IRR-Aware EEG Modeling

This repository contains code and analyses supporting the study "Modeling Rater Variability to Improve the Reliability of Visual EEG Grading in Encephalopathy."

We quantify inter-rater reliability (IRR) of visually assessed EEG features and incorporate empirically measured rater disagreement into the training of an interpretable EEG-based encephalopathy severity score.

Overview

• Feature-level inter-rater reliability analysis across EEG patterns
• IRR-aware model training via probabilistic feature perturbation
• Evaluation under simulated rater variability
• Comparison of original and IRR-adjusted EEG severity models

Key scripts

The main analyses in the manuscript are reproduced by the following scripts:

• irr_aware_modeling_vecams.py
  Implements the IRR-aware VE-CAM-S modeling framework, including Monte Carlo simulation of rater disagreement and training of the rater-variability-adjusted model.
• generate_modeling_figures.py
  Generates the figures used in the manuscript from precomputed results (model comparison, performance metrics, and robustness analyses).

Requirements

Python >= 3.9 Dependencies are specified in environment.yml.

Citation

If you use this code, please cite the accompanying manuscript.