def pre_process(self, dictionnaire: dict) -> BytesIO:
"""[Pre_process for pytorch]
Args:
dictionnaire (dict): [Dictionary containing the id of the images]
Returns:
BytesIO: [description]
"""
dict = dictionnaire
data_path = settings.STORAGE_DIR
idImage = str(dict['id'])
path_image = data_path + '/image/image_' + idImage + '_CT.nii'
objet = Nifti(path_image)
resampled = objet.resample(shape=(256, 256, 1024))
mip_generator = MIP_Generator(resampled)
array = mip_generator.project(angle=0)
array[np.where(array < 500)] = 0 #500 UH
array[np.where(array > 1024)] = 1024 #1024 UH
array = array[:, :, ] / 1024
array = np.expand_dims(array, axis=0)
array = array.astype(np.double)
np_bytes = BytesIO()
np.save(np_bytes, array, allow_pickle=True)
np_bytes = np_bytes.getvalue()
return np_bytes
def preprocessing_image_to_bytes(image: str,
output_shape: tuple,
angle: int = 0) -> torch.Tensor:
"""
Preprocessing applied to an image to fit the input data for classification neural network
Resample to the output shape
Scale Range intensity
Transform to torch array
Args:
image (str): path to the nifti file
output_shape (tuple): shape of the image required by the neural network
angle (int): angle to be applied to the CT image
Returns:
(torch.Tensor): Tensor of the preprocessed image (input for neural classification neural network)
"""
objet = Nifti(image)
resampled = objet.resample(output_shape)
mip_generator = MIP_Generator(resampled)
array = mip_generator.project(angle=angle)
array[np.where(array < 500)] = 0 #500 UH
array[np.where(array > 1024)] = 1024 #1024 UH
array = array[:, :, ] / 1024
array = np.expand_dims(array, axis=0)
array = array.astype(np.double)
array = np.expand_dims(array, axis=0)
array = torch.from_numpy(np.array(array))
return array
def generate_mip(self,idImage:str):
data_path = settings.STORAGE_DIR
directory=settings.STORAGE_DIR+'/image'
path_ct =data_path+'/image/image_'+idImage+'.nii'
objet = Nifti(path_ct)
resampled = objet.resample(shape=(256, 256, 1024))
resampled[np.where(resampled < 500)] = 0 #500 UH
normalize = resampled[:,:,:,]/np.max(resampled)
mip_generator = MIP_Generator(normalize)
mip_generator.project(angle=0)
print(mip_generator.project(angle=0))
mip_generator.save_as_png('image_2D_'+idImage, directory, vmin=0, vmax=1)
def preprocessing(path_image, output_shape, angle):
objet = Nifti(path_image)
resampled = objet.resample(shape=output_shape)
mip_generator = MIP_Generator(resampled)
array = mip_generator.project(angle=angle)
array[np.where(array < 500)] = 0 #500 UH
array[np.where(array > 1024)] = 1024 #1024 UH
array = array[:, :, ] / 1024
array = np.expand_dims(array, axis=0)
array = array.astype(np.double)
np_bytes = BytesIO()
np.save(np_bytes, array, allow_pickle=True)
np_bytes = np_bytes.getvalue()
return np_bytes
def __call__(self, sample:dict)->dict:
"""
Reshape and scale intensity range of the image.
Transforms the data into torch tensors for the neural network.
Args:
sample (dict): dictionnary of the input data with the original image
Returns:
(dict): dictionnary of the input data with the transformed image. Ready for the neural network.
"""
objet = Nifti(sample['image'])
resampled = objet.resample(self.output_shape)
mip_generator = MIP_Generator(resampled)
array=mip_generator.project(angle=self.angle)
array[np.where(array < 500)] = 0 #500 UH
array[np.where(array > 1024)] = 1024 #1024 UH
array = array[:,:,]/1024
array = np.expand_dims(array, axis=0)
sample = {'image': torch.from_numpy(np.array(array)), 'head': torch.from_numpy(np.array(sample['head'])),'leg': torch.from_numpy(np.array(sample['leg'])),'right_arm': torch.from_numpy(np.array(sample['right_arm'])),'left_arm': torch.from_numpy(np.array(sample['left_arm'])) }
return sample
def generate_instance_from_nifti(img_dict: dict) -> tuple:
"""prepare a batch for classification model from a dict
Args:
img_dict ([dict]): [{ct_img : value(a nifti image.nii), upper_limit : value, lower_limit : value, right_arm : value, left_arm : value}]
Returns:
[tuple]: [return the 2D MIP of the CT image and its encoded label]
"""
resampled_array = Nifti(img_dict['ct_img']).resample(
shape_matrix=(256, 256, 1024), shape_physic=(700, 700, 2000))
resampled_array[np.where(resampled_array < 500)] = 0 #500 UH
normalize = resampled_array[:, :, :, ] / np.max(resampled_array)
mip_generator = MIP_Generator(normalize)
mip = mip_generator.project(angle=0)
mip = np.expand_dims(mip, -1)
label = encoding_instance(img_dict)
return mip, label
def pre_process(self, dictionnaire: dict) -> TensorProto:
"""[Pre_process for TF]
Args:
dictionnaire (dict): [Dictionary containing the id of the images]
Returns:
TensorProto: [description]
"""
dict = dictionnaire
data_path = settings.STORAGE_DIR
idImage = str(dict['id'])
nifti_path = data_path + '/image/image_' + idImage + '.nii'
resampled_array = Nifti(nifti_path).resample((256, 256, 1024))
resampled_array[np.where(resampled_array < 500)] = 0 #500 UH
normalize = resampled_array[:, :, :, ] / np.max(
resampled_array) #normalize
mip_generator = MIP_Generator(normalize)
mip = mip_generator.project(angle=0)
mip = np.expand_dims(mip, -1)
mip = np.array(mip).astype('float32')
return tf.make_tensor_proto(mip, shape=[1, 1024, 256, 1])