Bleachcarte/JaundiceVSHealthy

JaundiceDetectorV0 API

Overview

This is a lightweight Python module entrypoint designed to assess the likelihood of jaundice (scleral icterus) from smartphone images of human eyes.

Instead of full containerization, this service is deployed as a standalone FastAPI web server that wraps the raw PyTorch weights. It includes built-in pre-inference quality gates that analyze images for blur and extreme lighting conditions before running the model, perfectly aligning with the client application's quality-check flow.

Model Details

• Base Architecture: EfficientNet-B0 (PyTorch state_dict)

Limitations & Disclaimer

Not a Diagnostic Device: This tool is intended as a screening and support feature only. It is not a bilirubin estimator and has not been cleared for clinical diagnosis.

Heuristic Limitations: The image quality checks rely on mathematical heuristics (like variance and RMS) which may not capture complex obstructions like closed eyelids or severe flash glare.

Author

Daniel Ekong — Daniel's Git