Hello,
I have created a DeepSSM model for each bone of the lumbar spine.
I now want to use it to segment CT scans that are not part of the train/validate/test dataset.
However, I am bumping against the problem of having to have my CT scans precisely aligned with the reference mesh/scan.
Is there a way to overcome this necessity?
Could you give me any suggestions?
Thank you very much,
Best wishes,
Simão
It is currently challenging with DeepSSM. The best you can do is careful cropping and centering around the target.
In the somewhat near future, we will be integrating some new changes that will make this much easier.
Hello Alan Morris,
Thank you very much for the quick response.
I will do as instructed.
I will also try to fit my DeepSSM with different cropped data to see if there is any improvement.
Could I also ask if there is a metric that can be used to know where the output of the DeepSSM model is within the shape space? Could I get a metric of where the shape lies within the distribution of shapes I used to fit the model?
Thank you very much,
Best wishes,
Simão
The output of the vanilla DeepSSM will be a combination of PCA modes from the original training data. Are you looking to determine where on each one of these modes it exists. This can be done by projecting to the eigenvectors of the shape model (which you can export from ShapeWorks Studio)