PROGRESSA

This repository contains the code to reproduce the results of the paper Sanabria, M, Tastet, L, Pelletier, S, Leclercq, M, Ohl, L, Hermann, L, Mattei, P-A, Precioso, F, Coté, N, Pibarot, P, Droit, A. AI-Enhanced Prediction of Aortic Stenosis Progression: Insights From the PROGRESSA Study. JACC Adv. 2024 Oct, 3 (10) . https://doi.org/10.1016/j.jacadv.2024.101234

Install

Install the package and necessary dependencies by running the command:

pip3 install .

or

python3 setup.py build
python3 setup.py install

PREPROCESS

Before training the models, the first step is to run the script to extract the features using the command:

python3 progressa/preprocess/extract_features.py

The next step is to create a file than will contain the indices for the 10 splits that will be used for the repeated holdout method. To achieve this, run the script:

python3 progressa/preprocess/create_splits.py

Then, the features importance is calculated using the script :

python3 progressa/preprocess/feature_importance.py

The most important features returned from running this script were then entered in the select_features.py script. To get the file with selected features only, which will be found in data/features-22.pkl, run:

python3 progressa/preprocess/select_features.py

Train models

To train the RNN model (GRU):

python3 progressa/train_models/RNN.py

To train the machine learning models compared with RNN, use the following command lines:

python3 progressa/train_models/sklearn_models.py --model=naiveBayes python3 progressa/train_models/sklearn_models.py --model=Logistic_Regression python3 progressa/train_models/sklearn_models.py --model=lightgbm python3 progressa/train_models/sklearn_models.py --model=xgboost

To train on 2 visits at a time, modify the precedent commands by adding the command --n_visits=2. For example:

python3 progressa/train_models/sklearn_models.py --model=naiveBayes --n_visits=2

Analysis

To reproduce the analysis from the paper, run the following commands

python3 progressa/analysis/analyse_results.py

python3 progressa/analysis/analyse_results_per_sex.py

python3 progressa/analysis/severity_baseline.py

python3 progressa/analysis/stats.py

python3 progressa/analysis/calibration_plot.py

Create images

To reproduce the images from the paper, run the following commands

python3 progressa/create_images/plot_rocs.py

python3 progressa/create_images/plot_tsne.py

Funding support

This work has been supported by MITACS Globalink (IT25650), Canadian Institutes of Health Research (#FDN-143225 and MOP-114997), Foundation of the Québec Heart and Lung Institute, Fonds de Recherche du Québec en Santé (FRQS), France Health Data Hub (HDH), and institutional research funds held by Drs Droit and Precioso