Skip to content

Foundation PLR Documentation

Foundation Models for Pupillary Light Reflex Analysis

Evaluating how preprocessing choices affect downstream prediction quality in glaucoma screening

What is Foundation PLR?

Foundation PLR is a comprehensive research framework that investigates whether generic time-series foundation models (MOMENT, UniTS, TimesNet, SAITS) can improve biosignal preprocessing compared to traditional methods in clinical applications.

Research Focus

This is NOT about comparing classifiers. The research question is:

How do preprocessing choices (outlier detection → imputation) affect ALL STRATOS-compliant downstream metrics when using handcrafted physiological features?

Key Findings

Finding Value Interpretation
Best AUROC 0.913 With ground truth preprocessing + CatBoost
Preprocessing effect η²=0.15 Preprocessing choice matters
Handcrafted vs Embeddings 0.830 vs 0.740 Embeddings underperform by 9pp
FM for preprocessing Competitive Foundation models useful for outlier/imputation

The Pipeline

graph LR
    A[Raw PLR Signal] --> B[Outlier Detection<br/>11 methods]
    B --> C[Imputation<br/>7 methods]
    C --> D[Featurization<br/>Handcrafted]
    D --> E[Classification<br/>CatBoost fixed]
    E --> F[STRATOS Metrics]

    style B fill:#e1f5fe
    style C fill:#e1f5fe
    style F fill:#fff3e0

  • Getting Started

    Install Foundation PLR and run your first experiment

    Installation

  • User Guide

    Understand the pipeline stages and configuration

    Pipeline Overview

  • API Reference

    Auto-generated documentation from source code

    API Docs

  • Tutorials

    Step-by-step guides for common tasks

    Tutorials

Data Provenance

Source: Najjar et al. 2023, Br J Ophthalmol (DOI: 10.1136/bjophthalmol-2021-319938)

Task N Subjects Description
Outlier Detection 507 All subjects with ground truth masks
Imputation 507 All subjects with denoised signals
Classification 208 152 control + 56 glaucoma with labels

Citation

If you use this code in your research, please cite:

@software{foundation_plr,
  title = {Foundation PLR: Foundation Models for Pupillary Light Reflex Analysis},
  year = {2026},
  url = {https://github.com/petteriTeikari/foundation_PLR}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.