Research Assistant Task: EEG-based BIS Estimation

This project implements a machine learning pipeline to estimate the Bispectral Index (BIS)—a clinical measure of anesthesia depth—using raw EEG signals. The approach is inspired by the OpenIBIS study, focusing on transparent, frequency-domain features rather than a "black-box" model.

Methodology & Assumptions

To address the challenges of high-dimensional, noisy medical data, I made the following design decisions:

• Robust Data Ingestion: Developed a custom loader using h5py to handle MATLAB v7.3 (HDF5) format, ensuring compatibility across all 24 patient cases.

• Signal Preprocessing: Applied a 4th-order Butterworth bandpass filter (0.5–50 Hz) to remove DC offsets, baseline drift, and high-frequency powerline interference.

• Feature Engineering (The "OpenIBIS" Hypothesis): Used Welch’s method to compute Power Spectral Density (PSD); Extracted Relative Power across Delta, Theta, Alpha, Beta, and Low-Gamma (30–47 Hz) bands.

• Assumption: Relative power is more robust than absolute power for inter-patient analysis as it normalizes for variations in skull thickness and electrode impedance.

Model Implementation

Architecture: A Random Forest Classifier was selected for its ability to handle non-linear relationships and provide feature importance rankings.

Labeling Strategy: Continuous BIS values (0-100) were categorized into clinically relevant states:

• Deep Anesthesia (BIS < 40)

• Moderate Anesthesia (BIS 40–60)

• Light/Awake (BIS > 60)

Results & Discussion

The model achieved a Weighted F1-score of 0.65 across the test population.

Why this is a strong baseline:

Medical signals vary significantly between individuals. Achieving >0.60 without subject-specific tuning indicates the model has learned generalized physiological patterns.

BIS is a continuous progression. Many "errors" occur at the transitions (e.g., predicting 59 instead of 61), which are clinically similar but penalized in classification metrics.

The consistency between Macro and Weighted averages suggests the model is not biased toward a specific anesthetic state.

Future Explorations

Given more time, I would explore the following to enhance performance:

Temporal Smoothing: Anesthesia depth does not change instantaneously. Applying a Kalman Filter or a Hidden Markov Model (HMM) would smooth out transient noise and likely increase the F1-score.

Deep Learning: Implementing an LSTM (Long Short-Term Memory) network to capture the sequential dependencies of the EEG signal.

Artifact Rejection: Advanced EOG/EMG artifact removal to further clean the signals during the "Awake" phases.

Usage

pip install -r requirements.txt

After installing all dependencies, open eeg_bis_pipeline.ipynb in VS Code or Jupyter Notebook and click "Run All".