Final year Masters project at Imperial College on tackling Crohn's Disease
arXiv: https://arxiv.org/abs/1909.00276
Paper to be presented at MICCAI 2019, Shenzhen
In this work we establish a baseline for binary prediction of terminal ileal Crohn's disease in abnormal and healthy MRI volumes, using deep learning
To this end we use a small 3D ResNet with added soft attention layers
Brief explanation of important files
/run_crohns.sh - Run config specifying training and model parameters (root of execution)
/run.py - Parses config options and builds TF Record decode function, starts training procedure
/pipeline.py - Builds TF Record load pipeline using decode function
/trainer.py - Constructs and iteratively trains TF network, continually loading TF Record data through pipeline
/model/resnet.py - Specification for 3D Resnet
/model/attention.py - Specification of soft attention mechanism
Files under /preprocessing/ generate the TF Records that are consumed in training
/preprocessing/metadata.py Loads labels and MRI metadata into memory
/preprocessing/preprocess.py Crops and rescales MRI volumes
/preprocessing/tfrecords.py Generates a series of training and test TF Records for cross-fold evaluation (introducing duplication)
/preprocessing/generate_tfrecords.py Configures and executes the generation process (i.e. how many cross folds)