Epochs and Event-Related Analysis

Epochs and Event-Related Analysis

Overview

Event-related analysis examines physiological responses time-locked to specific stimuli or events. NeuroKit2 provides tools for event detection, epoch creation, averaging, and event-related feature extraction across all signal types.

Event Detection

events_find()

Automatically detect events/triggers in a signal based on threshold crossings or changes.

events = nk.events_find(event_channel, threshold=0.5, threshold_keep='above',
                        duration_min=1, inter_min=0)

Parameters:

• threshold: Detection threshold value
• threshold_keep: 'above' or 'below' threshold
• duration_min: Minimum event duration (samples) to keep
• inter_min: Minimum interval between events (samples)

Returns:

• Dictionary with:
  • 'onset': Event onset indices
  • 'offset': Event offset indices (if applicable)
  • 'duration': Event durations
  • 'label': Event labels (if multiple event types)

Common use cases:

TTL triggers from experiments:

# Trigger channel: 0V baseline, 5V pulses during events
events = nk.events_find(trigger_channel, threshold=2.5, threshold_keep='above')

Button presses:

# Detect when button signal goes high
button_events = nk.events_find(button_signal, threshold=0.5, threshold_keep='above',
                               duration_min=10)  # Debounce

State changes:

# Detect periods above/below threshold
high_arousal = nk.events_find(eda_signal, threshold='auto', duration_min=100)

events_plot()

Visualize event timing relative to signals.

nk.events_plot(events, signal)

Displays:

• Signal trace
• Event markers (vertical lines or shaded regions)
• Event labels

Use case:

• Verify event detection accuracy
• Inspect temporal distribution of events
• Quality control before epoching

Epoch Creation

epochs_create()

Create epochs (segments) of data around events for event-related analysis.

epochs = nk.epochs_create(data, events, sampling_rate=1000,
                          epochs_start=-0.5, epochs_end=2.0,
                          event_labels=None, event_conditions=None,
                          baseline_correction=False)

Parameters:

• data: DataFrame with signals or single signal
• events: Event indices or dictionary from events_find()
• sampling_rate: Signal sampling rate (Hz)
• epochs_start: Start time relative to event (seconds, negative = before)
• epochs_end: End time relative to event (seconds, positive = after)
• event_labels: List of labels for each event (optional)
• event_conditions: List of condition names for each event (optional)
• baseline_correction: If True, subtract baseline mean from each epoch

Returns:

• Dictionary of DataFrames, one per epoch
• Each DataFrame contains signal data with time relative to event (Index=0 at event onset)
• Includes 'Label' and 'Condition' columns if provided

Typical epoch windows:

• Visual ERP: -0.2 to 1.0 seconds (200 ms baseline, 1 s post-stimulus)
• Cardiac orienting: -1.0 to 10 seconds (capture anticipation and response)
• EMG startle: -0.1 to 0.5 seconds (brief response)
• EDA SCR: -1.0 to 10 seconds (1-3 s latency, slow recovery)

Event Labels and Conditions

Organize events by type and experimental conditions:

# Example: Emotional picture experiment
event_times = [1000, 2500, 4200, 5800]  # Event onsets in samples
event_labels = ['trial1', 'trial2', 'trial3', 'trial4']
event_conditions = ['positive', 'negative', 'positive', 'neutral']

epochs = nk.epochs_create(signals, events=event_times, sampling_rate=1000,
                          epochs_start=-1, epochs_end=5,
                          event_labels=event_labels,
                          event_conditions=event_conditions)

Access epochs:

# Epoch by number
epoch_1 = epochs['1']

# Filter by condition
positive_epochs = {k: v for k, v in epochs.items() if v['Condition'][0] == 'positive'}

Baseline Correction

Remove pre-stimulus baseline from epochs to isolate event-related changes:

Automatic (during epoch creation):

epochs = nk.epochs_create(data, events, sampling_rate=1000,
                          epochs_start=-0.5, epochs_end=2.0,
                          baseline_correction=True)  # Subtracts mean of entire baseline

Manual (after epoch creation):

# Subtract baseline period mean
baseline_start = -0.5  # seconds
baseline_end = 0.0     # seconds

for key, epoch in epochs.items():
    baseline_mask = (epoch.index >= baseline_start) & (epoch.index < baseline_end)
    baseline_mean = epoch[baseline_mask].mean()
    epochs[key] = epoch - baseline_mean

When to baseline correct:

• ERPs: Always (isolates event-related change)
• Cardiac/EDA: Usually (removes inter-individual baseline differences)
• Absolute measures: Sometimes not desired (e.g., analyzing absolute amplitude)

Epoch Analysis and Visualization

epochs_plot()

Visualize individual or averaged epochs.

nk.epochs_plot(epochs, column='ECG_Rate', condition=None, show=True)

Parameters:

• column: Which signal column to plot
• condition: Plot only specific condition (optional)

Displays:

• Individual epoch traces (semi-transparent)
• Average across epochs (bold line)
• Optional: Shaded error (SEM or SD)

Use cases:

• Visualize event-related responses
• Compare conditions
• Identify outlier epochs

epochs_average()

Compute grand average across epochs with statistics.

average_epochs = nk.epochs_average(epochs, output='dict')

Parameters:

• output: 'dict' (default) or 'df' (DataFrame)

Returns:

• Dictionary or DataFrame with:
  • 'Mean': Average across epochs at each time point
  • 'SD': Standard deviation
  • 'SE': Standard error of mean
  • 'CI_lower', 'CI_upper': 95% confidence intervals

Use case:

• Compute event-related potentials (ERPs)
• Grand average cardiac/EDA/EMG responses
• Group-level analysis

Condition-specific averaging:

# Separate averages by condition
positive_epochs = {k: v for k, v in epochs.items() if v['Condition'][0] == 'positive'}
negative_epochs = {k: v for k, v in epochs.items() if v['Condition'][0] == 'negative'}

avg_positive = nk.epochs_average(positive_epochs)
avg_negative = nk.epochs_average(negative_epochs)

epochs_to_df()

Convert epochs dictionary to unified DataFrame.

epochs_df = nk.epochs_to_df(epochs)

Returns:

• Single DataFrame with all epochs stacked
• Includes 'Epoch', 'Time', 'Label', 'Condition' columns
• Facilitates statistical analysis and plotting with pandas/seaborn

Use case:

• Prepare data for mixed-effects models
• Plotting with seaborn/plotly
• Export to R or statistical software

epochs_to_array()

Convert epochs to 3D NumPy array.

epochs_array = nk.epochs_to_array(epochs, column='ECG_Rate')

Returns:

• 3D array: (n_epochs, n_timepoints, n_columns)

Use case:

• Machine learning input (epoched features)
• Custom array-based analysis
• Statistical tests on array data

Signal-Specific Event-Related Analysis

NeuroKit2 provides specialized event-related analysis for each signal type:

ECG Event-Related

ecg_epochs = nk.epochs_create(ecg_signals, events, sampling_rate=1000,
                              epochs_start=-1, epochs_end=10)
ecg_results = nk.ecg_eventrelated(ecg_epochs)

Computed metrics:

• ECG_Rate_Baseline: Heart rate before event
• ECG_Rate_Min/Max: Minimum/maximum rate during epoch
• ECG_Phase_*: Cardiac phase at event onset
• Rate dynamics across time windows

EDA Event-Related

eda_epochs = nk.epochs_create(eda_signals, events, sampling_rate=100,
                              epochs_start=-1, epochs_end=10)
eda_results = nk.eda_eventrelated(eda_epochs)

Computed metrics:

• EDA_SCR: Presence of SCR (binary)
• SCR_Amplitude: Maximum SCR amplitude
• SCR_Latency: Time to SCR onset
• SCR_RiseTime, SCR_RecoveryTime
• EDA_Tonic: Mean tonic level

RSP Event-Related

rsp_epochs = nk.epochs_create(rsp_signals, events, sampling_rate=100,
                              epochs_start=-0.5, epochs_end=5)
rsp_results = nk.rsp_eventrelated(rsp_epochs)

Computed metrics:

• RSP_Rate_Mean: Average breathing rate
• RSP_Amplitude_Mean: Average breath depth
• RSP_Phase: Respiratory phase at event
• Rate/amplitude dynamics

EMG Event-Related

emg_epochs = nk.epochs_create(emg_signals, events, sampling_rate=1000,
                              epochs_start=-0.1, epochs_end=1.0)
emg_results = nk.emg_eventrelated(emg_epochs)

Computed metrics:

• EMG_Activation: Presence of activation
• EMG_Amplitude_Mean/Max: Amplitude statistics
• EMG_Onset_Latency: Time to activation onset
• EMG_Bursts: Number of activation bursts

EOG Event-Related

eog_epochs = nk.epochs_create(eog_signals, events, sampling_rate=500,
                              epochs_start=-0.5, epochs_end=2.0)
eog_results = nk.eog_eventrelated(eog_epochs)

Computed metrics:

• EOG_Blinks_N: Number of blinks during epoch
• EOG_Rate_Mean: Blink rate
• Temporal blink distribution

PPG Event-Related

ppg_epochs = nk.epochs_create(ppg_signals, events, sampling_rate=100,
                              epochs_start=-1, epochs_end=10)
ppg_results = nk.ppg_eventrelated(ppg_epochs)

Computed metrics:

• Similar to ECG: rate dynamics, phase information

Practical Workflows

Complete Event-Related Analysis Pipeline

import neurokit2 as nk

# 1. Process physiological signals
ecg_signals, ecg_info = nk.ecg_process(ecg, sampling_rate=1000)
eda_signals, eda_info = nk.eda_process(eda, sampling_rate=100)

# 2. Align sampling rates if needed
eda_signals_resampled = nk.signal_resample(eda_signals, sampling_rate=100,
                                           desired_sampling_rate=1000)

# 3. Merge signals into single DataFrame
signals = pd.concat([ecg_signals, eda_signals_resampled], axis=1)

# 4. Detect events
events = nk.events_find(trigger_channel, threshold=0.5)

# 5. Add event labels and conditions
event_labels = ['trial1', 'trial2', 'trial3', ...]
event_conditions = ['condition_A', 'condition_B', 'condition_A', ...]

# 6. Create epochs
epochs = nk.epochs_create(signals, events, sampling_rate=1000,
                          epochs_start=-1.0, epochs_end=5.0,
                          event_labels=event_labels,
                          event_conditions=event_conditions,
                          baseline_correction=True)

# 7. Signal-specific event-related analysis
ecg_results = nk.ecg_eventrelated(epochs)
eda_results = nk.eda_eventrelated(epochs)

# 8. Merge results
results = pd.merge(ecg_results, eda_results, left_index=True, right_index=True)

# 9. Statistical analysis by condition
results['Condition'] = event_conditions
condition_comparison = results.groupby('Condition').mean()

Handling Multiple Event Types

# Different event types with different markers
event_type1 = nk.events_find(trigger_ch1, threshold=0.5)
event_type2 = nk.events_find(trigger_ch2, threshold=0.5)

# Combine events with labels
all_events = np.concatenate([event_type1['onset'], event_type2['onset']])
event_labels = ['type1'] * len(event_type1['onset']) + ['type2'] * len(event_type2['onset'])

# Sort by time
sort_idx = np.argsort(all_events)
all_events = all_events[sort_idx]
event_labels = [event_labels[i] for i in sort_idx]

# Create epochs
epochs = nk.epochs_create(signals, all_events, sampling_rate=1000,
                          epochs_start=-0.5, epochs_end=3.0,
                          event_labels=event_labels)

# Separate by type
type1_epochs = {k: v for k, v in epochs.items() if v['Label'][0] == 'type1'}
type2_epochs = {k: v for k, v in epochs.items() if v['Label'][0] == 'type2'}

Quality Control and Artifact Rejection

# Remove epochs with excessive noise or artifacts
clean_epochs = {}
for key, epoch in epochs.items():
    # Example: reject if EDA amplitude too high (movement artifact)
    if epoch['EDA_Phasic'].abs().max() < 5.0:  # Threshold
        # Example: reject if heart rate change too large (invalid)
        if epoch['ECG_Rate'].max() - epoch['ECG_Rate'].min() < 50:
            clean_epochs[key] = epoch

print(f"Kept {len(clean_epochs)}/{len(epochs)} epochs")

# Analyze clean epochs
results = nk.ecg_eventrelated(clean_epochs)

Statistical Considerations

Sample Size

• ERP/averaging: 20-30+ trials per condition minimum
• Individual trial analysis: Mixed-effects models handle variable trial counts
• Group comparisons: Pilot data for power analysis

Time Window Selection

• A priori hypotheses: Pre-register time windows based on literature
• Exploratory: Use full epoch, correct for multiple comparisons
• Avoid: Selecting windows based on observed data (circular)

Baseline Period

• Should be free of anticipatory effects
• Sufficient duration for stable estimate (500-1000 ms typical)
• Shorter for fast dynamics (e.g., startle: 100 ms sufficient)

Condition Comparison

• Repeated measures ANOVA for within-subject designs
• Mixed-effects models for unbalanced data
• Permutation tests for non-parametric comparisons
• Correct for multiple comparisons (time points/signals)

Common Applications

Cognitive psychology:

• P300 ERP analysis
• Error-related negativity (ERN)
• Attentional blink
• Working memory load effects

Affective neuroscience:

• Emotional picture viewing (EDA, HR, facial EMG)
• Fear conditioning (HR deceleration, SCR)
• Valence/arousal dimensions

Clinical research:

• Startle response (orbicularis oculi EMG)
• Orienting response (HR deceleration)
• Anticipation and prediction error

Psychophysiology:

• Cardiac defense response
• Orienting vs. defensive reflexes
• Respiratory changes during emotion

Human-computer interaction:

• User engagement during events
• Surprise/violation of expectation
• Cognitive load during task events

References

• Luck, S. J. (2014). An introduction to the event-related potential technique (2nd ed.). MIT press.
• Bradley, M. M., & Lang, P. J. (2000). Measuring emotion: Behavior, feeling, and physiology. In R. D. Lane & L. Nadel (Eds.), Cognitive neuroscience of emotion (pp. 242-276). Oxford University Press.
• Boucsein, W. (2012). Electrodermal activity (2nd ed.). Springer.
• Gratton, G., Coles, M. G., & Donchin, E. (1983). A new method for off-line removal of ocular artifact. Electroencephalography and clinical neurophysiology, 55(4), 468-484.