def load(self, kwargs):
""" Parameters
----------
depth : int, optional
Specified the layers deep the proposed U-Net should go.
Layer depth is symmetric on both upsampling and downsampling
arms.
max_filter: int, optional
Specifies the number of filters at the bottom level of the U-Net.
"""
add_parameter(self, kwargs, 'same_mask', True)
add_parameter(self, kwargs, 'reference_channel', [0, 1])
add_parameter(self, kwargs, 'model',
None) # TODO: Replace with load(skull_strip_model)
add_parameter(self, kwargs, 'name', 'SkullStrip_Model')
add_parameter(self, kwargs, 'preprocessor_string', '_SkullStripped')
self.array_input = True
self.mask_string = '_Skullstrip_Mask'
self.mask_filename = None
if type(self.reference_channel) is not list:
self.reference_channel = [self.reference_channel]
def load(self, kwargs):
super(ZeroMeanNormalization, self).load(kwargs)
# Naming Parameters
add_parameter(self, kwargs, 'name', 'ZeroMeanNormalization')
add_parameter(self, kwargs, 'preprocessor_string', '_ZeroNorm')
def load(self, kwargs):
""" Parameters
----------
name : str, optional
Preprocessor name for internal use. Default is 'SkullStrip_Model'
preprocessor_string: str, optional
Appended suffix to filenames saved out from this preprocessor.
Default is '_SkullStripped'
reference_channel: int or list, optional
model: DeepNeuroModel, optional
DeepNeuroModel from which to run inference in this preprocessor.
"""
add_parameter(self, kwargs, 'reference_channel', [0, 1])
add_parameter(self, kwargs, 'model', None)
# Data Output Parameters
add_parameter(self, kwargs, 'output_filename',
'skullstrip_mask.nii.gz')
add_parameter(self, kwargs, 'name', 'SkullStrip_Model')
add_parameter(self, kwargs, 'preprocessor_string', '_SkullStripped')
self.array_input = True
self.mask_string = '_Skullstrip_Mask'
self.mask_filename = None
self.mask_numpy = None
if type(self.reference_channel) is not list:
self.reference_channel = [self.reference_channel]
def load(self, kwargs):
""" Parameters
----------
depth : int, optional
Specified the layers deep the proposed U-Net should go.
Layer depth is symmetric on both upsampling and downsampling
arms.
max_filter: int, optional
Specifies the number of filters at the bottom level of the U-Net.
"""
super(PGGAN, self).load(kwargs)
add_parameter(self, kwargs, 'initial_size', (4, 4))
if type(self.initial_size) not in [list, tuple]:
self.initial_size = (4, ) * self.dim
# PGGAN Parameters
self.starting_depth = 1
self.transition_dict = {True: '_Transition', False: ''}
if self.dim == 3:
raise NotImplementedError
def load(self, kwargs):
# Naming Parameters
add_parameter(self, kwargs, 'name', 'Merge')
# Dropping Parameters
add_parameter(self, kwargs, 'channels', [0, 1, 2, 3])
self.output_shape = {}
self.array_input = True
def load(self, kwargs):
"""
"""
# Flip and Rotate options
add_parameter(self, kwargs, 'rotation_axes', [1, 2, 3])
# Derived Parameters
self.rotation_generator = {}
self.augmentation_num = 0
def load(self, kwargs):
super(BinaryNormalization, self).load(kwargs)
add_parameter(self, kwargs, 'intensity_range', [-1, 1])
add_parameter(self, kwargs, 'threshold', 0)
add_parameter(self, kwargs, 'single_value', None)
# Naming Parameters
add_parameter(self, kwargs, 'name', 'BinaryNormalization')
add_parameter(self, kwargs, 'preprocessor_string', '_Binary')
def __init__(self, **kwargs):
# Default Variables
add_parameter(self, kwargs, 'verbose', False)
add_parameter(self, kwargs, 'raw_data', None)
add_parameter(self, kwargs, 'ground_truth', 'ground_truth')
# Naming Variables
add_parameter(self, kwargs, 'name', 'Postprocesser')
add_parameter(self, kwargs, 'postprocessor_string', '_postprocess')
self.load(kwargs)
def load(self, kwargs):
# Naming Parameters
add_parameter(self, kwargs, 'name', 'MaskValues')
add_parameter(self, kwargs, 'preprocessor_string', 'Masked')
# Masking Parameters
add_parameter(self, kwargs, 'mask_threshold', 0)
add_parameter(self, kwargs, 'mask_value', 0)
add_parameter(self, kwargs, 'mask_mode', 'lower')
self.array_input = True
def load(self, kwargs):
super(RangeNormalization, self).load(kwargs)
add_parameter(self, kwargs, 'intensity_range', [-1, 1])
add_parameter(self, kwargs, 'input_intensity_range', None)
add_parameter(self, kwargs, 'outlier_percent', None) # Not Implemented
# Naming Parameters
add_parameter(self, kwargs, 'name', 'RangeNormalization')
add_parameter(self, kwargs, 'preprocessor_string', '_Range')
def load(self, kwargs):
# Naming Parameters
add_parameter(self, kwargs, 'name', 'ReorderAxes')
# Dropping Parameters
add_parameter(self, kwargs, 'axis_ordering', None)
assert self.axis_ordering is not None, 'You must provide an axis ordering for ReorderAxes.'
self.output_shape = {}
self.array_input = True
def load(self, kwargs):
# Naming Parameters
add_parameter(self, kwargs, 'name', 'OneHotClasses')
# Dropping Parameters
add_parameter(self, kwargs, 'channels', None)
add_parameter(self, kwargs, 'output_data_group', None)
add_parameter(self, kwargs, 'max_centroid', 1)
add_parameter(self, kwargs, 'aggregate_centroid', False)
self.output_shape = {}
self.array_input = True
def load(self, kwargs):
# Naming Parameters
add_parameter(self, kwargs, 'name', 'SqueezeAxes')
# Dropping Parameters
add_parameter(self, kwargs, 'axes', None)
if type(self.axes) is not list and self.axes is not None:
self.axes = [self.axes]
self.output_shape = {}
self.array_input = True
def load(self, kwargs):
# A lot of this functionality is vague and messy, revisit
# Downsample Parameters
add_parameter(self, kwargs, 'channel', 0)
add_parameter(self, kwargs, 'axes', {})
add_parameter(self, kwargs, 'factor', 2)
add_parameter(self, kwargs, 'random_sample', True)
add_parameter(self, kwargs, 'num_downsampled', 1)
self.input_shape = {}
self.augmentation_string = '_resample_'
def load(self, kwargs):
# Flip and Rotate options
add_parameter(self, kwargs, 'flip', True)
add_parameter(self, kwargs, 'rotate', True)
add_parameter(self, kwargs, 'flip_axis', 2)
add_parameter(self, kwargs, 'rotate_axis', (1, 2))
# TODO: This is incredibly over-elaborate, return to fix.
self.transforms_list = []
if self.flip:
self.flip_list = [False, True]
else:
self.flip_list = [False]
if self.rotate:
self.rotations_90 = [0, 1, 2, 3]
else:
self.rotations_90 = [0]
self.available_transforms = np.array(
np.meshgrid(self.flip_list, self.rotations_90)).T.reshape(-1, 2)
self.total_transforms = self.available_transforms.shape[0]
self.augmentation_string = '_rotate2D_'
def load(self, kwargs):
""" Parameters
----------
depth : int, optional
Specified the layers deep the proposed U-Net should go.
Layer depth is symmetric on both upsampling and downsampling
arms.
max_filter: int, optional
Specifies the number of filters at the bottom level of the U-Net.
"""
# Evaluation Params
add_parameter(self, kwargs, 'ground_truth', None)
# Saving Params
add_parameter(self, kwargs, 'postprocessor_string',
'_pseudoprobability')
# Model Parameters
add_parameter(self, kwargs, 'channels_first', False)
add_parameter(self, kwargs, 'input_channels', None)
if 'channels_dim' in kwargs:
self.channels_dim = kwargs.get('channels_dim')
elif self.channels_first:
self.channels_dim = 1
else:
self.channels_dim = -1
def __init__(self, case_list=None, verbose=False, **kwargs):
# Data sources
add_parameter(self, kwargs, 'data_sources', None)
self.data_types = ['directories', 'filepaths', 'csv', 'numpy', 'hdf5']
# File location variables
add_parameter(self, kwargs, 'source', 'directories')
add_parameter(self, kwargs, 'recursive', False)
add_parameter(self, kwargs, 'file_identifying_chars', None)
self.open_hdf5 = None
self.case_list = case_list
self.verbose = verbose
# Special behavior for augmentations
self.augmentations = []
self.preprocessors = []
self.multiplier = 1
# Empty vars
self.cases = []
self.total_cases = 0
self.current_case = None
# Data group variables
self.data_groups = {}
if self.data_sources is not None:
for data_type in self.data_types:
if data_type not in list(self.data_sources.keys()):
self.data_sources[data_type] = None
self.fill_data_groups()
def load(self, kwargs):
# Flip and Rotate options
add_parameter(self, kwargs, 'shift', True)
add_parameter(self, kwargs, 'squeeze', True)
add_parameter(self, kwargs, 'shift_amount', [-.5, .5])
add_parameter(self, kwargs, 'squeeze_factor', [.7, 1.3])
# TODO: This is incredibly over-elaborate, return to fix.
self.transforms_list = []
if self.shift:
self.shift_list = [False, True]
else:
self.shift_list = [False]
if self.squeeze:
self.squeeze_list = [False, True]
else:
self.squeeze_list = [False]
self.available_transforms = np.array(
np.meshgrid(self.shift_list, self.squeeze_list)).T.reshape(-1, 2)
self.total_transforms = self.available_transforms.shape[0]
self.augmentation_string = '_shift_squeeze_'
def load(self, kwargs):
""" Parameters
----------
depth : int, optional
Specified the layers deep the proposed U-Net should go.
Layer depth is symmetric on both upsampling and downsampling
arms.
max_filter: int, optional
Specifies the number of filters at the bottom level of the U-Net.
"""
super(MinimalKerasCNN, self).load(kwargs)
add_parameter(self, kwargs, 'filter_size', 1)
def load(self, kwargs):
""" Parameters
----------
depth : int, optional
Specified the layers deep the proposed U-Net should go.
Layer depth is symmetric on both upsampling and downsampling
arms.
max_filter: int, optional
Specifies the number of filters at the bottom level of the U-Net.
"""
add_parameter(self, kwargs, 'sess', None)
add_parameter(self, kwargs, 'saver', None)
self.tensorflow_optimizer_dict = {'Adam': tf.train.AdamOptimizer}
def load(self, kwargs):
# Naming parameter
add_parameter(self, kwargs, 'name', 'Binarization')
add_parameter(self, kwargs, 'postprocessor_string', 'largest_components')
add_parameter(self, kwargs, 'component_number', 1)
add_parameter(self, kwargs, 'connectivity', 2)
def load(self, kwargs):
# Naming parameter
add_parameter(self, kwargs, 'name', 'Rescale')
add_parameter(self, kwargs, 'postprocessor_string', 'rescaled')
add_parameter(self, kwargs, 'input_intensity_range', None)
add_parameter(self, kwargs, 'output_intensity_range', [0, 255])
def load(self, kwargs):
super(KerasModel, self).load(kwargs)
# Basic Keras Model Params
add_parameter(self, kwargs, 'output_activation', True)
# Specific Cost Function Params
add_parameter(self, kwargs, 'categorical_weighting', {0: 0.1, 1: 3.0})
self.keras_optimizer_dict = {'Nadam': Nadam, 'Adam': Adam, 'SGD': SGD}
if self.input_tensor is None:
self.inputs = Input(self.input_shape)
else:
self.inputs = self.input_tensor
def load(self, kwargs):
# Naming Parameters
add_parameter(self, kwargs, 'name', 'OneHotEncode')
# Class Parameters
add_parameter(self, kwargs, 'num_classes', 3)
add_parameter(self, kwargs, 'input_classes', None)
add_parameter(self, kwargs, 'class_dictionary', {})
self.output_shape = {}
self.array_input = True
def load(self, kwargs):
""" Parameters
----------
depth : int, optional
Specified the layers deep the proposed U-Net should go.
Layer depth is symmetric on both upsampling and downsampling
arms.
max_filter: int, optional
Specifies the number of filters at the bottom level of the U-Net.
"""
super(UNet, self).load(kwargs)
add_parameter(self, kwargs, 'num_blocks', 2)
add_parameter(self, kwargs, 'block_type', 'basic')
add_parameter(self, kwargs, 'block_filter_growth_ratio', 2)
add_parameter(self, kwargs, 'depth', 4)
add_parameter(self, kwargs, 'output_channels', 1)
def load(self, kwargs):
# Evaluation Params
add_parameter(self, kwargs, 'ground_truth', None)
# Saving Params
add_parameter(self, kwargs, 'postprocessor_string', '_inference')
# Model Parameters
add_parameter(self, kwargs, 'input_channels', None)
add_parameter(self, kwargs, 'output_channels', None)
add_parameter(self, kwargs, 'channels_dim', None)
if self.channels_dim is None:
if self.channels_first:
self.channels_dim = 1
else:
self.channels_dim = -1
def load(self, kwargs):
# Add functionality for masking multiples axes.
# Mask Parameters
add_parameter(self, kwargs, 'mask_channels', {})
add_parameter(self, kwargs, 'num_masked', 1)
add_parameter(self, kwargs, 'masked_value', -10)
add_parameter(self, kwargs, 'random_sample', True)
# Derived Parameters
self.input_shape = {}
self.augmentation_string = '_mask_'
def load(self, kwargs):
# Add functionality for choosing multiple axes
# Choose Parameters
add_parameter(self, kwargs, 'axis', {})
add_parameter(self, kwargs, 'choices', None)
add_parameter(self, kwargs, 'num_chosen', 1)
add_parameter(self, kwargs, 'random_sample', True)
# Derived Parameters
self.input_shape = {}
self.augmentation_string = '_choose_'
def load(self, kwargs):
# Naming parameter
add_parameter(self, kwargs, 'name', 'Binarization')
add_parameter(self, kwargs, 'postprocessor_string', 'binarized')
add_parameter(self, kwargs, 'binarization_threshold', 0.5)
def __init__(self, data_directory=None, data_storage=None, data_group_dict=None, spreadsheet_dict=None, value_dict=None, case_list=None, verbose=False, **kwargs):
# Input vars
self.data_directory = data_directory
self.data_storage = data_storage
self.data_group_dict = data_group_dict
self.spreadsheet_dict = spreadsheet_dict
# File location variables
add_parameter(self, kwargs, 'source', 'directories')
add_parameter(self, kwargs, 'recursive', False)
add_parameter(self, kwargs, 'file_identifying_chars', None)
self.value_dict = value_dict
self.case_list = case_list
self.verbose = verbose
# Special behavior for augmentations
self.augmentations = []
self.preprocessors = []
self.multiplier = 1
# Empty vars
self.cases = []
self.preprocessed_cases = {}
self.total_cases = 0
self.current_case = None
# Data group variables
self.data_groups = {}
if data_group_dict is not None or data_storage is not None:
self.fill_data_groups()
