Datasets and notebooks for reproducibility of manuscript "Exploring de-anonymization risks in PET imaging: Insights from a comprehensive analysis of 853 patient scans"

This folder contains everything needed to reproduce the figures from our paper.<b>/notebooks</b>The /notebooks folder contains the code that generated our figures.Results_Denoising.ipynb can be used to reproduce Figure 3, which represents our landmark placement performance with and without denoising step. We also plot the training and validation losses across all folds of our denoising model.Landmark_matching.ipynb can be used to reproduce Figure 4, which illustrates the performance of our landmark matching step between PETs and CTs. This step is a proxy for face detection in the absence of ground truth patient faces, as described in our manuscript.<b>/data</b>This folder contains all the data necessary to reproduce the figures. This data is used in the notebooks. /data/training_metrics contains .npy and .csv files with raw training data (training and validation losses per fold), as well as evaluation metrics such as our normalized version of the Mean Absolute Distance between ground truth landmarks (on CTs) and PET landmarks ( quality_5_fold.csv ), as well as the pourcentage of well placed landmarks before and after denoising ( perc_5_fold.csv )./data/landmarks contains the raw coordinates of each landmark on each morphological reconstruction (that we could not share as it would breach our Data Usage Agreement as explained in the Methods section of our paper) per fold. /data/nearest_neigh_landmarks_repositioning contains the distance matrix from each PET to each CT according to our landmark matching metric for each fold x in [1,5] , with ( distances_fold{x}_aligned.npy ) and without alignement ( distances_fold{x}.npy ).