ClinicaDL CNN Image based classifie

[CAMILO ANDRES CASTRILLON CALDERON]

Hi, I chose the Conv5 FC3 structure to classify between AD, CN and MCI but this is a binary classifier. I'm searching help about how to classify with this 3 classes. 

[CAMILO ANDRES CASTRILLON CALDERON]

Although, I'm doing the interpretation with silency maps but I'm not sure what's wrong because initially, the map is plotted in a upper corner of the image (the cordinates doesn't match) but when i set the yz cordinates manually with freesurfer GUI the two images match. Any idea to fix it?

Thanks, Camilo

[Elina Thibeau-Sutre]

Hi,

Thank you for your interest in ClinicaDL!

Right now the project was only meant to perform binary classification.

I can propose a temporary version to solve your problem if you want.

Concerning the saliency maps, to have correct coordinates you may want to give an example Nifti file with correct coordinates in `--nifti_template_path` argument. This will set the affine of your saliency map to the same value than this image.

Hope this helps,

Elina

[CAMILO ANDRES CASTRILLON CALDERON]

Hi, Elina.

Thanks for your response.

If you could give me the temporary version, I appreciate it. Btw I'm already solved the plotting issue

All my best,

Camilo

[Elina Thibeau-Sutre]

You can follow advancement here.

We are preparing a new release so it will be next week...

Best regards,

Elina