The paper, “Optimization for Multi-Region Segmentation of Cardiac MRI”, is based on Andrew Delong and Yuri Boykov’s multi-region paper from ICCV 2009. We segment the left and right ventricles, myocardium and the papillary muscles jointly.
1. The Model
The above figure shows the model. We use five regions: background (0), myocardium (1), left and right ventricles (2,3) and papillary muscles (4). As it turns out, the required geometric constraints do not result in a submodular function. While roof duality would be a possible minimization approach, we instead use supergradient minimization, which requires less memory.
We evaluate the method on two databases, one publicly available and one proprietary from the University Hospital in Lund. To our knowledge there is no publicly available database with ground truth for both ventricles.
We obtain results close to the state of the art for the left ventricle and very good results for the database with complete ground truth. This paper demonstrates that it is possible to apply global optimization techniques for segmentation of cardiac MRI using a sophisticated model of the heart. See the full paper for details.