🩺 Pneumonia Detection using Deep Learning

This project uses deep learning techniques to detect pneumonia from chest X-ray images. Early diagnosis can save lives, and this automated approach aims to assist clinicians in identifying pneumonia efficiently.

🔍 Project Overview

• Model Architecture: VGG16 (pre-trained on ImageNet)
• Transfer Learning: Only custom top layers are trained; all convolutional layers are frozen
• Dataset: Chest X-ray images labeled as NORMAL or PNEUMONIA
• Platform: Built for use in Kaggle or Colab

🛠️ Technologies Used

• Python & Keras
• VGG16: CNN for feature extraction
• ImageDataGenerator: For rescaling and augmentation
• Matplotlib: For accuracy/loss visualization
• NumPy & SciPy: For preprocessing and predictions
• KaggleHub: To fetch public datasets

🧠 Model Pipeline

[Chest X-Ray Dataset]
   ↓
[Preprocessing]
   ↓
[Train-Test Split]
   ↓
[Load VGG16 (no top)]
   ↓
[Freeze Layers]
   ↓
[Add Flatten + Dense + Softmax]
   ↓
[Compile Model (Adam + CrossEntropy)]
   ↓
[Train for 5 Epochs]
   ↓
[Evaluate]
   ↓
[Save & Load Model]
   ↓
[Predict on New X-ray]
   ↓
[Classification Output: NORMAL or PNEUMONIA]

🚀 How to Run

1. Set Up Dataset

Use kagglehub to download:

path = kagglehub.dataset_download("paultimothymooney/chest-xray-pneumonia")

2. Build and Compile Model

vgg_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
for layer in vgg_model.layers:
    layer.trainable = False

Add custom layers:

x = Flatten()(vgg_model.output)
x = Dense(8, activation='relu')(x)
output = Dense(2, activation='softmax')(x)
model = Model(inputs=vgg_model.input, outputs=output)

Compile:

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

3. Load Data

training_set = train_datagen.flow_from_directory(..., class_mode='categorical')
testing_set = test_datagen.flow_from_directory(..., class_mode='categorical')

4. Train and Evaluate

model.fit(training_set, validation_data=testing_set, epochs=5)
model.save('our_model.h5')

5. Predict

img = image.load_img(image_path, target_size=(224, 224))
image_data = preprocess_input(np.expand_dims(image.img_to_array(img), axis=0))
prediction = model.predict(image_data)

📈 Visualization

Accuracy and loss curves for training and validation are plotted using matplotlib.

🔮 Future Enhancements

• ✅ Add more CNN architectures like ResNet or EfficientNet for better performance
• ✅ Hyperparameter tuning for optimal learning rates and batch sizes
• ✅ Deploy model as a web app using Flask or Streamlit
• ✅ Integrate Grad-CAM visualizations to interpret model predictions
• ✅ Enable ensemble learning by combining predictions from multiple models
• ✅ Build mobile-friendly diagnostic tools for use in remote areas
• ✅ Expand dataset to include multiple pneumonia stages or other diseases

---

📁 License

This project is for educational and research purposes. Refer to the source datasets’ licensing terms on Kaggle.