Dense Design Matrix of volumetric Images (m*n*p)

[VIVEK RUHELA]

Hello guys,
                Presently I am playing with volumetric images and using pylearn2 for convnets. But I am facing some problems in the design of dense_design_matrix.

Let me explain my problem:

I have an Patient MRI consisting of four volumetric Images i.e. T1, T2, T1C, Flair. Size of all four volumetric images is (160,216,176). Labels are given pixel wise i.e. an volumetric image of same size as T1 or T2 etc. There are such 30 patients MRI with five images(T1, T2, T1C, Flair,Label)

As per the pylearn2 tutorials, I got success for dense design matrix configuration like in MNIST or CIFAR-10 dataset. The dense design matrix for mnist dataset is [10000,28,28] ndarray where [1,28,28] is an image and in cifar10 the dense design matrix is [10000,3028] ndarray where each [32,32,3] cifar10 image is converted into flattening mode and then design matrix is configured.

So if I try for volumetric images I mentioned above how can I configure the design matrix where labels depends not on one but on four images.

First I thought for [1*numpy.prod(160,216,176)] but not successful.

[poorna....@gmail.com]

Hi Vivek ,
Please share the Pylearn2 tutorials for me..