Friday, September 02, 2005

[Thesis] Current Status

1. I have experimented with a lot of variation of the method and never think that optimization can reliably lead to good result. The following shows what I did and result:
1.1. Using bifurcation point of trachea as a reference point gives the same quality of aorta output as using carina. The most difficult part may be finding correct positions of two end points of model.
1.2. Using more model does not mean that we have better chance to succeed. This is really bad news. In CR003 case, using the first model give better output because it detects correct postions of the two end points, although the middle is not quite correct. Using only one model, however, induces high risk of failure.
1.3. Model selection method is not accurate when the image is not isotropic, I think. I must report this to Dr. Higgins.

2. Aortic arch region growing is good in some cases. I'd like to report this stuff honestly in my report. It is good when the edges are quite clear (16-detector cases for example) , but not good at all when the edge is not clear (h005). This suggests me to use allow new parameters for aorta. User shoud know if edges are clear enough or not.

No comments: