How do i get started with the implementation of Label propagation (Multi Atlas based Image segmentation) using sparse coding? The same has been implemented in the paper Conference Paper Representation Learning: A Unified Deep Learning Framework f...
Are there any resources that I can refer to that have implemented this approach? Can somebody guide me on this.
Hello Aparna Sathya Murthy
Thank you for the reply. I did go through the research paper you mentioned. I found the explanation complicated and also, the paper seems to implement a different method. I am interested in implementing Label propagation using sparse coding like the one implemented in the paper Conference Paper Representation Learning: A Unified Deep Learning Framework f...
and a basic approach would be helpful for me to get started as I am very new to this topic.
Regarding the website nitrc, its very resourceful. Thank you for recommending it. Also, I wasn't able to find a similar method in that webite. Could you maybe provide a link to it?
Hello Aparna Sathya Murthy https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3939619/ This is the link to the original paper.
The ASETS tool, is it for label propagation 3D segmentation?
Hello Aparna Sathya Murthy
So far I've been able to extract the patches and features from those patches. My next step is to implement Label Propagation which is an atlas based Image segmentation method (different from the scikit learn label propagation). I'm currently stuck at this step.
Hello Aparna Sathya Murthy
I have already implemented equation 1. I'm stuck at eqn 2 and 3.
Yes. I have the labelled data.
Hello Aparna Sathya Murthy ,
I'm not sure how to incorporate the features extracted from stacked ISA into the wt(x,y). My feature map obtained is of size (nxf) where n is the number of patches and f is the number of components required to describe the patch. So that means I have 200 feature components for each patch. But in equation 2 and 3, it requires feature signature for each voxel. Whereas I have a feature map of size 200 representing an entire patch.
Hello Aparna Sathya Murthy ,
Each entry of nxf cannot be w(x,y) since w(x,y) considers the weight for a voxel and its neighbouring voxel. The feature map output is for an entire patch. Also, the feature map output cannot be reshaped as its a stacked hierarchical network with pca in between. So assuming my input patch is of size 32*32*10, number of patches is 100 and number of components in the output is set to 200 . Then after the stacked ISA network along with the PCA, the output returned (feature map) is of size 100*200. so for each patch the output is of size 1*200 which cannot be reshaped to input patch size.
Also, the ground truth vector is the same size of input image. it containes label for each voxel.
- Badges
- Science topic