MUSTACHE
So far, I've been working on pulling together all the pieces of MUSTACHE. Updates to that project include:
- Organizing the data so each rater has one mask for each patient, and each patient is rated by both raters
- Creating a GUI for manual raters to classify steatosis images
- Implementing histogram mean and standard deviation as features as mentioned by Lee et al. 2006
- Submitting an abstract to SPIE
Next steps for it are:
- Follow up with Soumya and Mike on incorporating different machine learning algorithms (and maybe look at SVM-Perf)
- Extend it to other fields to show generality
Speaking of which, I present...
NAUSICAA
NAUSICAA (Not Actually US Image Characterization And Analysis) is the stroke branch under the MUSTACHE project. It aims to quantify the amount of brain tissue affected from CT images, and predict the severity. Currently, I've shown that on one set of images on the third day after the stroke:
- Standard deviation of an affected region is markedly lower than normal tissue using manually drawn masks.
- Mean is not significantly different
Next steps:
- Implement some sort of sliding window to see if the stroke region can be detected automatically
- Identify additional features to discern stroke from normal tissue
- Look for indications from images of the first day to predict the impact on the third
- Extend the algorithm to 3D
- Get more images to see if observations are general
A few more things
In all my spare time, I did a literature review on texture analysis and elastography. The texture analysis gave a baseline to compare our MUSTACHE results to and ideas for features to implement. Hopefully, we'll start playing with the elastography machine this week and see what results we can get from it compared to literature.