Modeling the Mechanical Behavior of the Breast Tissues Under Compression in Real Time

This work presents a data-driven model to simulate the mechanical behavior of the breast tissues in real time. The aim of this model is to speed up some multimodal registration algorithms, as well as some image-guided interventions. Ten virtual breast phantoms were used in this work. Their deformation during a mammography was performed off-line using the finite element method. Three machine learning models were trained with the data from those simulations. Then, they were used to predict the deformation of the breast tissues. The models were a decision tree and two ensemble methods (extremely randomized trees and random forest). Four experiments were designed to assess the performance of these models. The mean 3D euclidean distance between the nodal displacements predicted by the models and those extracted from the FE simulations were used for the assessment. The mean error committed by the three models were under 3mm for all the experiments, although extremely randomized trees performed better than the other two models. Breast compression prediction takes on average 0.05 s, 0.33 s and 0.43 s with decision tree, random forest and extremely randomized trees respectively, thus proving the suitability of the three models for clinical practice.