I used the MRI dataset provided by Left Atrial Segmentation Challenge 2013which was part of the STACOM’13 workshop
this is the data I used
test_data
training_data
I am suppose to segment the left atrium wall (epicrdium and endocardium ) using deep learning approach and geodesic active contour
So I did segment the epicardium with the deep learning approach and I got my generated labels .
the result of the first segmentation should be the input for the GeodesicactiveContour to perform the segmentation of the epicardium ! I am a bit confused how should I give it as the input ?
my work is similar to this task
shuman Jia et al
Could anyone help me please ?
Did you take a look at the example? It uses only one seed point, but you can use all the voxels of your segmentation as seeds. Or maybe erode your mask and use the remainder as sees.
I don’t like level sets, because it is hard to tweak their parameters to work well.
yes in the example it uses a region growing approach ‘flood fill’ as the first approach and then the result of it’s segmentation is used as the initial contour for the geodesic active contour
but I am using a deep learning approach and I have my segmentation result
my problem is how to provide my results to the geodesic active contour to use it as the initial contour and reache the outer boundary ???
Remove fast marching filter and replace the line
geodesicActiveContour->SetInput(fastMarching->GetOutput());
by something like
geodesicActiveContour->SetInput(segmentationFromDeepLearning);
thank you for the reply ! from what I understand there has to be a learning phase first or how can the contour stop evolving when it reaches th borders of the object ?
You pre-set number of iterations, and tune it to your problem.