This repo does not contain the data used in experiments. Data can be downloaded from https://physionet.org/content/mimic-cxr-jpg/2.0.0/ after signing Physionet's data use agreement.
Scripts for the original 28-concept experiment can be found in the 28-concept directory. Scripts for the fully clustered experiment can be found in the 6-cluster directory.
Data must be in the following format:
- data
- attributes.txt (attribute id, attribute name)
- classes.txt (class id, class name)
- image_attribute_labels.txt (img id, attribute id, is_present)
- image_class_labels.txt (img id, class id)
- images.txt (img id, filename)
- train_test_split.txt (img id, is_training)
- images
- (JPG IMAGE FILES - NOT INCLUDED)
Concept extraction scripts can be found in the concept extraction directory.