OCDR — Optic Cup-to-Disc Ratio Monitor 🔬

Early Demo (BETA) [work in progress]

Automated tool for monitoring optic cup-to-disc ratio (CDR) changes over time, removing subjective unreliablity between professionals.

---

Model

Built on the DRISHTI-GS dataset

The optic disc and optic cup were trained as two separate segmentation models using an EfficientNetB4 U-Net architecture, allowing independent mask prediction before CDR is calculated from vertical diameter ratio.

---

Stack

• Backend — Flask REST API
• ML — Keras / TensorFlow (EfficientNetB4 U-Net)
• Frontend — Jinja2 + Alpine.js
• Database — SQLAlchemy / SQLite (PostgreSQL-ready)

Everything is in the backend/ folde for demo purposes.

---

Current Limitations

• Optimised for centralised ONH fundus photos only (Standard fundography images off-centre or will produce unreliable results)
• Models require further training and validation across diverse image types and noise management
• Demo dataset is limited — not yet validated for clinical use

---

Roadmap

• AWS deployment with image storage (S3 + RDS)
• Secure 2FA login
• Support for peripheral and wide-field fundus images
• Extended model training across broader datasets

---

Disclaimer

This is a research/portfolio project. Not validated for clinical decision-making.