def main(args):
Checker.check_args(args.part, args.task)
input_kwargs = dict(
task=args.task,
weight=args.weight,
image_path=args.image_path,
save_path=args.save_path,
gpus=args.gpus)
if args.part == "Brain":
if args.task in ["mribet", "mrabet"]:
BrainBET(**input_kwargs)
elif args.task == "blackblood_segmentation":
BrainBlackBloodSegmentation(**input_kwargs)
elif args.part == "Chest":
if args.task == "lrmark_detection":
ChestLRmarkDetection(**input_kwargs)
elif args.task == "viewpoint_classification":
ViewpointClassifier(**input_kwargs)
elif args.task == "enhance_classification":
EnhanceCTClassification(**input_kwargs)
elif args.part == "Abdomen":
if args.task == "liver_segmentation":
AbdomenLiverSegmentation(**input_kwargs)
elif args.task == "kidney_tumor_segmentation":
# AbdomenKidneyTumorSegmentation(**input_kwargs)
elif args.part == "Colon":
pass
def _preprocessing(self, path: str) -> np.array:
"""
Preprocess the image from the path
Args:
(string) path : absolute path of image
Return:
(numpy ndarray) image
"""
mean_std = [29.311405133024834, 43.38181786843102]
if Checker.check_input_type_bool(path, 'nii'):
image = sitk.ReadImage(path)
self.space = image.GetSpacing()
image = sitk.GetArrayFromImage(image).astype('float32')
warnings.warn(
'.nii is not recommended as an image format '
'due to be not clear abour horizontal or vertical shape. '
'Please check the sample in README.md.', UserWarning)
elif Checker.check_input_type_bool(path, 'dcm'):
raise ValueError('.dcm is not supported. '
'Please convert dcm dummies to analyze format.')
elif Checker.check_input_type_bool(path, 'img') or \
Checker.check_input_type_bool(path, 'hdr'):
image = sitk.ReadImage(path)
self.space = image.GetSpacing()
image = np.squeeze(
sitk.GetArrayFromImage(image).astype('float32')) # (d, w, h)
elif Checker.check_input_type_bool(path, 'npy'):
image = np.load(path)
self.space = [1., 1., 1.]
warnings.warn(
'.npy is not recommended as an image format.'
'Since spacing cannot be identified from .npy, '
'spacing is set as [1., 1., 1.].', UserWarning)
else:
input_ext = path.split('.')[-1]
raise ValueError(f'.{input_ext} format is not supported.')
self.img_shape = image.shape
_, h, w = self.img_shape
imageo = image.copy()
image = zoom(image,
[self.space[-1] / 5., 256. / float(w), 256. / float(h)],
order=1,
mode='constant')
image = np.clip(image, 10, 190)
image = (image - mean_std[0]) / mean_std[1]
image = image[np.newaxis, ..., np.newaxis] # (1, d, w, h, 1)
return imageo, image
def _preprocessing(self, path):
"""
Preprocess the image from the path
Args:
(string) path : absolute path of image
Return:
(numpy ndarray) image
"""
if Checker.check_input_type_bool(path, 'dcm'):
raise ValueError(
'.dcm is not supported. '
'Please convert dcm dummies to analyze format.')
Img = img_as_float(io.imread(path))
Img = transform.resize(Img, (1024,1024))
Img = np.expand_dims(Img, -1)
FileName = os.path.split(path)[-1]
Img_mean = Img.mean()
Img_std = np.sqrt(((Img**2).mean()) - (Img.mean())**2)
Img -= Img_mean
Img /= Img_std
print("Preprocessing done on {} files...".format(str(FileName)))
return Img, FileName
def ChestLRmarkDetection(
task: str,
weight: str,
image_path: str,
save_path: Optional[str] = None,
gpus: str = "-1"
) -> np.array:
"""
"""
from medimodule.Chest import ChestLRmarkDetection
Checker.check_input_type(image_path, ["png", "jpg", "bmp"])
Checker.set_gpu(gpu_idx=gpus, framework="tf")
model = ChestLRmarkDetection(weight)
image = model.predict(
os.path.abspath(image_path),
)
return image
def AbdomenLiverSegmentation(
task: str,
weight: str,
image_path: str,
save_path: Optional[str] = None,
gpus: str = "-1"
) -> Tuple[np.array, np.array]:
"""
"""
from medimodule.Liver import LiverSegmentation
Checker.check_input_type(image_path, ["hdr", "img", "nii"])
Checker.set_gpu(gpu_idx=gpus, framework="tf")
model = LiverSegmentation(weight)
image, mask = model.predict(
os.path.abspath(image_path),
save_path=save_path)
return image, mask
def BrainBlackBloodSegmentation(
task: str,
weight: str,
image_path: str,
save_path: Optional[str] = None,
gpus: str = "-1"
) -> Tuple[np.array, np.array]:
"""
"""
from medimodule.Brain import BlackbloodSegmentation
Checker.check_input_type(image_path, "nii")
Checker.set_gpu(gpu_idx=gpus, framework="tf")
model = BlackbloodSegmentation(weight)
image, mask = model.predict(
os.path.abspath(image_path),
save_path=save_path)
return image, mask
def BrainBET(
task: str,
weight: str,
image_path: str,
save_path: Optional[str] = None,
gpus: str = "-1"
) -> Tuple[np.array, np.array]:
"""
"""
from medimodule.Brain import MRI_BET
Checker.check_input_type(image_path, "nii")
Checker.set_gpu(gpu_idx=gpus, framework="pytorch")
model = MRI_BET(weight)
image, mask = model.predict(
os.path.abspath(image_path),
img_type="T1" if "mri" in task else "MRA",
save_path=save_path)
return image, mask
def _preprocessing(self, path: str) -> np.array:
"""
Preprocess the image from the path
Args:
(string) path : absolute path of image
Return:
(numpy ndarray) image shape (1, h, w, d, 1)
"""
if Checker.check_input_type_bool(path, 'nii'):
image = sitk.ReadImage(path)
self.space = image.GetSpacing()
image = sitk.GetArrayFromImage(image).astype('float32')
elif Checker.check_input_type_bool(path, 'npy'):
image = np.load(path)
self.space = [1., 1., 1.]
warnings.warn(
'.npy is not recommended as an image format.'
'Since spacing cannot be identified from .npy, spacing is set as [1., 1., 1.].',
UserWarning)
elif Checker.check_input_type_bool(path, 'dcm'):
raise ValueError('.dcm is not supported.'
'Please convert dcm dummies to analyze format.')
else:
input_ext = path.split('.')[-1]
raise ValueError(f'.{input_ext} format is not supported.')
self.img_shape = image.shape
# normalize
windowing_range = [-40., 120.]
windowing_min = windowing_range[0] - windowing_range[1] // 2
windowing_max = windowing_range[0] + windowing_range[1] // 2
image = ndimage.zoom(image, [.5, .5, .5], order=1, mode='constant')
image = np.clip(image, windowing_min, windowing_max)
image = (image - windowing_min) / (windowing_max - windowing_min)
image = image[np.newaxis, ..., np.newaxis]
return image