Bharath1434/Pneumonia-Detection-1

---

tags: - image-classification - medical - pneumonia-detection - tensorflow - keras - vgg16 - transfer-learning

VGG-16 (Base + FFNN) Pneumonia Detector

This repository hosts a VGG-16 based Convolutional Neural Network (CNN) model with a custom Feed-Forward Neural Network (FFNN) head developed for detecting pneumonia from chest X-ray images. The model leverages transfer learning from the VGG-16 architecture, fine-tuned for this specific medical imaging task.

Model Details

• Architecture: VGG-16 (Base) + Custom Feed-Forward Neural Network
• Task: Binary image classification (Pneumonia vs. No Pneumonia)
• Input Image Size: 200x200 pixels, 3 channels (RGB, replicated grayscale)
• Framework: TensorFlow/Keras
• Dataset: RSNA Pneumonia Detection Challenge (subset)

How to use

This model can be loaded using TensorFlow/Keras and used for inference on preprocessed chest X-ray images.

import tensorflow as tf
from huggingface_hub import from_pretrained_keras

# Load the model from Hugging Face Hub
model = from_pretrained_keras("Bharath1434/Pneumonia-Detection-1")

# Example: Preprocess your image (assuming 'your_image' is a 200x200x3 numpy array, normalized 0-1)
# preprocessed_image = your_preprocessing_function(your_image)

# Make a prediction
# prediction = model.predict(tf.expand_dims(preprocessed_image, axis=0))[0][0]

# if prediction > 0.5:
#     print("Pneumonia Detected (Probability: {prediction:.4f})")
# else:
#     print("No Pneumonia (Probability: {prediction:.4f})")

Training Information

• Training Data Size: A subset of the RSNA Pneumonia Detection Challenge dataset.
• Validation Accuracy: Approximately 0.7844
• Test Accuracy: Approximately 0.7867
• Key Features: Transfer learning approach with VGG-16, followed by a custom FFNN and Dropout layers. Class weights were used to handle dataset imbalance.

For more details on the training process, EDA, and model evaluation, please refer to the associated Colab notebook.