CODEWITHSUNDEEP
databasedeep-learningconv-neural-network
Tariq S.
Tariq S.

Reputation: 11

Can I train a CNN on 3d images?

I am working with the 2015 BRATS Dataset of MRI Brain Tumors. The files in the dataset are in .mha format and have many slices of 2d images that make up a 3d brain. Is there a way I can train the model on these images or do I have to convert them somehow? If so, how do I convert them?

Upvotes: 1

Views: 187

Answers (1)

Nopileos
Nopileos

Reputation: 2097

A normal RGB image is already a 3 dimensional input. So you can just stack all the images and instead of maybe 3 channels you have 30 (for 10 stacked rgb images). You can process this data like you would process any other imgae.

What you can also try with this kind of data is using 3D convolutions on them. This way you have a 3 dimensional kernel like 3x3x3 (maybe even more in channel direction) and you slide it over the input in x,y and channel direction. This can help and boost the performance but will increase the runtime.

Upvotes: 2

Related Questions