Pneumonia Detection using Convolutional Neural Networks (CNN)

This deep learning project applies convolutional neural networks (CNNs) to classify chest X-ray images as either Normal or Pneumonia. The model is trained, validated, and evaluated using a well-known medical imaging dataset.

---

Dataset

• Source: Kaggle - Chest X-Ray Images (Pneumonia)
• Structure: Contains chest X-ray images labeled as either NORMAL or PNEUMONIA
• Splits: Pre-divided into train, val, and test sets
• Added pictures to val from train and test for better training

---

Objectives

• Build a CNN model to detect pneumonia in X-ray images
• Improve performance using data augmentation
• Evaluate model using accuracy, confusion matrix, and classification report

---

Model Architecture

• 3 Convolutional layers with ReLU activation and MaxPooling
• Flatten layer followed by Dense layers
• Output layer with Softmax activation (for 2-class classification)
• Compiled with Adam optimizer and categorical cross-entropy loss

---

Techniques Applied

• Data augmentation (rotation, zoom, shift, horizontal flip)
• Batch image processing using Keras ImageDataGenerator
• Visualization of training samples and performance metrics
• Model saving for reuse and deployment
• Added early stopping and model checkpoint callbacks

---

Visualizations

• Sample augmented training images
• Confusion matrix on test set
• Classification report with precision, recall, F1-score

---

Results

• Test Accuracy: ~83%
• Performance: Balanced precision/recall between NORMAL and PNEUMONIA classes
• Saved Model: pneumonia_model.keras

---

What I Learned

• How to preprocess and augment image datasets for CNNs
• How to build and tune CNN architecture for medical image classification
• The importance of validation and evaluation on unseen data

---

Future Improvements

• Experiment with transfer learning (e.g. VGG16, ResNet)

• Explore Grad-CAM for interpretability of CNN predictions

• Credits: Dataset originally published by Paul Mooney on Kaggle