A population-based approach to the simulation of realistic 3D myocardial infarction
Résumé
Large amounts of data with ground truth are generally needed for training and/or validating machine learning algorithms. In medical imaging, image datasets are often of limited size, hence the high interest in simulated data. However, for the effective use of such data, the simulation models should be personalized to the real population. Population-wise personalization recently showed as a viable alternative to individual personalization, which is not always possible, in particular, for models involving randomness. The recent 2D method showed promising results for the personalization of two simple 2D geometrical models of infarction: 1) iterative, consisting of the union of random number of random spheres within the myocardium, and 2) elliptical, consisting of one random ellipsoid within the myocardium. These 2D approaches had a fast computing time and allowed easy visualization to validate the method, but with limited complexity of infarct patterns. In this work, we demonstrate the extension of this 2D method to 3D infarct patterns.
Origine | Fichiers produits par l'(les) auteur(s) |
---|