def my_dePierroMap(image, obj_fun, beta, filter, num_subsets,
num_subiterations, weights, sensitivity_image):
# Check that weights are normalised
if (np.abs(np.sum(weights, axis=1) - 1) > 1.0e-6).any():
raise ValueError("Weights should sum to 1 for each voxel")
# Create OSEM reconstructor
OSEM_reconstructor = pet.OSMAPOSLReconstructor()
OSEM_reconstructor.set_output_filename_prefix('subiter')
OSEM_reconstructor.set_objective_function(obj_fun)
OSEM_reconstructor.set_num_subsets(num_subsets)
OSEM_reconstructor.set_num_subiterations(num_subiterations)
OSEM_reconstructor.set_up(image)
current_image = image.clone()
for iter in range(1, num_subiterations + 1):
print('\n------------- Subiteration %d' % iter)
# Calculate imageReg and return as an array
imageReg_array = dePierroReg(current_image.as_array(), weights)
# OSEM image update
OSEM_reconstructor.update(current_image)
imageEM_array = current_image.as_array()
# Final image update
imageUpdated_array = dePierroUpdate \
(imageEM_array, imageReg_array, beta, sensitivity_image.as_array())
# Fill image and truncate to cylindrical field of view
current_image.fill(imageUpdated_array)
filter.apply(current_image)
image_out = current_image.clone()
return image_out
imshow(acquisition_array[0, 0, :, :, ], [], 'Forward projection')
#%% close all plots
plt.close('all')
#%% create objective function
obj_fun = pet.make_Poisson_loglikelihood(acquired_data)
# We could set acquisition model but the default (ray-tracing) is in this case ok
# obj_fun.set_acquisition_model(am)
#obj_fun.set_prior(prior)
#%% create OSMAPOSL reconstructor
# This implements the Ordered Subsets Maximum A-Posteriori One Step Late
# Since we are not using a penalty, or prior in this example, it
# defaults to using MLEM, but we will modify it to OSEM
recon = pet.OSMAPOSLReconstructor()
recon.set_objective_function(obj_fun)
recon.set_num_subsets(4)
num_iters = 10
recon.set_num_subiterations(num_iters)
#%% create initial image
# we could just use a uniform image but here we will create a disk with a different
# initial value (this will help the display later on)
init_image = image.clone()
init_image.fill(cmax / 4)
make_cylindrical_FOV(init_image)
# display
idata = init_image.as_array()
slice_num = idata.shape[0] // 2
plt.figure()
imshow(idata[slice_num, :, :], [0, cmax], 'initial image')