Active Machine Learning for Pre-procedural Prediction of Time-Varying Boundary Condition After Fontan Procedure Using Generative Adversarial Networks