Chapter 2: EEG Technology and Brain Wave Patterns

Electroencephalography (EEG) represents the cornerstone technology underlying all neurofeedback applications. This chapter explores the technical foundations of EEG signal acquisition, the neurophysiology of brain wave generation, electrode placement systems, signal processing methods, artifact identification and removal, and spectral analysis techniques. Understanding these technical aspects is essential for clinicians and researchers implementing neurofeedback systems that comply with WIA-MENTAL-009 standards.

EEG measures electrical potential differences on the scalp surface that reflect the summed activity of millions of cortical neurons firing in synchrony. Unlike individual action potentials that last only milliseconds, EEG captures slower oscillating patterns of post-synaptic potentials in cortical pyramidal neurons arranged in parallel. When large populations of these neurons fire rhythmically together, their electrical fields summate and become detectable at the scalp surface, albeit at extremely small amplitudes measured in microvolts (millionths of a volt).

The generation of detectable EEG signals requires several conditions. First, large populations of neurons (thousands to millions) must be active simultaneously. Second, these neurons must be spatially aligned in parallel orientations—cortical pyramidal cells with their dendrites perpendicular to the brain surface are ideally configured for this. Third, the neurons must fire synchronously to generate coherent electrical fields rather than random activity that cancels out. These requirements mean that EEG primarily reflects activity from superficial cortical layers rather than deep brain structures, and captures synchronized oscillatory activity rather than individual neuronal spikes.

The International 10-20 Electrode Placement System

Standardized electrode placement is crucial for consistency, reproducibility, and comparison of results across different neurofeedback sessions, practitioners, and research studies. The International 10-20 System, developed in 1958 by Herbert Jasper, provides a standardized method for electrode placement based on anatomical landmarks that ensure electrodes are positioned over consistent brain regions across individuals despite variations in head size and shape.

The system derives its name from the spacing intervals used—electrodes are placed at 10% or 20% intervals along lines measured between specific skull landmarks. The primary landmarks are the nasion (bridge of the nose), inion (bony prominence at the back of the skull), and left and right preauricular points (depressions just in front of the ears). From these fixed reference points, measurements are taken to calculate electrode positions that correspond to underlying brain regions.

Table 2.1: Key Electrode Positions in the 10-20 System

Position	Location	Underlying Brain Region	Common Neurofeedback Uses
Fp1, Fp2	Prefrontal	Orbitofrontal cortex	Executive function, emotional regulation
F3, F4	Frontal	Dorsolateral prefrontal cortex	Working memory, depression (alpha asymmetry)
F7, F8	Lateral frontal	Inferior frontal cortex	Language, emotional processing
Fz	Midline frontal	Medial frontal cortex	Attention, error monitoring
C3, C4	Central	Sensorimotor cortex	SMR training, motor control, ADHD
Cz	Vertex (midline)	Supplementary motor area	General training, reference electrode
T3, T4 (T7, T8)	Temporal	Temporal cortex	Memory, language, emotion
P3, P4	Parietal	Posterior parietal cortex	Spatial attention, integration
Pz	Midline parietal	Precuneus	Self-referential processing, attention
O1, O2	Occipital	Primary visual cortex	Alpha training, visual processing

The naming convention combines a letter indicating the region (Fp=frontopolar, F=frontal, C=central, T=temporal, P=parietal, O=occipital) with a number or letter indicating lateralization (odd numbers for left hemisphere, even for right, 'z' for midline positions). This systematic nomenclature enables clear communication about electrode positions across different systems and protocols.

Modern high-density EEG systems extend beyond the 10-20 system using the 10-10 or 10-5 systems, which place electrodes at finer intervals to achieve higher spatial resolution. Research-grade systems may employ 64, 128, or even 256 electrodes to capture detailed spatial patterns of brain activity. However, most clinical neurofeedback applications use between 1 and 4 channels, focusing on specific positions relevant to the targeted condition rather than comprehensive brain mapping.

EEG Signal Acquisition and Amplification

Acquiring usable EEG signals presents significant technical challenges due to their extremely small amplitude (typically 10-100 microvolts) compared to environmental electrical noise (often millivolts or more from power lines, muscle activity, and electronic devices). Specialized EEG amplifiers must boost these tiny signals by factors of 10,000 to 100,000 while maintaining signal integrity and rejecting noise.

EEG amplifiers employ differential recording—measuring the voltage difference between two electrodes rather than the absolute voltage at a single electrode. This approach helps cancel common-mode noise (electrical interference present equally at both electrodes) while preserving the brain signal of interest. Each differential amplifier channel connects to an active electrode (recording site), a reference electrode, and a ground electrode. The reference provides a baseline against which the active electrode is compared, while the ground helps reduce common-mode interference.

Reference electrode selection significantly influences the recorded signal characteristics. Common reference configurations include:

• Linked ears or mastoids: Reference electrodes on both earlobes or mastoid processes, electrically linked. Provides a relatively neutral reference with low brain activity, though not completely inactive.
• Cz reference: Vertex position (top center of head) used as reference. Convenient for single-channel recordings but introduces bias from the reference location's own activity.
• Average reference: The mean of all electrode signals serves as the reference. Provides a balanced reference in multi-channel systems but requires simultaneous recording from multiple sites.
• Laplacian derivation: Each electrode is referenced to the average of surrounding electrodes, emphasizing local activity and reducing volume conduction from distant sources.

Electrode impedance—the resistance to electrical current flow at the electrode-skin interface—critically affects signal quality. High impedance increases noise susceptibility and signal attenuation. The WIA-MENTAL-009 standard specifies maximum impedance thresholds of 5 kΩ (5,000 ohms) for clinical-grade neurofeedback, with lower values (under 2 kΩ) preferred for research applications. Achieving low impedance requires proper skin preparation: cleaning with alcohol to remove oils, gentle abrasion to remove dead skin cells, and application of conductive gel or paste to ensure good electrical contact.

{
  "eeg_acquisition_parameters": {
    "WIA_MENTAL_009_requirements": {
      "sampling_rate": {
        "minimum": 250,
        "recommended": 500,
        "unit": "Hz",
        "rationale": "Nyquist theorem requires >2x highest frequency (125 Hz for gamma)"
      },
      "resolution": {
        "minimum": 16,
        "recommended": 24,
        "unit": "bits",
        "dynamic_range": "96 dB @ 16-bit, 144 dB @ 24-bit"
      },
      "electrode_impedance": {
        "maximum": 5000,
        "preferred": 2000,
        "unit": "ohms",
        "measurement_frequency": "10 Hz"
      },
      "common_mode_rejection_ratio": {
        "minimum": 90,
        "preferred": 110,
        "unit": "dB",
        "purpose": "Rejection of environmental noise"
      },
      "input_noise": {
        "maximum": 1.0,
        "unit": "microvolts RMS",
        "bandwidth": "0.5-100 Hz"
      },
      "amplifier_gain": {
        "typical": 24000,
        "range": "10000-100000",
        "adjustable": true
      },
      "frequency_response": {
        "lowpass_filter": {
          "cutoff": 100,
          "rolloff": 24,
          "unit_cutoff": "Hz",
          "unit_rolloff": "dB/octave"
        },
        "highpass_filter": {
          "cutoff": 0.5,
          "rolloff": 12,
          "unit_cutoff": "Hz",
          "unit_rolloff": "dB/octave",
          "purpose": "Remove DC drift and slow artifacts"
        },
        "notch_filter": {
          "frequency": 60,
          "alternative": 50,
          "bandwidth": 2,
          "unit": "Hz",
          "purpose": "Remove power line interference",
          "note": "Use 50 Hz in regions with 50 Hz power"
        }
      }
    }
  }
}
        

Frequency Domain Analysis and Spectral Decomposition

While raw EEG waveforms provide temporal information about brain activity, most neurofeedback applications require decomposition of signals into their frequency components to target specific brain wave bands. The Fourier transform—specifically the Fast Fourier Transform (FFT) algorithm—converts time-domain signals into frequency-domain representations, revealing the amplitude and power of oscillations at different frequencies.

The FFT operates on short time windows (epochs) of EEG data, typically 1-4 seconds in duration. Each epoch is transformed from a sequence of amplitude values over time into a power spectrum showing how much energy is present at each frequency. The resulting spectrum displays power (usually in microvolts squared or decibels) as a function of frequency, with peaks at frequencies where strong rhythmic activity occurs.

Spectral resolution—the ability to distinguish nearby frequencies—depends on epoch length. Longer epochs provide better frequency resolution but poorer temporal resolution, creating a fundamental trade-off. For neurofeedback applications requiring rapid feedback, epoch lengths of 0.5-2 seconds represent a practical compromise, providing adequate frequency resolution (0.5-2 Hz bins) while maintaining responsiveness to changes in brain state.

Table 2.2: Spectral Analysis Parameters for Neurofeedback

Parameter	Typical Range	Impact on Analysis	WIA-MENTAL-009 Recommendation
Epoch Length	0.5-4 seconds	Longer = better frequency resolution, worse temporal	1-2 seconds for real-time feedback
Epoch Overlap	0-75%	Smoother updates, higher computation	50% overlap recommended
Windowing Function	Hanning, Hamming, Blackman	Reduces spectral leakage, affects resolution	Hanning window standard
Frequency Resolution	0.25-2 Hz	Ability to distinguish nearby frequencies	0.5-1 Hz typical for clinical work
Power Calculation	Absolute, relative, ratio	Normalization affects interpretation	Absolute power for most protocols
Update Rate	2-10 Hz	Feedback smoothness vs. latency	4-8 Hz update rate optimal

Window functions (mathematical functions applied to each epoch before FFT) reduce spectral leakage—artificial spreading of power across adjacent frequencies caused by analyzing finite-duration signals. The Hanning window, which gradually tapers signal amplitude to zero at epoch boundaries, is most commonly used in neurofeedback applications. Without windowing, abrupt transitions at epoch boundaries introduce high-frequency artifacts that contaminate the spectrum.

Alternative time-frequency analysis methods offer advantages over standard FFT for certain applications. Wavelet decomposition provides better temporal resolution for high frequencies while maintaining good frequency resolution for low frequencies, matching the natural time-frequency characteristics of brain signals. However, wavelets require more computational resources and are less commonly implemented in real-time neurofeedback systems compared to FFT-based approaches.

Artifact Detection and Removal

EEG recordings are contaminated by numerous sources of non-brain electrical activity collectively termed artifacts. Effective neurofeedback requires robust artifact detection and handling to prevent rewarding or training artifact patterns rather than genuine brain activity. Common artifact sources include eye movements and blinks, muscle tension, movement, electrode problems, and environmental electromagnetic interference.

Ocular artifacts from eye movements and blinks produce large amplitude signals (50-200 microvolts) that propagate across frontal electrode sites. Blinks generate characteristic sharp transients, while lateral eye movements produce slower deflections with opposite polarity at left versus right frontal sites. Since ocular artifacts can be larger than genuine EEG signals, they can dominate recorded activity and trigger false rewards if not properly handled.

Muscle artifacts (EMG - electromyography) result from facial muscle tension, jaw clenching, or neck muscle activity. These artifacts appear as high-frequency noise (primarily above 30 Hz) that increases broad-spectrum power. Muscle artifacts particularly affect temporal electrode sites adjacent to jaw muscles. Since muscle tension often increases during concentration or stress, muscle artifacts can inadvertently be rewarded in neurofeedback protocols targeting beta frequencies if not carefully monitored.

Automated artifact detection algorithms analyze multiple features to identify contaminated data segments. Amplitude thresholding flags epochs where any sample exceeds maximum acceptable values. Gradient thresholding detects sharp transitions characteristic of eye blinks or movement artifacts. Frequency domain analysis identifies muscle artifacts by detecting power increases in high-frequency ranges (typically 30-50 Hz) that exceed ratios relative to lower frequencies. Multi-channel systems can employ spatial filters or independent component analysis (ICA) to separate artifact sources from brain signals.

class ArtifactDetector:
    """
    Real-time artifact detection for neurofeedback
    Implements WIA-MENTAL-009 standard detection methods
    """
    
    def __init__(self, config):
        self.amplitude_threshold = config.get('amplitude_max', 100)  # microvolts
        self.gradient_threshold = config.get('gradient_max', 50)     # uV/sample
        self.emg_ratio_threshold = config.get('emg_ratio', 3.0)      # ratio
        self.sampling_rate = config['sampling_rate']
    
    def detect_artifacts(self, eeg_epoch, spectrum):
        """
        Analyzes EEG epoch for artifacts
        Returns: (is_clean, artifact_types)
        """
        artifacts = []
        
        # Amplitude threshold detection
        if max(abs(eeg_epoch)) > self.amplitude_threshold:
            artifacts.append('amplitude_exceeded')
        
        # Gradient/slope detection (blinks, movement)
        gradient = np.diff(eeg_epoch)
        if max(abs(gradient)) > self.gradient_threshold:
            artifacts.append('sharp_transient')
        
        # EMG contamination detection
        low_freq_power = np.mean(spectrum[1:8])   # 1-8 Hz
        high_freq_power = np.mean(spectrum[30:50]) # 30-50 Hz
        
        if high_freq_power > (self.emg_ratio_threshold * low_freq_power):
            artifacts.append('muscle_artifact')
        
        # Eye movement detection (requires multi-channel)
        # Omitted for brevity - would analyze Fp1-Fp2 differences
        
        is_clean = len(artifacts) == 0
        return is_clean, artifacts
    
    def handle_artifact_period(self, feedback_system):
        """
        Suspend feedback during artifacts per WIA-MENTAL-009 protocol
        """
        feedback_system.suspend_reward()
        feedback_system.display_artifact_warning()
        # Do not include epoch in session statistics or training
        
    def calculate_artifact_percentage(self, session_data):
        """
        Computes percentage of training time affected by artifacts
        High artifact rates may indicate need for relaxation training
        """
        total_epochs = len(session_data.epochs)
        clean_epochs = sum([e.is_clean for e in session_data.epochs])
        artifact_percent = 100 * (1 - clean_epochs/total_epochs)
        
        return {
            'total_epochs': total_epochs,
            'clean_epochs': clean_epochs,
            'artifact_percentage': artifact_percent,
            'interpretation': self.interpret_artifact_rate(artifact_percent)
        }
    
    def interpret_artifact_rate(self, percent):
        """Provides clinical interpretation of artifact rates"""
        if percent < 10:
            return "Excellent - minimal artifacts"
        elif percent < 25:
            return "Good - acceptable artifact levels"
        elif percent < 40:
            return "Fair - consider relaxation techniques"
        else:
            return "Poor - artifacts compromising training, address tension/movement"
        

Quantitative EEG (QEEG) Assessment

Quantitative EEG (QEEG), also called brain mapping, involves comprehensive analysis of multi-channel EEG recordings to create statistical representations of an individual's brain activity patterns. QEEG assessment typically involves recording 19 or more channels using the full 10-20 electrode array during eyes-open and eyes-closed resting conditions, and sometimes during cognitive tasks. The recorded data is then processed to extract numerous quantitative features and compared to normative databases containing thousands of recordings from healthy individuals.

QEEG analysis generates various maps and metrics including absolute power (magnitude of activity in each frequency band at each location), relative power (percentage of total power in each band), coherence (correlation between activity at different electrode pairs), phase relationships (timing differences between oscillations), and asymmetry measures (differences between homologous left and right hemisphere sites). These metrics are standardized against age-matched normative data to identify statistically significant deviations.

In clinical neurofeedback practice, QEEG assessment serves multiple purposes. It identifies specific dysregulation patterns associated with clinical symptoms, guides selection of appropriate training protocols and electrode sites, provides objective baseline measurements for outcome tracking, and helps rule out underlying neurological conditions requiring medical evaluation. However, QEEG interpretation requires specialized training, and overreliance on automated reports without clinical context can lead to inappropriate protocol selection.

Signal Processing Pipeline

A complete neurofeedback system implements a multi-stage signal processing pipeline that transforms raw EEG voltage readings into meaningful metrics for feedback. Understanding this pipeline is essential for optimizing system performance, troubleshooting technical issues, and ensuring compliance with WIA-MENTAL-009 data processing standards.

Stage 1 - Acquisition and Digitization: Analog EEG signals from amplifiers undergo analog-to-digital conversion at the specified sampling rate (typically 250-1000 Hz). The raw digital signal is stored in a circular buffer that maintains the most recent several seconds of data for processing.

Stage 2 - Preprocessing and Filtering: High-pass filtering (typically 0.5-1 Hz cutoff) removes DC drift and very slow artifacts. Low-pass filtering (typically 50-100 Hz cutoff) attenuates high-frequency noise and muscle artifacts. Notch filtering at power line frequency (50 or 60 Hz) removes electromagnetic interference. These filters can be implemented in hardware (analog filtering before digitization) or software (digital filtering), with digital filtering now standard in modern systems.

Stage 3 - Artifact Detection: Real-time algorithms analyze amplitude, gradients, and spectral characteristics to identify contaminated epochs as described in the previous section. Clean epochs proceed to further processing, while artifact-contaminated epochs are flagged and excluded from feedback calculations.

Stage 4 - Spectral Decomposition: Fast Fourier Transform or wavelet decomposition converts time-domain signals into frequency-domain representations. This yields power values for each frequency bin, which are then integrated across target frequency bands (delta, theta, alpha, SMR, beta, gamma) according to the specific protocol.

Stage 5 - Metric Calculation: Protocol-specific calculations derive the training metrics. Simple amplitude/power protocols calculate the power in target bands. Ratio protocols compute ratios like theta/beta for ADHD. Coherence protocols calculate correlation between electrode pairs. Z-score protocols compare current metrics to baseline statistics.

Stage 6 - Reward Logic: The computed metrics are compared against threshold criteria to determine whether current brain activity warrants reward delivery. Thresholds may be fixed values or adaptive percentiles that adjust based on ongoing performance.

Stage 7 - Feedback Presentation: Visual, auditory, or combined feedback is updated to reflect current reward status. Games advance, tones play, or visual displays animate when reward criteria are met.

Stage 8 - Data Logging: All raw data, processed metrics, artifact annotations, reward events, and session metadata are recorded according to WIA-MENTAL-009 data format specifications for outcome tracking, quality assurance, and research purposes.

The entire pipeline must execute within strict latency constraints. From the time a brain signal occurs to when feedback reflects that signal should not exceed 250 milliseconds total system latency. This requires optimized algorithms, sufficient computational resources, and careful system architecture to minimize processing delays.