Machine Learning - MATLAB

This project implements a deep learning algorithm in MATLAB to classify diabetic retinopathy images using transfer learning on the InceptionV3 convolutional neural network (CNN), using data from IDRiD Sub-challenge 2 on the Grand Challenge platform.

Overview

  • Classified diabetic retinopathy images into 5 severity classes using the IDRiD dataset (516 images).
  • Retrained InceptionV3 multiple times and combined predictions across runs.
  • Achieved over 90% accuracy on the test set.
  • Results include a confusion matrix, a combined probability table, and an accuracy graph.

Methodology

  • Retrained InceptionV3 CNN 11 times: once on all classes and 10 times on combinations of 2 classes
  • Combined predicted probabilities from all retrainings to create a final probability matrix
  • Assigned each sample to the class with the highest total probability
  • Misclassifications typically occur in smaller classes; the correct class is often the second-highest probability

Report

Note: Running the MATLAB script without retraining the CNNs will produce this same report. Retraining the CNNs generates an updated PDF report and an Excel file with the new results.

Project Files

The full project, including MATLAB scripts and supporting files, is available on GitHub:

→ diabetic-retinopathy-classifier Repository

← Back to University Projects