def train():
args = util.get_args_shape_training()
# Params / Config
learning_rate = 1e-3
momentums_cae = (0.99, 0.999)
criterion = metrics.BatchDiceLoss([1.0]) # nn.BCELoss()
path_training_metrics = args.continuetraining # --continuetraining /share/data_zoe1/lucas/Linda_Segmentations/tmp/tmp_shape_f3.json
path_saved_model = args.caepath
channels_cae = args.channelscae
n_globals = args.globals # type(core/penu), tO_to_tA, NHISS, sex, age
resample_size = int(args.xyoriginal * args.xyresample)
pad = args.padding
pad_value = 0
leakage = 0.01
cuda = True
# CAE model
enc = Enc3DCtp(size_input_xy=resample_size, size_input_z=args.zsize,
channels=channels_cae, n_ch_global=n_globals, leakage=leakage, padding=pad)
dec = Dec3D(size_input_xy=resample_size, size_input_z=args.zsize,
channels=channels_cae, n_ch_global=n_globals, leakage=leakage)
cae = Cae3DCtp(enc, dec)
if cuda:
cae = cae.cuda()
# Model params
params = [p for p in cae.parameters() if p.requires_grad]
print('# optimizing params', sum([p.nelement() * p.requires_grad for p in params]),
'/ total: cae', sum([p.nelement() for p in cae.parameters()]))
# Optimizer with scheduler
optimizer = torch.optim.Adam(params, lr=learning_rate, weight_decay=1e-5, betas=momentums_cae)
if args.lrsteps:
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, args.lrsteps)
else:
scheduler = None
# Data
common_transform = [data.ResamplePlaneXY(args.xyresample),
data.HemisphericFlipFixedToCaseId(split_id=args.hemisflipid),
data.PadImages(pad[0], pad[1], pad[2], pad_value=pad_value)]
train_transform = common_transform + [data.ElasticDeform(), data.ToTensor()]
valid_transform = common_transform + [data.ToTensor()]
modalities = ['_CBV_reg1_downsampled', '_TTD_reg1_downsampled']
labels = ['_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled',
'_FUCT_MAP_T_Samplespace_subset_reg1_downsampled']
ds_train, ds_valid = data.get_stroke_shape_training_data(modalities, labels, train_transform, valid_transform,
args.fold, args.validsetsize, batchsize=args.batchsize)
print('Size training set:', len(ds_train.sampler.indices), 'samples | Size validation set:', len(ds_valid.sampler.indices),
'samples | Capacity batch:', args.batchsize, 'samples')
print('# training batches:', len(ds_train), '| # validation batches:', len(ds_valid))
# Training
learner = CaeReconstructionLearner(ds_train, ds_valid, cae, path_saved_model, optimizer, scheduler,
path_outputs_base=args.outbasepath)
learner.run_training()
def test(args):
# Params / Config
modalities = ['_CBV_reg1_downsampled', '_TTD_reg1_downsampled']
labels = [
'_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled',
'_FUCT_MAP_T_Samplespace_subset_reg1_downsampled'
]
normalization_hours_penumbra = args.normalize
pad = args.padding
pad_value = 0
for idx in range(len(args.path)):
# Data
transform = [
data.ResamplePlaneXY(args.xyresample),
data.PadImages(pad[0], pad[1], pad[2], pad_value=pad_value),
data.ToTensor()
]
ds_test = data.get_testdata(modalities=modalities,
labels=labels,
transform=transform,
indices=args.fold[idx])
print('Size test set:', len(ds_test.sampler.indices), '| # batches:',
len(ds_test))
# Single case evaluation
tester = CaeReconstructionTester(ds_test, args.path[idx],
args.outbasepath,
normalization_hours_penumbra)
tester.run_inference()
def test():
args = util.get_args_shape_testing()
assert len(args.fold) == len(
args.path
), 'You must provide as many --fold arguments as caepath model arguments\
in the exact same order!'
# Params / Config
modalities = ['_CBV_reg1_downsampled', '_TTD_reg1_downsampled']
labels = [
'_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled',
'_FUCT_MAP_T_Samplespace_subset_reg1_downsampled'
]
normalization_hours_penumbra = args.normalize
steps = range(6) # fixed steps for tAdmission-->tReca: 0-5 hrs
pad = args.padding
pad_value = 0
# Data
transform = [
data.ResamplePlaneXY(args.xyresample),
data.PadImages(pad[0], pad[1], pad[2], pad_value=pad_value),
data.ToTensor()
]
# Fold-wise evaluation according to fold indices and fold model for all folds and model path provided as arguments:
for i, path in enumerate(args.path):
print('Model ' + path + ' of fold ' + str(i + 1) + '/' +
str(len(args.fold)) + ' with indices: ' + str(args.fold[i]))
ds_test = data.get_testdata(modalities=modalities,
labels=labels,
transform=transform,
indices=args.fold[i])
print('Size test set:', len(ds_test.sampler.indices), '| # batches:',
len(ds_test))
# Single case evaluation for all cases in fold
tester = CaeReconstructionTesterCurve(ds_test, path, args.outbasepath,
normalization_hours_penumbra,
steps)
tester.run_inference()
def test(args):
# Params / Config
modalities = ['_CBV_reg1_downsampled', '_TTD_reg1_downsampled']
labels = ['_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled']
path_saved_model = args.unetpath
pad = args.padding
pad_value = 0
# Data
# Trained on patches, but fully convolutional approach let us apply on bigger image (thus, omit patch transform)
transform = [data.ResamplePlaneXY(args.xyresample),
data.PadImages(pad[0], pad[1], pad[2], pad_value=pad_value),
data.ToTensor()]
ds_test = data.get_testdata(modalities=modalities, labels=labels, transform=transform, indices=args.fold)
print('Size test set:', len(ds_test.sampler.indices), '| # batches:', len(ds_test))
# Single case evaluation
tester = UnetSegmentationTester(ds_test, path_saved_model, args.outbasepath, None)
tester.run_inference()
def train(args):
# Params / Config
learning_rate = 1e-3
momentums_cae = (0.9, 0.999)
weight_decay = 1e-5
criterion = metrics.BatchDiceLoss([1.0]) # nn.BCELoss()
channels_cae = args.channelscae
n_globals = args.globals # type(core/penu), tO_to_tA, NHISS, sex, age
resample_size = int(args.xyoriginal * args.xyresample)
alpha = 1.0
cuda = True
# CAE model
cae = torch.load(args.caepath)
cae.freeze(True)
enc = Enc3DStep(size_input_xy=resample_size,
size_input_z=args.zsize,
channels=channels_cae,
n_ch_global=n_globals,
alpha=alpha)
enc.encoder = cae.enc.encoder # enc.step will be trained from scratch for given shape representations
dec = cae.dec
cae = Cae3D(enc, dec)
if cuda:
cae = cae.cuda()
# Model params
params = [p for p in cae.parameters() if p.requires_grad]
print('# optimizing params',
sum([p.nelement() * p.requires_grad for p in params]),
'/ total: cae', sum([p.nelement() for p in cae.parameters()]))
# Optimizer with scheduler
optimizer = torch.optim.Adam(params,
lr=learning_rate,
weight_decay=weight_decay,
betas=momentums_cae)
if args.lrsteps:
scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer, args.lrsteps)
else:
scheduler = None
# Data
common_transform = [data.ResamplePlaneXY(args.xyresample)
] # before: FixedToCaseId(split_id=args.hemisflipid)]
train_transform = common_transform + [
data.HemisphericFlip(),
data.ElasticDeform(),
data.ToTensor()
]
valid_transform = common_transform + [data.ToTensor()]
modalities = ['_CBV_reg1_downsampled', '_TTD_reg1_downsampled'
] # dummy data only needed for visualization
labels = [
'_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled',
'_FUCT_MAP_T_Samplespace_subset_reg1_downsampled'
]
ds_train, ds_valid = data.get_stroke_shape_training_data(
modalities,
labels,
train_transform,
valid_transform,
args.fold,
args.validsetsize,
batchsize=args.batchsize)
print('Size training set:',
len(ds_train.sampler.indices), 'samples | Size validation set:',
len(ds_valid.sampler.indices), 'samples | Capacity batch:',
args.batchsize, 'samples')
print('# training batches:', len(ds_train), '| # validation batches:',
len(ds_valid))
# Training
learner = CaeStepLearner(ds_train,
ds_valid,
cae,
optimizer,
scheduler,
n_epochs=args.epochs,
path_previous_base=args.inbasepath,
path_outputs_base=args.outbasepath,
criterion=criterion)
learner.run_training()
def train():
args = util.get_args_unet_training()
# Params / Config
batchsize = 6 # 17 training, 6 validation
learning_rate = 1e-3
momentums_cae = (0.99, 0.999)
criterion = metrics.BatchDiceLoss([1.0]) # nn.BCELoss()
path_saved_model = args.unetpath
channels = args.channels
pad = args.padding
cuda = True
# Unet model
unet = Unet3D(channels)
if cuda:
unet = unet.cuda()
# Model params
params = [p for p in unet.parameters() if p.requires_grad]
print('# optimizing params',
sum([p.nelement() * p.requires_grad for p in params]),
'/ total: unet', sum([p.nelement() for p in unet.parameters()]))
# Optimizer with scheduler
optimizer = torch.optim.Adam(params,
lr=learning_rate,
weight_decay=1e-5,
betas=momentums_cae)
if args.lrsteps:
scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer, args.lrsteps)
else:
scheduler = None
# Data
train_transform = [
data.ResamplePlaneXY(args.xyresample),
data.HemisphericFlipFixedToCaseId(split_id=args.hemisflipid),
data.PadImages(pad[0], pad[1], pad[2], pad_value=0),
data.RandomPatch(104, 104, 68, pad[0], pad[1], pad[2]),
data.ToTensor()
]
valid_transform = [
data.ResamplePlaneXY(args.xyresample),
data.HemisphericFlipFixedToCaseId(split_id=args.hemisflipid),
data.PadImages(pad[0], pad[1], pad[2], pad_value=0),
data.RandomPatch(104, 104, 68, pad[0], pad[1], pad[2]),
data.ToTensor()
]
ds_train, ds_valid = data.get_stroke_shape_training_data(
train_transform,
valid_transform,
args.fold,
args.validsetsize,
batchsize=batchsize)
print('Size training set:',
len(ds_train.sampler.indices), 'samples | Size validation set:',
len(ds_valid.sampler.indices), 'samples | Capacity batch:',
batchsize, 'samples')
print('# training batches:', len(ds_train), '| # validation batches:',
len(ds_valid))
# Training
learner = UnetSegmentationLearner(ds_train,
ds_valid,
unet,
path_saved_model,
optimizer,
scheduler,
criterion,
path_previous_base=args.inbasepath,
path_outputs_base=args.outbasepath)
learner.run_training()
def infer():
args = util.get_args_sdm()
print('Evaluate validation set', args.fold)
# Params / Config
normalization_hours_penumbra = 10
#channels_unet = args.channels TODO Unet live segmentation
#pad = args.padding TODO Unet live segmentation
transform = [data.ResamplePlaneXY(args.xyresample),
data.HemisphericFlipFixedToCaseId(split_id=args.hemisflipid),
#data.PadImages(pad[0], pad[1], pad[2], pad_value=0), TODO Unet live segmentation
data.ToTensor()]
ds_test = data.get_testdata(modalities=['_unet_core', '_unet_penu'], # modalities=['_CBV_reg1_downsampled', '_TTD_reg1_downsampled'], TODO Unet live segmentation
labels=['_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled',
'_FUCT_MAP_T_Samplespace_subset_reg1_downsampled'],
transform=transform,
indices=args.fold)
# Unet
#unet = None TODO Unet live segmentation
#if not args.groundtruth: TODO Unet live segmentation
# unet = Unet3D(channels=channels_unet) TODO Unet live segmentation
# unet.load_state_dict(torch.load(args.unet)) TODO Unet live segmentation
# unet.train(False) # fixate regularization for forward-only! TODO Unet live segmentation
for sample in ds_test:
case_id = sample[data.KEY_CASE_ID].cpu().numpy()[0]
nifph = nib.load('/share/data_zoe1/lucas/Linda_Segmentations/' + str(case_id) + '/train' + str(case_id) +
'_CBVmap_reg1_downsampled.nii.gz').affine
to_to_ta, normalization = get_normalized_time(sample, normalization_hours_penumbra)
lesion = Variable(sample[data.KEY_LABELS][:, 2, :, :, :].unsqueeze(data.DIM_CHANNEL_TORCH3D_5))
if args.groundtruth:
core = Variable(sample[data.KEY_LABELS][:, 0, :, :, :].unsqueeze(data.DIM_CHANNEL_TORCH3D_5))
penu = Variable(sample[data.KEY_LABELS][:, 1, :, :, :].unsqueeze(data.DIM_CHANNEL_TORCH3D_5))
else:
#dto = UnetDtoInit.init_dto(Variable(sample[data.KEY_IMAGES]), None, None) TODO Unet live segmentation
#dto = unet(dto) TODO Unet live segmentation
#core = dto.outputs.core TODO Unet live segmentation
#penu = dto.outputs.penu, TODO Unet live segmentation
core = Variable(sample[data.KEY_IMAGES][:, 0, :, :, :].unsqueeze(data.DIM_CHANNEL_TORCH3D_5))
penu = Variable(sample[data.KEY_IMAGES][:, 1, :, :, :].unsqueeze(data.DIM_CHANNEL_TORCH3D_5))
ta_to_tr = sample[data.KEY_GLOBAL][:, 1, :, :, :].squeeze().unsqueeze(data.DIM_CHANNEL_TORCH3D_5)
time_to_treatment = Variable(ta_to_tr.type(torch.FloatTensor) / normalization)
del to_to_ta
del normalization
recon_core, recon_intp, recon_penu, latent_core, latent_intp, latent_penu = \
sdm_interpolate_numpy(core.data.cpu().numpy(), penu.data.cpu().numpy(), threshold=0.5,
interpolation=time_to_treatment.data.cpu().numpy().squeeze(), zoom=12,
resample=args.downsample)
print(int(sample[data.KEY_CASE_ID]), 'TO-->TR', float(time_to_treatment))
if args.visualinspection:
fig, axes = plt.subplots(3, 4)
axes[0, 0].imshow(core.cpu().data.numpy()[0, 0, 16, :, :], cmap='gray', vmin=0, vmax=1)
axes[1, 0].imshow(lesion.cpu().data.numpy()[0, 0, 16, :, :], cmap='gray', vmin=0, vmax=1)
axes[2, 0].imshow(penu.cpu().data.numpy()[0, 0, 16, :, :], cmap='gray', vmin=0, vmax=1)
axes[0, 1].imshow(latent_core[16, :, :], cmap='gray')
axes[1, 1].imshow(latent_intp[16, :, :], cmap='gray')
axes[2, 1].imshow(latent_penu[16, :, :], cmap='gray')
axes[0, 2].imshow(recon_core[16, :, :], cmap='gray')
axes[1, 2].imshow(recon_intp[16, :, :], cmap='gray')
axes[2, 2].imshow(recon_penu[16, :, :], cmap='gray')
axes[0, 3].imshow(recon_core[16, :, :] < 0, cmap='gray', vmin=0, vmax=1)
axes[1, 3].imshow(recon_intp[16, :, :] > 0, cmap='gray', vmin=0, vmax=1)
axes[2, 3].imshow(recon_penu[16, :, :] > 0, cmap='gray', vmin=0, vmax=1)
plt.show()
results = metrics.binary_measures_numpy((recon_intp > 0).astype(np.float),
lesion.cpu().data.numpy()[0, 0, :, :, :], binary_threshold=0.5)
c_res = metrics.binary_measures_numpy((recon_core < 0).astype(np.float),
core.cpu().data.numpy()[0, 0, :, :, :], binary_threshold=0.5)
p_res = metrics.binary_measures_numpy((recon_penu > 0).astype(np.float),
penu.cpu().data.numpy()[0, 0, :, :, :], binary_threshold=0.5)
with open('/data_zoe1/lucas/Linda_Segmentations/tmp/sdm_results.txt', 'a') as f:
print('Evaluate case: {} - DC:{:.3}, HD:{:.3}, ASSD:{:.3}, Core recon DC:{:.3}, Penu recon DC:{:.3}'.format(case_id,
results.dc, results.hd, results.assd, c_res.dc, p_res.dc), file=f)
zoomed = ndi.interpolation.zoom(recon_intp.transpose((2, 1, 0)), zoom=(2, 2, 1))
nib.save(nib.Nifti1Image((zoomed > 0).astype(np.float32), nifph), args.outbasepath + '_' + str(case_id) + '_lesion.nii.gz')
del zoomed
zoomed = ndi.interpolation.zoom(lesion.cpu().data.numpy().astype(np.int8).transpose((4, 3, 2, 1, 0))[:, :, :, 0, 0], zoom=(2, 2, 1))
nib.save(nib.Nifti1Image(zoomed, nifph), args.outbasepath + '_' + str(case_id) + '_fuctgt.nii.gz')
del zoomed
zoomed = ndi.interpolation.zoom(recon_core.transpose((2, 1, 0)), zoom=(2, 2, 1))
nib.save(nib.Nifti1Image((zoomed < 0).astype(np.float32), nifph), args.outbasepath + '_' + str(case_id) + '_core.nii.gz')
del zoomed
zoomed = ndi.interpolation.zoom(recon_penu.transpose((2, 1, 0)), zoom=(2, 2, 1))
nib.save(nib.Nifti1Image((zoomed > 0).astype(np.float32), nifph), args.outbasepath + '_' + str(case_id) + '_penu.nii.gz')
del nifph
del sample
def train(args):
# Params / Config
learning_rate = 1e-3
momentums_cae = (0.9, 0.999)
weight_decay = 1e-5
criterion = metrics.BatchDiceLoss([1.0]) # nn.BCELoss()
resample_size = int(args.xyoriginal * args.xyresample)
n_globals = args.globals # type(core/penu), tO_to_tA, NHISS, sex, age
channels_enc = args.channelsenc
alpha = 1.0
cuda = True
# TODO assert initbycae XOR channels_enc
# CAE model
path_saved_model = args.caepath
cae = torch.load(path_saved_model)
cae.freeze(True)
if args.initbycae:
enc = torch.load(path_saved_model).enc
else:
enc = Enc3D(size_input_xy=resample_size,
size_input_z=args.zsize,
channels=channels_enc,
n_ch_global=n_globals,
alpha=alpha)
if cuda:
cae = cae.cuda()
enc = enc.cuda()
# Model params
params = [p for p in enc.parameters() if p.requires_grad]
print(
'# optimizing params',
sum([p.nelement() * p.requires_grad for p in params]),
'/ total new enc + old cae',
sum([p for p in enc.parameters()] +
[p.nelement() for p in cae.parameters()]))
# Optimizer with scheduler
optimizer = torch.optim.Adam(params,
lr=learning_rate,
weight_decay=weight_decay,
betas=momentums_cae)
if args.lrsteps:
scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer, args.lrsteps)
else:
scheduler = None
# Data
common_transform = [
data.ResamplePlaneXY(args.xyresample),
data.HemisphericFlipFixedToCaseId(split_id=args.hemisflipid)
]
train_transform = common_transform + [
data.ElasticDeform(apply_to_images=True),
data.ToTensor()
]
valid_transform = common_transform + [data.ToTensor()]
modalities = ['_unet_core', '_unet_penu']
labels = [
'_CBVmap_subset_reg1_downsampled', '_TTDmap_subset_reg1_downsampled',
'_FUCT_MAP_T_Samplespace_subset_reg1_downsampled'
]
ds_train, ds_valid = data.get_stroke_prediction_training_data(
modalities,
labels,
train_transform,
valid_transform,
args.fold,
args.validsetsize,
batchsize=args.batchsize)
print('Size training set:',
len(ds_train.sampler.indices), 'samples | Size validation set:',
len(ds_valid.sampler.indices), 'samples | Capacity batch:',
args.batchsize, 'samples')
print('# training batches:', len(ds_train), '| # validation batches:',
len(ds_valid))
# Training
learner = CaePredictionLearner(ds_train,
ds_valid,
cae,
enc,
optimizer,
scheduler,
n_epochs=args.epochs,
path_previous_base=args.inbasepath,
path_outputs_base=args.outbasepath,
criterion=criterion)
learner.run_training()