DICOM-CT-PD (3rd Generation CT)#

[1]:

from __future__ import annotations
from pytomography.metadata import ObjectMeta
from pytomography.algorithms import SART
from pytomography.io.CT import dicom_ct_pd
from pytomography.projectors.CT import CTGen3SystemMatrix
import matplotlib.pyplot as plt
import os

First, lets specify the paths of all the projection data:

[2]:

save_path = '/disk1/ct_proj/DICOM-CT-PD_FD' # Path where you downloaded the CT-PD data
paths = [os.path.join(save_path, file) for file in os.listdir(save_path) if '.dcm' in file]

We can load the projections and corresponding metadata as follows

This may take a couple minutes as the dataset is huge

[3]:

proj, proj_meta = dicom_ct_pd.get_projections_and_metadata_gen3(paths)

We can view some of the projections:

[4]:

plt.subplots(5,1,figsize=(5,5))
plt.subplot(511)
plt.pcolormesh(proj[0].cpu().T, cmap='Greys_r')
plt.axis('off')
plt.subplot(512)
plt.pcolormesh(proj[1400].cpu().T, cmap='Greys_r')
plt.axis('off')
plt.subplot(513)
plt.pcolormesh(proj[4800].cpu().T, cmap='Greys_r')
plt.axis('off')
plt.subplot(514)
plt.pcolormesh(proj[6400].cpu().T, cmap='Greys_r')
plt.axis('off')
plt.subplot(515)
plt.pcolormesh(proj[8800].cpu().T, cmap='Greys_r')
plt.axis('off')
plt.show()

Now lets set up our object space metadata

[5]:

dx = dy = dz =  1.0 # mm
Nx = Ny = Nz = 512
object_meta = ObjectMeta(dr=(dx,dx,dx), shape=(Nx,Ny,Nz))

We can create the system matrix as follows. The N_splits argument can save on memory if you are using a small number of subsets or have little GPU RAM. The device='cpu' means that the projections are expected on CPU (the internal projector code will still use pytomography.device and transfer the projections to GPU if needed)

[6]:

system_matrix = CTGen3SystemMatrix(object_meta, proj_meta, N_splits=1, device='cpu')

Here we reconstruct using SART with 3 iterations and 40 subsets (this takes around 10 mins)

[7]:

recon_algorithm = SART(system_matrix, proj)
recon = recon_algorithm(n_iters=3, n_subsets=40)

Now we can look at an axial slice:

[8]:

plt.figure(figsize=(5.5,5))
plt.imshow(recon[:,255].cpu().numpy().T * 10, cmap='Greys_r', vmin=0.15, vmax=0.26, interpolation='gaussian', origin='lower')
plt.ylim(10,500)
plt.axis('off')
plt.colorbar(label='cm$^{-1}$')
plt.title('PyTomography: OS-SART(3it40ss)')

If anyone wants to implement FBP for the geometry of this scanner, please do and submit a pull request! (see the CTConeBeamFlatPanelSystemMatrix backward method for how filtered back projection should be implemented)
If anyone wants to write a DICOM exporter for these CT images, please do, and submit a pull request!

DICOM-CT-PD (3rd Generation CT)#

This Page