Week #12

The GitHub repository for UNet practise
I reviewed the course CE7454_2018 to correct the errors in my code and find out why the loss doesn’t descend.
As we know that for the image segmentation, the goal is to get the class label for every pixel of the input image. For the BMCC dataset, the mask is size: 1024x1024 2D tensor, image is size: 1024x1024x3 3D tensor. After finding the unique colour in the mask, I define a mapping for the mask, which indicate the class for every pixel, in my case, there are 3 classes, i.e. 0,1,2.
GitHub repository for MRIImage_segmentation

For the MRI image segmentation, the FreeSurfer is used as the first step to:

• Estimate the brain mask
• Extract cortical brain surfaces
• Get a first segmentation of the brain tissue(CSF, White Matter(VM), Gray Matter(GM))
• Get a first low scale (big regions) parcellation of the cortial gray Matter (Desikan-Killiany atlas)