khcs/learning-to-read

Image caption generation to diagnose chest x-rays using dataset of images and reports

Learning to Read Chest X-Rays

Source code for the CVPR 2016 paper:

Learning to Read Chest X-Rays: Recurrent Neural Cascade Model for Automated Image Annotation (arxiv)

The source code is based on the:

• cifar.torch by Sergey Zagoruyko, for the CNN training
• char-rnn by Andrej Karpathy, for the RNN training

The overall pipeline and codes for preparing data, training, and sampling can be found in the src/chestx

Dataset can be downloaded from the Open-i chest x-ray subset (link)

Trained models can be downloaded:

• CNN - GoogLeNet model
• RNN-LSTM, RNN-GRU

Required software packages:

• Torch
• Python 2.7 with the following packages:
  • Numpy
  • Matplotlib
  • Pandas
  • Scikit-Learn
  • Scikit-Image
  • iPython Notebook - Optional
  • NLTK - Optional