I suggest you use Theano and write your own code. You can refer to the attached github project, which works on video classification. You can use the same code and modify it to suit your 3D volumes. Tried and tested - it works!  

https://github.com/lpigou/Theano-3D-ConvNet

Dr. Dan Ciresan at Dalle Molle Institute for Artificial Intelligence has authored a couple of papers on deep learning for image segmentation. I'm not sure it has been applied directly to 3D in the past, but I know it has been used in 2D multi-slice segmentation applications. His page for the software is linked below. Two of his publications for this method are:

[1] D. Ciresan et al. - Deep, Big, Simple Neural Nets for Handwritten Digit Recognition, Neural Computation, 2010

[2] D. Ciresan et al. - Flexible, High Performance Convolutional Neural Networks for Image Classification, IJCAI, 2011

http://people.idsia.ch/~ciresan/academic.html

Dear Cory,

Thanks  for  help ,these  articles  seem useful especially  I'm  interested  with CNN.

Regards. 

Shaima. 

I suggest you use Theano and write your own code. You can refer to the attached github project, which works on video classification. You can use the same code and modify it to suit your 3D volumes. Tried and tested - it works!  

https://github.com/lpigou/Theano-3D-ConvNet

For now, it seems like the available open-source options are Caffe (https://github.com/BVLC/caffe/pull/2049) and Theano+3D-ConvNet+Pylearn2.  The usage of configuration for training in Pylearn2 is pretty neat.

@Suthirth, thanks for providing the link to the 3D-CNN library. Would you be willing to share us an example for using Theano-3D-ConvNet?

yeah..any example??

Maybe this library could be useful? https://github.com/Kamnitsask/deepmedic