Adelaide University Summer Research Scholarship for Radiology at the Royal Adelaide Hospital

The data files are not included in the repository and need to be manually added. The model files need to be generated before running the search engine.

If you are running Linux you need to install the following packages: gfortran libfreetype6-dev libpng-dev libyamal-dev libhdf5-dev You may also want: liblapack-dev libopenblas-dev

The search engine requires the following python dependencies: numpy scipy cython statsmodels gensim nltk matplotlib sklearn The RNN code also requires: pyyaml h5py keras (configured with theano backend) theano All of these can be installed with: pip install --upgrade packageName

For setup of theano it is recommended that you make use of a gpu, for setup of Theano on ubuntu see: http://deeplearning.net/software/theano/install_ubuntu.html#install-ubuntu

Once these are installed you need to download nltk's stopwords. Open python and run: import nltk nltk.download() The package identifier is "stopwords"

Word2Vec model has minimal training time using CPU for both datasets RNN sentence and report models were trained on GTX980. RNN Sentence model: 15-20 epochs on full dataset, 2hrs15min/epoch 15 epochs on small dataset, 25min/epoch RNN report model:

160 epochs on small dataset, 4-5min/epoch