def _visualize_outputs(c_img_recons, c_img_targets, smoothing_factor=8):
image_recons = complex_abs(c_img_recons)
image_targets = complex_abs(c_img_targets)
kspace_recons = make_k_grid(fft2(c_img_recons), smoothing_factor)
kspace_targets = make_k_grid(fft2(c_img_targets), smoothing_factor)
image_recons, image_targets, image_deltas = make_grid_triplet(
image_recons, image_targets)
return kspace_recons, kspace_targets, image_recons, image_targets, image_deltas
def forward(self, cmg_output, targets, extra_params):
if cmg_output.size(0) > 1:
raise NotImplementedError('Only one at a time for now.')
cmg_target = targets['cmg_targets']
cmg_recon = nchw_to_kspace(cmg_output)
assert cmg_recon.shape == cmg_target.shape, 'Reconstruction and target sizes are different.'
assert (cmg_recon.size(-3) % 2 == 0) and (cmg_recon.size(-2) % 2 == 0), \
'Not impossible but not expected to have sides with odd lengths.'
if self.residual_acs: # Adding the semi-k-space of the ACS as a residual. Necessary due to complex cropping.
raise NotImplementedError('Not ready yet.')
# cmg_acs = targets['cmg_acss']
# cmg_recon = cmg_recon + cmg_acs
kspace_recon = fft2(cmg_recon)
img_recon = complex_abs(cmg_recon)
recons = {
'kspace_recons': kspace_recon,
'cmg_recons': cmg_recon,
'img_recons': img_recon
}
if self.challenge == 'multicoil':
rss_recon = center_crop(img_recon, (
self.resolution, self.resolution)) * extra_params['cmg_scales']
rss_recon = root_sum_of_squares(rss_recon, dim=1).squeeze()
recons['rss_recons'] = rss_recon
return recons # recons are not rescaled except rss_recons.
def forward(self, outputs, targets, extra_params):
img_recon, phase_recon = outputs
if img_recon.size(0) > 1:
raise NotImplementedError('Only one at a time for now.')
img_target = targets['img_targets']
img_recon = F.relu(img_recon) # img_recons must be positive numbers.
assert img_recon.shape == img_target.shape, 'Reconstruction and target sizes are different.'
# Input transform had addition of pi as pre-processing.
phase_recon = phase_recon - math.pi # No clamping implemented since the loss is MSE.
cmg_recon = torch.stack([
img_recon * torch.cos(phase_recon),
img_recon * torch.sin(phase_recon)
],
dim=-1)
kspace_recon = fft2(cmg_recon)
recons = {
'img_recons': img_recon,
'phase_recons': phase_recon,
'cmg_recons': cmg_recon,
'kspace_recons': kspace_recon
}
if self.challenge == 'multicoil':
rss_recon = center_crop(
img_recon,
shape=(self.resolution,
self.resolution)) * extra_params['img_scales']
rss_recon = root_sum_of_squares(rss_recon, dim=1).squeeze()
recons['rss_recons'] = rss_recon
return recons # recons are not rescaled except rss_recons.
def test_fft2(shape):
shape = shape + [2]
tensor = create_tensor(shape)
out_torch = data_transforms.fft2(tensor).numpy()
out_torch = out_torch[..., 0] + 1j * out_torch[..., 1]
tensor_numpy = data_transforms.tensor_to_complex_np(tensor)
tensor_numpy = np.fft.ifftshift(tensor_numpy, (-2, -1))
out_numpy = np.fft.fft2(tensor_numpy, norm='ortho')
out_numpy = np.fft.fftshift(out_numpy, (-2, -1))
assert np.allclose(out_torch, out_numpy)
def _cmg_output(cmg_output, targets, extra_params):
cmg_target = targets['cmg_targets']
cmg_recon = nchw_to_kspace(
cmg_output) # Assumes data was cropped already.
assert cmg_recon.shape == cmg_target.shape, 'Reconstruction and target sizes are different.'
assert (cmg_recon.size(-3) % 2 == 0) and (cmg_recon.size(-2) % 2 == 0), \
'Not impossible but not expected to have sides with odd lengths.'
cmg_recon = cmg_recon + targets[
'cmg_inputs'] # Residual of complex input.
kspace_recon = fft2(cmg_recon)
img_recon = complex_abs(cmg_recon)
recons = {
'kspace_recons': kspace_recon,
'cmg_recons': cmg_recon,
'img_recons': img_recon
}
return recons
def forward(self, cmg_output: Tensor, targets: dict, extra_params: dict):
assert cmg_output.dim() == 5 and cmg_output.size(
1) == 2, 'Invalid shape!'
if cmg_output.size(0) > 1:
raise NotImplementedError('Only one at a time for now.')
kspace_target = targets['kspace_targets']
cmg_recon = cmg_output.permute(dims=(0, 2, 3, 4,
1)) # Convert back into NCHW2
if cmg_recon.shape != kspace_target.shape: # Cropping recon left-right.
left = (cmg_recon.size(-2) - kspace_target.size(-2)) // 2
cmg_recon = cmg_recon[..., left:left + kspace_target.size(-2), :]
assert cmg_recon.shape == kspace_target.shape, 'Reconstruction and target sizes are different.'
assert (cmg_recon.size(-3) % 2 == 0) and (cmg_recon.size(-2) % 2 == 0), \
'Not impossible but not expected to have sides with odd lengths.'
kspace_recon = fft2(cmg_recon)
if self.replace_kspace:
mask = extra_params['masks']
kspace_recon = kspace_target * mask + (1 - mask) * kspace_recon
cmg_recon = ifft2(kspace_recon)
img_recon = complex_abs(cmg_recon)
# recons = {'kspace_recons': kspace_recon, 'cmg_recons': cmg_recon, 'img_recons': img_recon}
recons = dict()
if self.challenge == 'multicoil':
rss_recon = center_crop(img_recon, (
self.resolution, self.resolution)) * extra_params['cmg_scales']
rss_recon = root_sum_of_squares(rss_recon, dim=1).squeeze()
recons['rss_recons'] = rss_recon
return recons # recons are not rescaled except rss_recons.
import torch import numpy as np from data.data_transforms import ifft2, fft2 a = torch.rand(20, 40, 60, 92, 2, device='cuda:1') b = ifft2(a * 1E8) c = fft2(b) * 1E-8 # print(torch.all(a == c)) # print(torch.allclose(a, c, rtol=0.01)) eps = np.finfo(np.float64).eps print(torch.max(c / (a + eps))) print(torch.min(c / (a + eps))) print(torch.mean(c / (a + eps)).cpu().numpy()) # print(torch.sum(a != c) / (20 * 40 * 60 * 92 * 2))
def check_invertible():
orig = torch.rand(4, 6, 8, 12, 2, dtype=torch.float64) * 1024 - 64
trans = fft2(orig) * 100
trans = ifft2(trans) / 100
print(torch.allclose(orig, trans))
import torch from data.data_transforms import ifft2, fft2, complex_abs image = torch.rand(10, 20, 30, 2) lr_flip = torch.flip(image, dims=[-2]) ud_flip = torch.flip(image, dims=[-3]) all_flip = torch.flip(image, dims=[-3, -2]) kspace = fft2(image) lr_kspace = fft2(lr_flip) ud_kspace = fft2(ud_flip) all_kspace = fft2(all_flip) absolute = torch.sum(complex_abs(kspace)) lr_abs = torch.sum(complex_abs(lr_kspace)) ud_abs = torch.sum(complex_abs(ud_kspace)) all_abs = torch.sum(complex_abs(all_kspace)) a = torch.allclose(absolute, lr_abs) b = torch.allclose(absolute, ud_abs) c = torch.allclose(absolute, all_abs) print(a, b, c)
def __call__(self, kspace_target, target, attrs, file_name, slice_num):
assert isinstance(
kspace_target, torch.Tensor
), 'k-space target was expected to be a Pytorch Tensor.'
if kspace_target.dim(
) == 3: # If the collate function does not expand dimensions for single-coil.
kspace_target = kspace_target.expand(1, 1, -1, -1, -1)
elif kspace_target.dim(
) == 4: # If the collate function does not expand dimensions for multi-coil.
kspace_target = kspace_target.expand(1, -1, -1, -1, -1)
elif kspace_target.dim(
) != 5: # Expanded k-space should have 5 dimensions.
raise RuntimeError('k-space target has invalid shape!')
if kspace_target.size(0) != 1:
raise NotImplementedError('Batch size should be 1 for now.')
with torch.no_grad():
seed = None if not self.use_seed else tuple(map(ord, file_name))
masked_kspace, mask, info = apply_info_mask(
kspace_target, self.mask_func, seed)
num_low_freqs = info['num_low_frequency']
acs_mask = self.find_acs_mask(kspace_target, num_low_freqs)
acs_kspace = kspace_target * acs_mask
semi_kspace_acs = fft1(complex_center_crop(
ifft2(acs_kspace), shape=(self.resolution, self.resolution)),
direction='width')
complex_image = ifft2(masked_kspace)
complex_image = complex_center_crop(complex_image,
shape=(self.resolution,
self.resolution))
# img_input is not actually an input but what the input would look like in the image domain.
img_input = complex_abs(complex_image)
# Direction is fixed due to challenge conditions.
semi_kspace = fft1(complex_image, direction='width')
weighting = self.weight_func(semi_kspace)
semi_kspace *= weighting
sk_scale = torch.std(semi_kspace)
semi_kspace /= sk_scale
inputs = kspace_to_nchw(semi_kspace)
extra_params = {
'sk_scales': sk_scale,
'masks': mask,
'weightings': weighting
}
extra_params.update(info)
extra_params.update(attrs)
# Recall that the Fourier transform is a linear transform.
cmg_target = ifft2(kspace_target)
cmg_target = complex_center_crop(cmg_target,
shape=(self.resolution,
self.resolution))
cmg_target /= sk_scale
img_target = complex_abs(cmg_target)
semi_kspace_target = fft1(cmg_target, direction='width')
kspace_target = fft2(cmg_target)
# Use plurals as keys to reduce confusion.
targets = {
'semi_kspace_targets': semi_kspace_target,
'kspace_targets': kspace_target,
'cmg_targets': cmg_target,
'img_targets': img_target,
'img_inputs': img_input,
'semi_kspace_acss': semi_kspace_acs
}
if self.challenge == 'multicoil':
targets['rss_targets'] = target
return inputs, targets, extra_params
def __call__(self, kspace_target, target, attrs, file_name, slice_num):
assert isinstance(
kspace_target, torch.Tensor
), 'k-space target was expected to be a Pytorch Tensor.'
if kspace_target.dim(
) == 3: # If the collate function does not expand dimensions for single-coil.
kspace_target = kspace_target.expand(1, 1, -1, -1, -1)
elif kspace_target.dim(
) == 4: # If the collate function does not expand dimensions for multi-coil.
kspace_target = kspace_target.expand(1, -1, -1, -1, -1)
elif kspace_target.dim(
) != 5: # Expanded k-space should have 5 dimensions.
raise RuntimeError('k-space target has invalid shape!')
if kspace_target.size(0) != 1:
raise NotImplementedError('Batch size should be 1 for now.')
with torch.no_grad():
# Apply mask
seed = None if not self.use_seed else tuple(map(ord, file_name))
masked_kspace, mask, info = apply_info_mask(
kspace_target, self.mask_func, seed)
# Complex image made from down-sampled k-space.
complex_image = ifft2(masked_kspace)
if self.crop_center:
complex_image = complex_center_crop(complex_image,
shape=(self.resolution,
self.resolution))
cmg_scale = torch.std(complex_image)
complex_image /= cmg_scale
extra_params = {'cmg_scales': cmg_scale, 'masks': mask}
extra_params.update(info)
extra_params.update(attrs)
# Recall that the Fourier transform is a linear transform.
kspace_target /= cmg_scale
cmg_target = ifft2(kspace_target)
if self.crop_center:
cmg_target = complex_center_crop(cmg_target,
shape=(self.resolution,
self.resolution))
# Data augmentation by flipping images up-down and left-right.
if self.augment_data: # No rotation implemented.
flip_lr = torch.rand(()) < 0.5
flip_ud = torch.rand(()) < 0.5
if flip_lr and flip_ud:
# Last dim is real/complex dimension for complex image and target.
complex_image = torch.flip(complex_image, dims=(-3, -2))
cmg_target = torch.flip(cmg_target, dims=(-3, -2))
target = torch.flip(target, dims=(
-2, -1)) # Has only two dimensions, height and width.
kspace_target = fft2(cmg_target)
elif flip_ud:
complex_image = torch.flip(complex_image, dims=(-3, ))
cmg_target = torch.flip(cmg_target, dims=(-3, ))
target = torch.flip(target, dims=(-2, ))
kspace_target = fft2(cmg_target)
elif flip_lr:
complex_image = torch.flip(complex_image, dims=(-2, ))
cmg_target = torch.flip(cmg_target, dims=(-2, ))
target = torch.flip(target, dims=(-1, ))
kspace_target = fft2(cmg_target)
# The image target is obtained after flipping the complex image.
# This removes the need to flip the image target.
img_target = complex_abs(cmg_target)
img_inputs = complex_abs(complex_image)
# Use plurals as keys to reduce confusion.
targets = {
'kspace_targets': kspace_target,
'cmg_targets': cmg_target,
'img_targets': img_target,
'cmg_inputs': complex_image,
'img_inputs': img_inputs
}
if self.challenge == 'multicoil':
targets['rss_targets'] = target
# Creating concatenated image of real/imag/abs channels.
concat_image = torch.cat(
[complex_image, img_inputs.unsqueeze(dim=-1)], dim=-1)
# Converting to NCHW format for CNN.
inputs = kspace_to_nchw(concat_image)
return inputs, targets, extra_params
def __call__(self, kspace_target, target, attrs, file_name, slice_num):
assert isinstance(
kspace_target, torch.Tensor
), 'k-space target was expected to be a Pytorch Tensor.'
if kspace_target.dim(
) == 3: # If the collate function does not expand dimensions for single-coil.
kspace_target = kspace_target.expand(1, 1, -1, -1, -1)
elif kspace_target.dim(
) == 4: # If the collate function does not expand dimensions for multi-coil.
kspace_target = kspace_target.expand(1, -1, -1, -1, -1)
elif kspace_target.dim(
) != 5: # Expanded k-space should have 5 dimensions.
raise RuntimeError('k-space target has invalid shape!')
if kspace_target.size(0) != 1:
raise NotImplementedError('Batch size should be 1 for now.')
with torch.no_grad():
# Apply mask
seed = None if not self.use_seed else tuple(map(ord, file_name))
masked_kspace, mask, info = apply_info_mask(
kspace_target, self.mask_func, seed)
complex_image = ifft2(masked_kspace)
cmg_target = ifft2(kspace_target)
if self.crop_center:
complex_image = complex_center_crop(complex_image,
shape=(self.resolution,
self.resolution))
cmg_target = complex_center_crop(cmg_target,
shape=(self.resolution,
self.resolution))
# Data augmentation by flipping images up-down and left-right.
if self.augment_data:
flip_lr = torch.rand(()) < 0.5
flip_ud = torch.rand(()) < 0.5
if flip_lr and flip_ud:
# Last dim is real/complex dimension for complex image and target.
complex_image = torch.flip(complex_image, dims=(-3, -2))
cmg_target = torch.flip(cmg_target, dims=(-3, -2))
target = torch.flip(target, dims=(
-2, -1)) # Has only two dimensions, height and width.
elif flip_ud:
complex_image = torch.flip(complex_image, dims=(-3, ))
cmg_target = torch.flip(cmg_target, dims=(-3, ))
target = torch.flip(target, dims=(-2, ))
elif flip_lr:
complex_image = torch.flip(complex_image, dims=(-2, ))
cmg_target = torch.flip(cmg_target, dims=(-2, ))
target = torch.flip(target, dims=(-1, ))
# Adding pi to angles so that the phase is in the [0, 2pi] range for better learning.
phase_input = torch.atan2(complex_image[..., 1], complex_image[...,
0])
phase_input += math.pi # Don't forget to remove the pi in the output transform!
img_input = complex_abs(complex_image)
img_scale = torch.std(img_input)
img_input /= img_scale
cmg_target /= img_scale
img_target = complex_abs(cmg_target)
kspace_target = fft2(
cmg_target
) # Reconstruct k-space target after cropping and image augmentation.
phase_target = torch.atan2(cmg_target[..., 1], cmg_target[..., 0])
extra_params = {'img_scales': img_scale, 'masks': mask}
extra_params.update(info)
extra_params.update(attrs)
# Use plurals as keys to reduce confusion.
targets = {
'kspace_targets': kspace_target,
'cmg_targets': cmg_target,
'img_targets': img_target,
'phase_targets': phase_target,
'img_inputs': img_input
}
if self.challenge == 'multicoil':
targets['rss_targets'] = target
# Converting to NCHW format for CNN. Also adding phase input.
inputs = (img_input, phase_input)
return inputs, targets, extra_params