EEG Analysis with MNE Sample Dataset

Level: Hard
Goal: Use EEG time series (one or a few channels) for analysis and simple modeling.

Library & Data

• MNE Python: https://mne.tools/stable/
• MNE Datasets: https://mne.tools/stable/documentation/datasets.html

Installation

pip install mne

Data Loading

import mne
import pandas as pd
import numpy as np

data_path = mne.datasets.sample.data_path()
raw_fname = data_path + "/MEG/sample/sample_audvis_raw.fif"

raw = mne.io.read_raw_fif(raw_fname, preload=True)
raw.pick_types(meg=False, eeg=True)

data, times = raw[:1, :]  # first EEG channel
sr = raw.info["sfreq"]
index = pd.to_datetime(times, unit="s")
eeg_series = pd.Series(np.ravel(data), index=index)

Implementation Steps

1. Library Setup and Data Loading

• Install MNE library
• Download sample dataset (automatic on first use)
• Load EEG data
• Understand MNE data structure

2. Data Exploration

• Inspect available EEG channels
• Select 1-3 channels for analysis
• Examine sampling rate (typically 100-1000 Hz)
• Plot raw EEG signals
• Understand signal characteristics

3. Data Preprocessing

• Filtering:
  • Apply band-pass filter (e.g., 1-40 Hz for EEG)
  • Remove line noise (50/60 Hz) if present
• Artifact Removal:
  • Identify and remove artifacts (eye blinks, muscle activity)
  • Use MNE's artifact detection methods
• Downsampling (optional):
  • May need to downsample for time series analysis
  • Preserve frequency content of interest

4. Feature Extraction

• Frequency Domain:
  • Calculate power spectral density (PSD)
  • Analyze frequency bands (delta, theta, alpha, beta, gamma)
  • Extract band power over time
• Time Domain:
  • Extract amplitude features
  • Calculate statistical features (mean, std, etc.)
• Create Lower-Frequency Series:
  • Extract features per time window (e.g., alpha power per second)
  • Create time series of extracted features

5. Exploratory Data Analysis (EDA)

• Plot raw and filtered signals
• Visualize frequency content (spectrogram, PSD)
• Analyze temporal patterns
• Examine stationarity of extracted features

6. Model Building

• Feature Time Series:
  • Apply ARIMA/SARIMA to extracted features (e.g., alpha power)
  • Model lower-frequency derived series
• Raw Signal (advanced):
  • May need specialized methods for high-frequency signals
  • Consider state-space models or specialized EEG analysis

7. Model Evaluation

• Split data temporally
• Generate forecasts for features
• Calculate accuracy metrics
• Visualize results
• Compare with neurophysiological expectations

8. Neurophysiological Interpretation

• Interpret results in neuroscience context
• Relate to known EEG patterns (alpha, beta waves)
• Discuss implications for signal analysis
• Compare with EEG literature

Expected Deliverables

1. EDA Report:

   • Raw and filtered EEG plots
   • Frequency analysis (PSD, spectrograms)
   • Feature extraction results
   • Temporal pattern analysis

2. Model Results:

   • Selected models for feature time series
   • Forecast accuracy
   • Visualization of results
   • Neurophysiological interpretation

3. Code:

   • Complete Python notebook
   • MNE data loading and preprocessing functions
   • Feature extraction utilities
   • Visualization tools

Tips

• EEG data is high-frequency and requires specialized preprocessing
• Filtering is essential (remove artifacts, focus on frequency bands of interest)
• Consider extracting features rather than modeling raw signals directly
• Frequency domain analysis is often more informative than time domain
• Use MNE's built-in functions for preprocessing and analysis
• Understand neurophysiological context (different frequency bands have different meanings)
• Downsampling may be necessary for standard time series methods
• Document all preprocessing steps (filtering, artifact removal)
• Consider event-related analysis in addition to continuous forecasting
• Consult neuroscience literature for appropriate analysis methods