Sanity Check for DICOM HDF5

Let's go through the HDF5 file we created with the DICOM Processing Script notebook.

We want to make sure that neon can correctly load the data from the HDF5 file.



In [1]:

    
from neon.data import HDF5Iterator  # Neon's HDF5 data loader
from neon.backends import gen_backend



In [2]:

    
be = gen_backend(backend='cpu', batch_size=1)



In [3]:

    
outFilename = 'dicom_out.h5'  # The name of our HDF5 data file



In [4]:

    
train_set = HDF5Iterator(outFilename)



In [5]:

    
train_set.get_description()









    Out[5]:





{'config': {'hdf_filename': 'dicom_out.h5', 'name': 'HDF5Iterator_0'},
 'type': 'neon.data.hdf5iterator.HDF5Iterator'}



In [6]:

    
train_set.ndata   # Number of patients in our dataset









    Out[6]:





5



In [7]:

    
train_set.lshape   # DICOM image tensor (C x H x W x D)









    Out[7]:





(1, 512, 512, 128)



In [8]:

    
from matplotlib import pyplot as plt, cm

%matplotlib inline

Test what the tensor looks like to the backend

Evren confirmed that I should be iterating through the train_set so that it will show me how the tensor will look to the backend.

This looks right.



In [15]:

    
i = 0
plt.figure(figsize=(40,40))

for x, t in train_set:
    
    print(x)
    plt.subplot(train_set.ndata,1,i+1)
    
    # Print out slice #100 for eah patient
    plt.title('Patient #{}, Slice #{}'.format(i, 100))
    plt.imshow(x.get().reshape(512,512,128)[:,:,100], cmap=cm.bone); 
    
    i += 1









    



CPUTensor(base 0x10b8693f0) name:None shape:(33554432, 1) dtype:float32 strides:(4, 4) is_c_contiguous:True
CPUTensor(base 0x10b8693f0) name:None shape:(33554432, 1) dtype:float32 strides:(4, 4) is_c_contiguous:True
CPUTensor(base 0x10b8693f0) name:None shape:(33554432, 1) dtype:float32 strides:(4, 4) is_c_contiguous:True
CPUTensor(base 0x10b8693f0) name:None shape:(33554432, 1) dtype:float32 strides:(4, 4) is_c_contiguous:True
CPUTensor(base 0x10b8693f0) name:None shape:(33554432, 1) dtype:float32 strides:(4, 4) is_c_contiguous:True

Interactive display

Use the slider to go through the slices for patient 0.



In [10]:

    
from ipywidgets import interact



In [16]:

    
def displaySlice(sliceNo):
    plt.figure(figsize=[10,10]);
    plt.title('Patient #0, Slice #{}'.format(sliceNo));
    plt.imshow(train_set.inp[0,:].reshape(512,512,128)[:,:,sliceNo-1], cmap=cm.bone); 
    plt.show()
    
interact(displaySlice, sliceNo=(1,128,1));



In [ ]: