Overview

This repository contains a binary X-ray classifier implemented in PyTorch, designed to classify X-ray images into two categories: normal and pneumonia-infected lungs. The model utilizes a Convolutional Neural Network (CNN) architecture and is trained on a dataset containing labeled X-ray images of normal and pneumonia cases.

Prerequisites:

• Python 3.7 or later
• PyTorch
• torch.optim
• torch.nn
• torchvision
• typer
• pathlib

Install the required dependencies using the following command:

pip install torch torchvision typer

Usage

Clone the repository:

https://github.com/RobyIm/binary-xray-classifier.git

Or type this in the terminal:

git clone https://github.com/RobyIm/binary-xray-classifier.git

Run the classifier using the provided run.py file:

python run.py run --save_dir /path/to/save/model --model_type X_Ray-cnn --train_model --enable_checkpoints --test_model

Parameters:

• save_dir: Directory to save the model and checkpoints. model_type: Choose the model type (currently only 'X_Ray-cnn' is available).
• train_model: Train the model (default: True).
• enable_checkpoints: Enable saving checkpoints during training (default: True).
• test_model: Test the model on the test set (default: True).
• test_model_path: Path to a pre-trained model for testing. If not provided, the latest trained model will be used.

Model Architecture

The model architecture is defined in BinaryX_RayCNN.py. It is a Convolutional Neural Network designed for binary classification of X-ray images.

Training and Testing

Training is performed using the X_Ray_cnn_train_fn function, and testing is performed using the X_Ray_cnn_test_fn function. Checkpoints are saved during training in the specified save_dir.

Citation

If you use the provided dataset, please cite the original source: Kermany, Daniel; Zhang, Kang; Goldbaum, Michael (2018), “Labeled Optical Coherence Tomography (OCT) and Chest X-Ray Images for Classification”, Mendeley Data, V2, doi: 10.17632/rscbjbr9sj.2