Built a deep learning pipeline, with Dhanrajsinh Parmar, to classify lung diseases from chest X-ray images. Three architectures — a custom Convolutional Neural Network, ResNet50, and VGG16 — were trained and compared, with each model's top layers fine-tuned to adapt to the classification task.
The workflow covers the full pipeline: loading and organizing the X-ray dataset into classes, applying image augmentation via Keras' Image Data Generator, training each model with early stopping to guard against overfitting, and evaluating results with accuracy and loss curves plotted across training epochs. The final trained ResNet50 model is saved as a portable HDF5 file for reuse without retraining.
The project design was informed by a review of related research on integrating deep learning with hand-crafted image features for diagnosing pneumonia and tuberculosis — comparing that paper's approach against this project's own architecture choices and identifying hand-crafted feature fusion as a promising direction for future improvement.