Classifying Chest X-Ray Images Using CNN Architectures

The project aims to harness the power of Convolutional Neural Networks (CNNs) to classify chest X-ray images for various diseases. Utilizing advanced deep learning models, specifically DenseNet-121 and ResNet-50, the project seeks to improve the accuracy and efficiency of diagnosing chest-related diseases from X-ray images. These models are known for their effectiveness in image classification tasks. The project involves training and fine-tuning these models on a comprehensive dataset of chest X-ray images, aiming to establish a reliable, automated system for early and accurate disease detection in the medical imaging field.

Dataset

The dataset for this project is a comprehensive amalgamation of three distinct datasets from Kaggle, enriched to form a new dataset with 18 classes:

• Chest X-ray – 17 Diseases: This dataset includes X-rays in both JPEG and DCM formats, organized into 17 folders named after specific diseases or conditions [1].
• Chest X-Ray Images (Pneumonia) Comprising 5863 X-ray images in JPEG format, this dataset categorizes images as either depicting pneumonia or being normal [2].
• Tuberculosis (TB) Chest X-ray Database Contains chest X-ray images of normal individuals (3500) and patients diagnosed with Tuberculosis [3].

From these datasets, a novel dataset was created, improving the scope of the project to encompass 18 disease categories. Its possible to download the novel dataset here: https://drive.google.com/file/d/1a-CUf1Zy5kj-c7gcNMf2KISq855c1s9D/view?pli=1

Technologies Used

The project utilizes a range of technologies and tools for its implementation. Key technologies include:

• TensorFlow and Keras: For building and training the CNN models.
• Google Colab: Used for leveraging GPU acceleration during model training.
• Nvidia-smi library: Employed to speed up the training process.

Setup

Cloning the repository

Run this command to download the repository:git clone https://github.com/FioroniCostanza/Chest_XRay.git

If you are working on Google Colab

Load the dataset in your Drive and remember to replace the code lines of the dataset path.

If you are working in a Local Environments

Remember just to replace the path of the dataset.

Dependencies

To Install the necessary libraries you just need to run their installation commands, like:

pip install tensorflow
pip install keras

already provided in this file Xray_chest.ipynb.

Repository File Descriptions

Preprocessing Scripts

The following files have been used for preprocessing the dataset, ensuring it is clean and standardized for analysis:

• rename_images.py: Script that rename the images after the merge of the three dataset in order to have an homogeneous dataset.
• black_border.py: Script to remove black borders from images.
• white_border.py: Script to remove white borders from images.
• white_boxes.py: Script to remove white boxes from images.

To used them is neccessary to manually pick the images with these features and put them in the structure given by the dataset divided in folders named by the class's names. The dataset in this repository github already provides the preprocessing, so it's not neccessary to run the three files cited before.

Main Project Notebook

• Xray_chest.ipynb: This Jupyter notebook contains the core of the project. It includes data loading, model building with CNN architectures (DenseNet-121 and ResNet-50), training, evaluation, and results presentation. All major analyses and findings of the project are documented here.

References

1. ”Chest X-ray – 17 Diseases” Kaggle.com. [Online]. Available: https://www.kaggle.com/datasets/trainingdatapro/chest- xray- 17- diseases
2. ”Chest X-Ray Images (Pneumonia)”, Kaggle.com. [Online]. Available: https://www.kaggle.com/datasets/paultimothymooney/chest-xray-pneumonia
3. ”Tuberculosis (TB) Chest X-ray Database”, Kaggle.com.[Online]. Available: https://www.kaggle.com/datasets/tawsifurrahman/ tuberculosis- tb- chest- xray- dataset