def gpu_transforms(dcm_list) -> DataLoader:
msk_trfm = A.Compose([A.Resize(256, 256)],
additional_targets={"image1": 'image', "mask1": 'mask'})
ctds = CTDicomSlices(dcm_list, img_and_mask_transform = msk_trfm, n_surrounding=0, trim_edges=False, self_supervised_mask=True)#, preprocessing = prep)
dl = DataLoader(ctds, batch_size=1, num_workers = 0, shuffle=True)
return dl
def cpu_transforms(dcm_list) -> DataLoader:
mean, std = [61.0249], [78.3195]
prep = transforms.Compose([Window(50, 200), Imagify(50, 200), Normalize(mean, std)])
resize_tsfm = A.Compose([A.Resize(256, 256)],
additional_targets={"image1": 'image', "mask1": 'mask'})
img_mask_tsfm = A.Compose([
A.ShiftScaleRotate(shift_limit=(0.1, 0.1), scale_limit=(0.9, 1.1), rotate_limit=15),
A.HorizontalFlip()],
additional_targets={"image1": 'image', "mask1": 'mask'})
ctds = CTDicomSlices(dcm_list, preprocessing = prep,
resize_transform = resize_tsfm, img_and_mask_transform = img_mask_tsfm, n_surrounding=0)
#ctds = CTDicomSlicesFelzenszwalb(dcm_list, resize_transform=resize_tsfm, preprocessing=prep, transform = img_tsfm, n_surrounding=0, trim_edges=False)
dl = DataLoader(ctds, batch_size=1, num_workers = 0, shuffle=True)
return dl
def get_dataset(dataset, model_dir):
'''
Builds the necessary datasets
'''
# create ds
#dcm_list = CTDicomSlicesJigsaw.generate_file_list(dataset,
# dicom_glob='/*/*/dicoms/*.dcm')
prep = transforms.Compose([Window(WL, WW), Imagify(WL, WW)]) #, Normalize(mean, std)])
if pre_train == 'jigsaw' or pre_train == 'jigsaw_ennead' or pre_train == 'jigsaw_softrank':
if pre_train == 'jigsaw_softrank':
n_perms = None
dsm = DatasetManager.generate_train_val_test(dataset, val_frac=0.05, test_frac=0, pretrain_ds=True)
if model_dir is not None:
dsm.save_lists(model_dir)
train_dicoms, val_dicoms, _ = dsm.get_dicoms()
datasets = {}
datasets['train'] = CTDicomSlicesJigsaw(train_dicoms, preprocessing=prep, return_tile_coords=True,
perm_path=Constants.default_perms, n_shuffles_per_image=num_shuffles, num_perms=n_perms)
datasets['val'] = CTDicomSlicesJigsaw(val_dicoms, preprocessing=prep, return_tile_coords=True,
perm_path=Constants.default_perms, n_shuffles_per_image=num_shuffles, num_perms=n_perms)
return datasets
elif pre_train == 'felz':
dcm_list = CTDicomSlicesJigsaw.generate_file_list(dataset,
dicom_glob='/*/*/dicoms/*.dcm')
# Felz masks were saved with foreground being 1,2,3,4 (instead of 255). mask_is_255 flag is CRITICAL
ctds = CTDicomSlices(dcm_list, preprocessing=prep, n_surrounding=in_channels // 2, mask_is_255=False)
return ctds
else:
raise Exception('Invalid pre_train mode of "{}"'.format(pre_train))
ax = fig.add_subplot(1, 3, 1)
ax.imshow(slices, cmap='gray')
ax2 = fig.add_subplot(1, 3, 2)
ax2.imshow(slices, cmap='gray')
ax2.imshow(mask, cmap='jet', alpha=0.5)
ax3 = fig.add_subplot(1, 3, 3)
ax3.imshow(mask, cmap='gray')
fig.show()
if __name__ == '__main__':
dataset = '/mnt/g/thesis/ct_only_cleaned_resized_mini/head-neck-radiomics'
#dataset = Constants.organized_dataset_2
dcm_list = CTDicomSlices.generate_file_list(dataset)
#dicom_glob='/*/*/*.dcm')
prep = transforms.Compose([Window(50, 200), Imagify(50, 200)])
tsfm = A.Compose([
A.SmallestMaxSize(max_size=256, always_apply=True, p=1),
A.CenterCrop(256, 256, always_apply=True, p=1.0)
])
#ctds = CTDicomSlices(dcm_list, preprocessing=prep, resize_transform=tsfm,
# n_surrounding=0)# , felz_crop=True)
ctds = CTDicomSlices(dcm_list, n_surrounding=0, mask_is_255=False)
show_dataset(ctds)
edgecolor='g',
facecolor='none')
ax.add_patch(p)
puzzle = ResnetJigsaw.tiles_to_image(tiles, in_channels=1)[0]
ax2 = fig.add_subplot(1, 2, 2)
ax2.imshow(puzzle[0], cmap='gray')
fig.show()
if __name__ == '__main__':
dataset = Constants.organized_dataset_2 #Constants.organized_dataset_2 #'/mnt/g/thesis/ct_only_cleaned_mini' #Constants.ct_only_cleaned #
dcm_list = CTDicomSlices.generate_file_list(
dataset,
dicom_glob='/*/dicoms/*.dcm') #dicom_glob='/*/*/dicoms/*.dcm')
prep = transforms.Compose(
[Window(50, 200),
Imagify(50, 200),
Normalize(61.0249, 78.3195)])
ctds = CTDicomSlicesJigsaw(dcm_list,
preprocessing=prep,
trim_edges=False,
return_tile_coords=True,
perm_path=Constants.default_perms,
n_shuffles_per_image=1,
num_perms=100)
#show_jigsaw_dataset(ctds)
def get_datasets(model_dir=None,
new_ds_split=True,
train_list="train.txt",
val_list="val.txt",
test_list="test.txt"):
'''
Builds the necessary datasets
model_dir is where model parameters will be stored
new_ds_split creates a new train/val/test split if True, and loads the relevant folders if false
dslist_in_pt_dir uses the pt dir as a reference for where the patient lists are located
'''
# Manage patient splits
if new_ds_split:
dsm = DatasetManager.generate_train_val_test(dataset_dir, val_frac,
test_frac)
if model_dir is not None:
dsm.save_lists(model_dir)
else:
dsm = DatasetManager.load_train_val_test(dataset_dir, train_list,
val_list, test_list)
#preprocess_fn = get_preprocessing_fn(backbone, pretrained=encoder_weights)
prep = transforms.Compose(
[Window(WL, WW), Imagify(WL, WW),
Normalize(mean, std)])
resize_tsfm = A.Compose([A.Resize(img_size, img_size)],
additional_targets={
"image1": 'image',
"mask1": 'mask'
})
img_mask_tsfm = A.Compose([
A.ShiftScaleRotate(
shift_limit=translate, scale_limit=scale, rotate_limit=rotate),
A.HorizontalFlip()
],
additional_targets={
"image1": 'image',
"mask1": 'mask'
})
# create ds
train_dicoms, val_dicoms, test_dicoms = dsm.get_dicoms(
train_frac=train_frac)
datasets = {}
datasets['train'] = CTDicomSlices(train_dicoms,
preprocessing=prep,
resize_transform=resize_tsfm,
img_and_mask_transform=img_mask_tsfm,
n_surrounding=in_channels // 2)
datasets['val'] = CTDicomSlices(val_dicoms,
preprocessing=prep,
resize_transform=resize_tsfm,
n_surrounding=in_channels // 2)
datasets['test'] = CTDicomSlices(test_dicoms,
preprocessing=prep,
resize_transform=resize_tsfm,
n_surrounding=in_channels // 2)
return datasets
selected_pixels = selected_pixels.astype('int32')
selected_segments = [segments[tuple(sp)] for sp in selected_pixels]
pre_mask = [segments == ss for ss in selected_segments]
mask = np.logical_or.reduce(pre_mask)
return segments, mask
use_gpu_transforms = False
if __name__ == '__main__':
only_positive = True
dataset = '/home/hussam/organized_dataset_2/'
dcm_list = CTDicomSlices.generate_file_list(dataset)
if use_gpu_transforms:
ct_dl = gpu_transforms(dcm_list)
else:
ct_dl = cpu_transforms(dcm_list)
for slices, mask, img_path, slice_n in ct_dl:
slices = slices.permute(0, 3, 1, 2)
if use_gpu_transforms:
slices, mask = do_inplace_transforms(slices, mask)
mask = mask[0]
slices = slices[0]
img_path = img_path[0]
slice_n = slice_n[0]