BARAKODE
Contact Us
All Case Studies
Computer Vision

Glaucoma Detection in Retinal Fundus Images

U-Net segmentation of optic cup & disc with automated CDR diagnosis.

2025
View full PDFDownload PDF
OVERVIEW

The Project

Glaucoma is the second-largest cause of blindness worldwide and needs early diagnosis. This project segments the optic cup and optic disc from retinal fundus images using a U-Net model, then derives a Cup-to-Disc Ratio (CDR) to flag likelihood of glaucoma — a system that works without excessive equipment or specialist time.

Objectives

  • Segment optic cup and optic disc from fundus images.
  • Compute the Cup-to-Disc Ratio for diagnosis.
  • Evaluate with accuracy, precision, recall and IoU.

    • Tools & Technologies

    PythonU-Net (encoder/decoder)OpenCVTensorFlow / Keras
    FIGURES

    Charts & Graphs

    U-Net architecture for optic cup & disc segmentation
    U-Net architecture for optic cup & disc segmentation
    Predicted vs original cup/disc masks
    Predicted vs original cup/disc masks
    Segmentation metrics with predicted masks
    Segmentation metrics with predicted masks
    Predicted vs original masks (sample 2)
    Predicted vs original masks (sample 2)
    Predicted vs original masks (sample 3)
    Predicted vs original masks (sample 3)
    Predicted vs original masks (sample 4)
    Predicted vs original masks (sample 4)
    METHODOLOGY

    The Approach

    1

    Preprocess: convert images to a uniform (256×256×1) grayscale format; build image, optic-cup and optic-disc lists by thresholding masks.

    2

    Build a U-Net from conv/encoder/decoder blocks (conv layers + BatchNorm + ReLU, max-pool down, transpose-conv up with skip connections).

    3

    Train on the dataset (batch size 2, 25–30 epochs).

    4

    Predict masks, compute CDR; CDR > 0.4 → high likelihood of glaucoma.

    OUTCOME

    Results & Learnings

  • High segmentation accuracy (optic disc & cup accuracy > 0.98).
  • Automated CDR-based diagnosis with per-task precision/recall/IoU.

    • Key Learnings

    • Consistent preprocessing (fixed input shape) is critical for U-Net.
    • Skip connections preserve spatial detail for fine segmentation.
    Related Case Studies
    Computer Vision

    Bag of Visual Words for Image Classification

    Computer Vision

    VOC Segmentation App with Custom U-Net

    Computer Vision

    Satellite Image Matcher & Coordinate Projector

    CONTACT US

    Let's Discuss YourNext Project

    Ready to build something exceptional? Share your idea and we'll respond with a clear plan, honest timeline, and competitive quote within 24 hours.