Robust exemplar model of respiratory liver motion and individualization using an additional breath-hold image

This study investigates the benefits of observing an example of the subject-specific 3D liver motion for individualizing a statistical population model and the influence of tracking errors on the model predictions based on 4D-MRI data from 16 subjects, simulated erroneous 2D tracking surrogates and leave-one-subject-out tests. A robust formulation of the exemplar model was proposed, which lead to mean (95%) prediction improvements of 6% (6%) for tracking errors with 1mm standard deviation. Individualization by an example 3D motion field from the 4D-MRI provided additional improvements (mean: 7%, 95%: 13%). Similar benefits could be achieved from breath-hold images, after automatically detecting and excluding unsuitable motion examples.