2014 Summer Project Week:Cardiac-Congenital
- Danielle Pace, MIT
- Adrian Dalca, MIT
- Polina Golland, MIT
This project involves semi-automatic segmentation of gated 3D magnetic resonance images of hearts with congenital heart defects. Our aim is to create patient-specific heart models for surgical planning, which can be viewed either graphically on a computer, or with a 3D printer to create a physical model for surgeons.
We have had initial success in significantly reducing segmentation time with the following pipeline: 1) User manually segments ~10 axial slices 2) Segment the remaining slices using patch-based majority voting.
- Continue testing the patch based methods on an additional five datasets.
- Address main remaining challenge: segmenting thin interior heart walls.
- Manually segment ~10 axial slices for each of the five additional datasets.
- Explore remaining parameters for the patch-based segmentation: e.g. weighted voting, varying k in k-nearest neighbors patch lookup
- Did manual initial segmentations and and ran patch-based segmentation on additional datasets.