Pneumonia Detection

Overview

This repository contains a complete end-to-end deep learning pipeline for detecting pneumonia from chest X-ray images using modern convolutional neural networks and GAN-based augmentation.

The project includes:

• Transfer learning models (EfficientNetB0, DenseNet121, CheXNet, ResNet50)
• Two-phase training (frozen base → fine-tuning)
• Dataset preprocessing & augmentation
• Automatic class balancing using classical augmentation
• Optional GAN-based synthetic image generation
• Evaluation tools (confusion matrix, ROC curve, PR curve, metrics, reports)
• Model checkpointing, history plots, and result saving

This project is structured for reproducible medical-imaging experiments and can be used both for research and for production-grade training workflows.

---

📦 Repository Structure

├── preprocessing.py                # Dataset analysis, visualization, augmentation
├── balancing.py                    # Classical augmentation-based balancing
├── transfer_learning_classifier.py # Transfer learning architectures and training logic
├── train_transfer_models.py        # Main script for training CNN models
├── gan_generator.py                # Script for generating synthetic X-ray images using GAN
├── models/                         # Saved models (.h5)
├── checkpoints/                    # Best-model checkpoints
├── results/                        # Confusion matrix, ROC, PR curves, metrics
└── chest_xray/                     # Expected dataset structure (train/val/test)

---

⚙️ Requirements

• Python 3.12
• TensorFlow / Keras
• NumPy, Matplotlib, Seaborn
• Scikit-learn
• PIL / OpenCV
• tqdm

Install dependencies:

pip install -r requirements.txt

---

📁 Dataset Structure

The project expects the Chest X-Ray dataset in the following format:

chest_xray/
    train/
        NORMAL/
        PNEUMONIA/
    val/
        NORMAL/
        PNEUMONIA/
    test/
        NORMAL/
        PNEUMONIA/

You may use the Kaggle Chest X-Ray Pneumonia dataset or any dataset matching this structure.

---

🚀 How to Run the Project

There are two main runnable scripts:

1. Train a Transfer Learning Model

File: train_transfer_models.py

This script:

• Analyzes the dataset
• Creates augmentation generators
• Optionally balances the dataset
• Builds and trains a selected model
• Saves:
  • best checkpoint
  • final model
  • training history plots
  • confusion matrix / ROC / PR curves
  • metrics JSON
  • classification report

Usage

python3.12 train_transfer_models.py --model densenet121 --dataset chest_xray

Supported models

• densenet121
• efficientnetb0
• resnet50
• chexnet (DenseNet121 with pretrained CheXNet weights)

---

2. Generate Synthetic Images Using GAN

File: gan_generator.py

This script loads a trained GAN generator and produces synthetic X-ray images (grayscale) for augmentation.

Usage

python3.12 gan_generator.py \
            --model path/to/generator.h5 \
            --num_images 2500 \
            --output_dir gan-generated \ 
            --upscale 224 \
            --prefix gan_ 

Generated images can then be copied to:

chest_xray/val/NORMAL/

or any other class folder depending on your needs.

---