Chest X-ray Pneumonia Detection

AI-powered diagnostic tool using deep learning to automatically detect pneumonia from chest X-ray images with high accuracy.

Overview

This project implements a convolutional neural network based on MobileNet architecture to detect pneumonia in chest X-ray images. It offers two training approaches: dataset balancing through oversampling and class-weight adjustment. The model achieves precision and recall metrics suitable for clinical decision support, demonstrating the potential of AI to assist healthcare professionals in pneumonia screening.

The project aims to demonstrate the effectiveness of deep learning in medical image analysis and its potential to assist healthcare professionals in diagnosing pneumonia accurately and efficiently.

Dataset

To reproduce the experiments conducted in this project, you need to download the Chest X-ray dataset from the following link: Mendeley.

After downloading, follow these steps:

1. Unzip the folder: Extract the contents of the downloaded file named "ChestXRay2017.zip".
2. Specify the dataset path: Copy the correct path to the folder containing the dataset inside the code. You may need to update the file paths in the code accordingly.

Make sure to have the dataset available in the specified directory to run the code successfully.

Usage

Before running the code, ensure you have downloaded the dataset and placed it in the appropriate directory. You should also choose which approach you want to use for training by commenting out the respective sections in the code.

Prerequisites

• Python 3.6+
• Jupyter Notebook
• TensorFlow 2.0+
• Keras 2.4+
• OpenCV
• NumPy
• Pandas
• Matplotlib
• Scikit-learn
• Seaborn

Installation

1. Clone the repository:

   git clone https://github.com/PeretzNiro/Chest-X-Ray-Images-Classification.git
   cd Chest-X-Ray-Images-Classification

2. Install the required libraries:

   pip install -r requirements.txt

3. Download the dataset and place it in the specified directory.

Running the Code

Open the Jupyter Notebook file (Chest_X_Ray_Medical_Images_Classification.ipynb) and execute the code cells, paying attention to the following:

1. Dataset Selection:

   • If using the original dataset: You should use the class-weight adjustment approach to handle class imbalance
   • If using the augmented dataset: The data augmentation code will generate additional images to balance the classes

2. Code Sections:

   • For data augmentation: Run the cell containing ImageDataGenerator code if you want to balance classes by generating additional images
   • For class weights: Use this approach only with the original unbalanced dataset

Make sure to update file paths in the notebook to match your local environment before running.

Results

The model demonstrates strong performance in pneumonia detection with the following metrics:

• High accuracy on the test set
• Balanced precision and recall
• Strong ROC curve and AUC score

License

This project is licensed under the MIT License.

Acknowledgements

The codebase for this project was adapted from Khan Fashee Monowar's Kaggle notebook. We acknowledge and appreciate their work in providing the foundational code for this project.