to radial sources
• The skull does not distort magnetic fields → better spatial resolution than EEG
• MEG requires a magnetically shielded room and cryogenic equipment → expensive, stationary systems

MEG and EEG provide complementary information and are often combined. MEG is particularly valuable for presurgical epilepsy mapping (pinpointing the seizure focus) and auditory/language neuroscience.

fMRI: The BOLD Signal

Functional MRI (fMRI) measures the BOLD (Blood-Oxygen-Level-Dependent) signal — changes in the ratio of oxygenated to deoxygenated hemoglobin, which have different magnetic properties. Neural activity → increased local metabolic demand → vasodilation and increased blood flow → change in oxy/deoxy hemoglobin ratio → BOLD signal change.

The Hemodynamic Response Function (HRF)

The BOLD signal is an indirect and temporally blurred measure of neural activity:

• The HRF peaks ~5–6 seconds after neural activity onset
• It has a characteristic undershoot after return to baseline
• Duration: full HRF lasts ~20–30 seconds

This ~5 second lag means fMRI cannot distinguish events closer than 2–3 seconds (temporal resolution). But spatial resolution (1–3 mm isotropic) and whole-brain coverage make fMRI the dominant tool for studying where in the brain things happen.

What fMRI Doesn't Tell You

A common misconception: BOLD activation means "this region is active." More precisely: BOLD signal change indicates a hemodynamic response, which correlates with net synaptic input (predominantly excitatory and inhibitory inputs to a region, not just spike output). An inhibitory neuron increasing its firing can produce BOLD activation.

Neurovascular coupling — the relationship between neural activity and blood flow — is not perfectly understood and can be modified by drugs, age, and disease, complicating interpretation.

Calcium Imaging: Watching Neurons Optically

Two-photon calcium imaging allows visualizing the activity of hundreds to thousands of individual neurons simultaneously in awake, behaving animals.

Mechanism: cells are loaded with a fluorescent calcium indicator (genetically encoded indicators like GCaMP are now standard). When a neuron fires, Ca²⁺ enters via voltage-gated channels → GCaMP fluorescence increases → imaged with a two-photon microscope.

Properties:

• Single-cell resolution in intact tissue
• Can image >1000 neurons simultaneously over weeks (chronic imaging with chronic windows)
• Temporal resolution ~50–100 ms (calcium signal kinetics) — slower than electrophysiology
• Requires surgery (cranial window) in mice

Two-photon imaging has revolutionized systems neuroscience — enabling large-scale population recordings in awake behaving animals to study motor control, sensory processing, and memory formation.

Data Properties and Analysis Implications

Each modality produces data with characteristic properties that determine valid analyses:

The multiple comparisons problem is severe in brain data: testing 50,000 voxels or 5000 neurons at p < 0.05 produces ~2500 or ~250 false positives by chance. Corrections (FWE, FDR, permutation testing) are essential.

Clinical Applications

These recording modalities are not only research tools:

EEG in clinical neurology: seizure diagnosis, encephalopathy grading, sleep staging, intraoperative monitoring, and increasingly, BCI for communication in ALS patients.

fMRI in presurgical planning: mapping language (Broca's/Wernicke's areas) and motor cortex relative to a tumor to guide safe resection margins.

Quantitative EEG (qEEG): biomarkers of anesthetic depth (bispectral index, BIS), sedation level in ICU patients, and prognostics in coma.

Understanding the signal physics behind each modality is not just theoretical — it determines what you can and cannot conclude from the data, which preprocessing steps are appropriate, and what artifacts might produce false results. The analysis chapters that follow use EEG data specifically, where you'll apply these concepts directly.