Apr 26 06 (project meeting)
From Utah Center for Neuroimage Analysis
Present: Ross, Tolga, Liz, Pavel, Bryan, Chi-Bin, Melissa
Contents |
Papers
Liz submitted a paper to MICCAI on axon tracking; Paul is working on a methods paper on slice-to-slice matching to be submitted to a microscopy workshop at MICCAI. Liz's paper will be dual-submitted to this workshop.
Microneuroma data
Bryan and Robert said that a complete reconstruction of the microneuroma dataset would be sufficient for a big paper. After registration, they would do some manual analysis and volume reconstruction.
It was decided that complete registration of this dataset should be a good goal. Paul says that this may be possible within two weeks. The outstanding issue at present is that the autocropper, which is having trouble recognizing borders in the current images, which are not well-defined.
The plan is to import the registered images into Imaris (a commercial volume-analysis package from Bitplane) for analysis. Paul points out that the registered microneuroma dataset at full resolution will likely be ~10 Gb. (i.e., resample the registered data onto a regular grid, and save out as TIFFs). Bryan says that they can probably use a subsampled version for their analysis.
Bryan needs to look into the memory limitations of Imaris.
Paul points out that there are some inconsistencies in the current dataset: different areas are missing in different slices. Bryan says that they still have the EM grids and can go back and reimage them to fill in these holes.
Axon tracking
Liz has been trying a 3D watershed algorithm as an alternative to her axon tracking algorithm. Using a subset of the post-UINTA dataset, she had pretty unsatisfactory results: axon segments tended not to extend through more than a few (~5) slices, ending when the axon got pinched, etc.
Problems with the dataset: noise, intensity variations between slices, limited resolution. Might be worth trying watershed again if we get a significantly improved dataset.
Liz's plan is to go back to the snakes, reading the literature to see if there are techniques that she can use to avoid some of the existing failure modes.
Melissa will be in Heidelberg for the middle two weeks of May, hopefully acquiring new data with better contrast and resolution. In the meantime, she is going to send Liz the reference for the Denk and Horstmann paper; perhaps there is a better dataset there that Liz could use to try her algorithms on.
New student
A new grad student (Sam Gerber) will be joining the project; currently doing a master's but planning to switch to Ph.D.
The next meeting will be May 24th at 10 am in SCI.
(notes by Chi-Bin)
