I have a stack of brain MRI with tumor present in some slices. I wanted to remove the tumor from that slices and fill these pixels with relevant values, such that it will be like non-tumor or normal brain MRI.
I tried with opencv's image inpainiting function for this, but the results are not that good enough. Please suggest the direction that I shall follow.
Hi Gunjan,
What is your motivation for removing the tumours from these? Are they being used as a means to train some classification method? General inpainting is likely to produce the unnatural extrapolation like you are seeing. Would it be valid to simply use a mirror image of the healthy half of the brain?
Best
If you don't like to mirror the tumor cells from the other healthy side (as the image seems to be symmetric), you can seed in that location where you have the tumor cells some new pixels with preselected values. Then it will be sufficient to apply one step of 2D interpolation to get a new safe tissue of cells. If the size of your image was doubled after interpolating the image you can resample it to restore its original size.
Hi Jonathan Mason, I wanted to create a template from non-tumor brain MRI based on shape and texture, such that, it will be used to find variation with a test image with tumor present. I have thought on mirroring, but I wanted to make the method generic, such that it can be used for unsymmetrical medical images also. So, that's why I am looking for other methods. This image is not perfectly symmetric.
Abdellatif Baba I will try and let you know what is the outcome.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images