This folder contains the datasets used for the paper "Transfer learning with weak labels from radiology reports: application to glioma change detection". A preprint of the manuscript can be found on arXiv. Inside the parent directory, you will find two datasets: one is the in-house dataset from the University Hospital of Lausanne (CHUV), and the second is made ofpreprocessed difference maps from the longitudinal patientsof the BraTS-TCIA-2015 dataset. The labels associated with both datasets can be found on the corresponding github repository (https://github.com/connectomicslab/Glioma_Change_Detection_T2w/tree/master/extra_files). Inside the in-house dataset directory, you will find two subdirectories (sub-datasets) named HAD_diffmaps and WAD_diffmaps. Both sub-datasets contain T2w difference maps of patients with high-grade gliomas. HAD stands for Human-Annotated Dataset: as explained in the paper, the labels for this sub-dataset were created manually from three radiologists by looking at the corresponding radiology reports. Conversely, WAD stands for Weakly-Annotated Dataset: for this sub-dataset, the labels were created automatically with a Natural Language Processing pipeline for radiology reports. Please refer to the paper for more details. For each subject inside any of the two sub-datasets you will find session pairs. This is because every report/difference map links two time points (i.e. two sessions), since we tackle longitudinal change detection. Then, inside each session pair folder, you will find the following files: - sub-XX_ses-001_vs_ses-002_comparative_bet_cropped_t2_n4.nii.gz -> is the comparative/previous (i.e. ses-001) T2w volume already with N4 bias field correction (bfc), and already skull-stripped (bet = brain extraction performed with HD-BET) - sub-XX_ses-001_vs_ses-002_comparative_bet_cropped_zscored_t2_n4.nii.gz -> is the comparative T2w volume with N4 bfc, bet, and z-score normalized - sub-XX_ses-001_vs_ses-002_current_bet_cropped_t2_n4.nii.gz -> is the current (i.e. ses-002) T2w volume with N4 bfc, and bet - sub-XX_ses-001_vs_ses-002_current_bet_cropped_zscored_t2_n4.nii.gz -> is the current Tw2 volume with N4 bfc, bet, and z-score normalized - sub-XX_ses-001_vs_ses-002_difference_t2_volumes.nii.gz -> is the voxel-wise difference maps between the two normalized volumes. It is the one used for classification in the paper - sub-XX_ses-001_vs_ses-002_out_prev_2_curr_0GenericAffine.mat -> is the affine matrix generated by ANTs when registering the previous comparative volume to the current - sub-XX_ses-001_vs_ses-002_reg_quality_metrics_comp2curr.csv -> it contains some metrics (Neighborhood Correlation and Mutual Information) to monitor the registration quality