def test_3d_shepp_logan_recon(self):
# Set threshold for test
threshold = 0.05
# Set simulation and phantom parameters
num_rows = 256
num_cols = num_rows - 1 # Test for nasty case of columns = rows-1
num_views = 144
num_slices = 8
# Reconstruction parameters
T = 0.1
p = 1.1
sharpness = 0.0
snr_db = 40.0
try:
phantom = svmbir.phantom.gen_shepp_logan_3d(
num_rows, num_cols, num_slices)
except Exception as e:
print(e)
assert 0
angles = np.linspace(-np.pi / 2.0,
np.pi / 2.0,
num_views,
endpoint=False)
try:
sino = svmbir.project(angles, phantom, max(num_rows, num_cols))
except Exception as e:
print(e)
assert 0
(num_views, num_slices, num_channels) = sino.shape
try:
recon = svmbir.recon(sino,
angles,
num_rows=num_rows,
num_cols=num_cols,
T=T,
p=p,
sharpness=sharpness,
snr_db=snr_db,
verbose=0)
except Exception as e:
print(e)
assert 0
# Compute normalized root mean squared error of reconstruction
nrmse = svmbir.phantom.nrmse(recon, phantom)
assert nrmse <= threshold
num_slices)
# Generate array of view angles form -180 to 180 degs
angles = np.linspace(-tilt_angle, tilt_angle, num_views)
# Generate sinogram by projecting phantom
sino = svmbir.project(angles, phantom, max(num_rows, num_cols))
# Determine resulting number of views, slices, and channels
(num_views, num_slices, num_channels) = sino.shape
# Perform fixed resolution MBIR reconstruction
recon = svmbir.recon(sino,
angles,
num_rows=num_rows,
num_cols=num_cols,
T=T,
p=p,
sharpness=sharpness,
snr_db=snr_db)
# Perform multi-resolution MBIR reconstruction
mr_recon = svmbir.recon(sino,
angles,
num_rows=num_rows,
num_cols=num_cols,
T=T,
p=p,
max_resolutions=max_resolutions,
sharpness=sharpness,
snr_db=snr_db)
phantom = svmbir.phantom.gen_shepp_logan_3d(num_rows_cols, num_rows_cols,
num_slices)
# Generate array of view angles form -180 to 180 degs
angles = np.linspace(-tilt_angle, tilt_angle, num_views, endpoint=False)
# Generate sinogram by projecting phantom
sino = svmbir.project(angles, phantom, num_rows_cols)
# Determine resulting number of views, slices, and channels
(num_views, num_slices, num_channels) = sino.shape
# Perform fixed resolution MBIR reconstruction
recon = svmbir.recon(sino,
angles,
T=T,
p=p,
sharpness=sharpness,
snr_db=snr_db)
# Compute Normalized Root Mean Squared Error
nrmse = svmbir.phantom.nrmse(recon, phantom)
# create output folder
os.makedirs('output', exist_ok=True)
# display phantom
title = f'Slice {display_slice:d} of 3D Shepp Logan Phantom.'
plot_image(phantom[display_slice],
title=title,
filename='output/3d_shepp_logan_phantom.png',
vmin=vmin,
def test_multires_microscopy(self):
# Set threshold for test
threshold = 0.05
# Set simulation and phantom parameters
num_rows = 256
num_cols = 64
num_slices = 16
# Simulated sinogram parameters
num_views = 64
tilt_angle = np.pi / 2.3 # Tilt range of +-60deg
# Reconstruction parameters
sharpness = 2.0
T = 0.25
snr_db = 30.0
p = 1.2
# Display parameters
vmin = 0.0
vmax = 1.1
# Generate phantom
phantom = svmbir.phantom.gen_microscopy_sample_3d(
num_rows, num_cols, num_slices)
# Generate array of view angles form -180 to 180 degs
angles = np.linspace(-tilt_angle, tilt_angle, num_views)
# Generate sinogram by projecting phantom
sino = svmbir.project(angles, phantom, max(num_rows, num_cols))
# Determine resulting number of views, slices, and channels
(num_views, num_slices, num_channels) = sino.shape
# Perform fixed resolution MBIR reconstruction
recon = svmbir.recon(sino,
angles,
num_rows=num_rows,
num_cols=num_cols,
max_resolutions=0,
T=T,
p=p,
sharpness=sharpness,
snr_db=snr_db,
stop_threshold=0.03,
verbose=0)
# Perform default MBIR reconstruction
mr_recon = svmbir.recon(sino,
angles,
num_rows=num_rows,
num_cols=num_cols,
max_resolutions=2,
T=T,
p=p,
sharpness=sharpness,
snr_db=snr_db,
stop_threshold=0.03,
verbose=0)
# Compute Normalized Root Mean Squared Error
nrmse = svmbir.phantom.nrmse(mr_recon, recon)
assert nrmse <= threshold
angles = np.linspace(-tilt_angle, tilt_angle, num_views, endpoint=False)
# Generate sinogram by projecting phantom
sino = svmbir.project(angles, phantom, num_rows_cols)
# Determine resulting number of views, slices, and channels
(num_views, num_slices, num_channels) = sino.shape
# Rotate image to use as input to proximal map
phantom_rot = np.swapaxes(phantom, 1, 2)
# Perform fixed resolution MBIR reconstruction using proximal map input
recon = svmbir.recon(sino,
angles,
max_resolutions=max_resolutions,
init_image=phantom_rot,
prox_image=phantom_rot,
positivity=False,
sigma_x=sigma_x,
snr_db=snr_db)
# create output folder
os.makedirs('output', exist_ok=True)
# display phantom
title = f'Slice {display_slice:d} of 3D Shepp Logan Phantom.'
plot_image(phantom[display_slice],
title=title,
filename='output/prox_phantom.png',
vmin=vmin,
vmax=vmax)
