jerryk42/Deep-Learning-Xray-Fracture-Classifier — reverse-engineered prompt

Build me a Google Colab project that can train deep learning models to read musculoskeletal X ray images and predict whether there is a fracture or not. It should use the MURA dataset from Stanford, with clear places where I can point to the train and validation folders in Google Drive.

I want the project to compare a simple custom model with stronger pretrained image models like ResNet50, DenseNet169, and EfficientNet B0. The better models should also learn which body part is shown, like elbow, finger, hand, shoulder, wrist, and so on, because that context should help fracture detection.

Please make the notebooks easy to run from top to bottom, with setup, data loading, training, validation, charts, accuracy, ROC AUC, and example predictions. Also include a short README that explains how to download the data, mount Google Drive, change paths, and run everything. Look up current PyTorch or Colab docs online if you need to.