Retinal Fundus Disease Classifier

Author: Andrew Feng

Project Overview

This project implements a multi-label disease classifier for retinal fundus images using a fine-tuned ResNet-50 model. It explores classification performance on a medical dataset through training, evaluation, and fine-tuning pipelines. The goal is to effectively detect key retinal diseases, addressing class imbalance and real-world diagnostic challenges.

Dataset

The dataset used in this project is available here:
🔗 Google Drive Dataset

After cloning this repository, download the entire data/ folder from the link above and place it inside the root directory of the project (Fundus-Classifier/). This is required to run the code successfully and reproduce results.

---

Getting Started

1. Clone the Repository

git clone https://github.com/Andyrooooo16/Fundus-Classifier.git cd ResNet50-Retinal-Fundus-Image-Classifier

2. Set Up the Dataset

Download the data/ folder from the Google Drive link Place the folder directly inside the project root:

css Copy Edit Fundus-Classifier/ ├── data/ ├── src/ ├── outputs/ └── README.md

3. Code Structure

All source code is located in the src/ directory:

css Copy Edit Fundus-Classifier/ └── src/ ├── crop_pictures.py ├── dataset.py ├── disease_breakdown.py ├── model.py ├── train_model.py ├── evaluate.py ├── finetuning_model.py ├── evaluate_finetuned.py └── test_finetuned.py

4. How to Run the Project

1. Image Preprocessing

Run 'crop_pictures.py' on the Training, Validation, and Test datasets to generate cropped images for model training, evaluation, and testing.

2. Data Preparation

Run 'dataset.py' to preprocess all three subsets of data.

3. Dataset Analysis

Run 'disease_breakdown.py' to generate three visualization graphs in your browser showing the dataset composition and top disease classes in each subset.

4. Model Setup

The 'model.py' file contains the pretrained ResNet50 model with adjusted hyperparameters for initial training.

5. Model Training

Run 'train_model.py' to train the multi-label classifier using the cropped Training Dataset and the model from 'model.py'. This will save 'best_fundus_model.pth' in the Fundus-Classifier/outputs/Training folder.

6. Model Evaluation

Run 'evaluate.py' to assess the performance of 'best_fundus_model.pth'. Evaluation metrics will be saved in the Fundus-Classifier/outputs/Evaluation folder.

7. Model Fine-tuning

Run 'finetuning_model.py' to fine-tune the model. This will generate 'fine_tuned_model.pth'.

8. Fine-tuned Model Evaluation

Run 'evaluate_finetuned.py' to evaluate the fine-tuned model. Results will be saved under Fundus-Classifier/outputs/Fine_Tuned.

9. Final Testing

Run 'test_finetuned.py' to perform the final evaluation of the fine-tuned model on the test dataset. Final evaluation metrics will be saved in the Fundus-Classifier/outputs/Test folder.