MICCAI 2015 Submission:
I worked on the development of a methodology that could allow us to validate segmentation results without requiring a previously annotated database (ground truth).
I have implemented the method proposed in Kohlberger et al (MICCAI 2012) to try to reproduce their results.
The obtained numerical results are as low as those in the paper. In the original method, they used categorical results (grouping of numerical results) to avoid the problem.
For every proposed feature, I have explored plotting the segmentation dice score as a function of alterations of the original segmentation (through dilation and emotion) . The results showed that there was not a relationship between these. For this reason, we decided not to submit the paper.
Additionally, I have been supporting Guotai Wang on his MICCAI submission on Placenta Segmentation.
Maria Alejandra Zuluaga V.