XRay Image Classifier Using CNN (ResNet50)

Overview

This project involves designing and training a deep learning classifier to detect lung conditions from X-ray images. The model classifies chest X-ray images into four categories:

• COVID-19 Positive
• Normal
• Lung Opacity
• Viral Pneumonia

The primary objective was to develop a model that accurately identifies these conditions from X-ray images, assisting in rapid diagnosis and healthcare support.

Dataset

The model was trained, tested, and validated on the COVID-19 Radiography Database, which consists of approximately 21,165 X-ray images spanning the four classes mentioned above.

Model Architecture

The deep learning model was implemented using ResNet50, a state-of-the-art convolutional neural network (CNN) architecture known for strong performance in image classification tasks.

• Input: X-ray images
• Output: One of four classes (COVID-19, Normal, Lung Opacity, Viral Pneumonia)
• Training Steps: 10 steps per epoch
• Epochs: 30
• Validation Steps: 16

The model’s performance was evaluated using standard image classification metrics and visualized through a confusion matrix.

Performance

• Accuracy: ~80.76%
• The model demonstrates strong capability in distinguishing between COVID-19, Viral Pneumonia, Lung Opacity, and Normal lung X-rays.

Platform

• Environment: Jupyter Notebook
• Libraries: TensorFlow, Keras, NumPy, Pandas, Matplotlib, and other standard Python ML libraries

Notes

• This project is intended for educational and research purposes.
• The model performance may vary depending on system configurations and dataset availability.
• Proper preprocessing of images is crucial for maintaining model accuracy.