Atlas-Based probabilistic fibroglandular tissue segmentation in breast MRI
MICCAI'12 Proceedings of the 15th international conference on Medical Image Computing and Computer-Assisted Intervention - Volume Part II
| Hi-index | 0.00 |
The parenchyma tissue in the breast has a strong relation with predictive biomarkers of breast cancer. To better segment parenchyma, we perform segmentation on estimated tissue property T1map. To improve the estimation of tissue property (T1) which is the basis for parenchyma segmentation, we present an integrated algorithm for simultaneous T1map estimation, T1map based parenchyma segmentation and group-wise registration on series of inversion recovery magnetic resonance (MR) breast images. The advantage of using this integrated algorithm is that the simultaneous T1map estimation (E-step) and group-wise registration (R-step) could benefit each other and jointly improve parenchyma segmentation. In particular, in E-step, T1map based segmentation could help perform an edge-preserving smoothing on the tentatively estimated noisy T1map, and could also help provide tissue probability maps to be robustly registered in R-step. Meanwhile, the improved estimation of T1map could help segment parenchyma in a more accurate way. In R-step, for robust registration, the group-wise registration is performed on the tissue probability maps produced in E-step, rather than the original inversion recovery MR images, since tissue probability maps are the intrinsic tissue property which is invariant to the use of different imaging parameters. The better alignment of images achieved in R-step can help improve T1map estimation and indirectly the T1map based parenchyma segmentation. By iteratively performing E-step and R-step, we can simultaneously obtain better results for T1map estimation,T1map based segmentation, group-wise registration, and finally parenchyma segmentation.