Source code for mermaid.example_generation

Package to create example images to test the image registration algorithms
from __future__ import absolute_import

from builtins import object
from abc import ABCMeta, abstractmethod
import numpy as np
from . import fileio
from future.utils import with_metaclass

[docs]class CreateExample(with_metaclass(ABCMeta, object)): """ Abstract base class. """ def __init__(self,dim): """ Constructor :param dim: Desired dimension of the example image """ self.dim = dim """Spatial dimension"""
[docs] @abstractmethod def create_image_pair(self,sz=None,params=None): """ Abstract method to create example image pairs :param params: Dictionary which contains parameters to create the images :return: Will return two images """ pass
[docs]class CreateSquares(CreateExample): """ Class to create two example squares in arbitrary dimension as registration test cases """ def __init__(self,dim, add_noise_to_bg=False): super(CreateSquares, self).__init__(dim) self.add_noise_to_bg = add_noise_to_bg
[docs] def create_image_pair(self,sz,params): """ Creates two square images in dimensions 1-3 :param sz: Desired size, e.g., [5,10] :param params: Parameter dictionary. Uses 'len_s' and 'len_l' to define the side-length of the small and the large squares which will be generated :return: Returns two images of squares and the spacing (I0,I1,spacing) """ if not self.add_noise_to_bg: I0 = np.zeros(sz, dtype='float32') I1 = np.zeros(sz, dtype='float32') else: I0 = np.random.rand(*sz).astype(np.float32)/5. I1 = np.random.rand(*sz).astype(np.float32)/5. # get parameters and replace with defaults if necessary # create a new category if it does not exist params[('square_example_images', {}, 'Controlling the size of a nD cube')] len_s = params['square_example_images'][('len_s',int(sz.min()//6),'Mimimum side-length of square')] len_l = params['square_example_images'][('len_l',int(sz.max()//4),'Maximum side-length of square')] c = sz//2 # center coordinates # create small and large squares if self.dim==1: I0[c[0]-len_s:c[0]+len_s]=1 I1[c[0]-len_l:c[0]+len_l]=1 elif self.dim==2: I0[c[0]-len_s:c[0]+len_s, c[1]-len_s:c[1]+len_s] = 1 I1[c[0]-len_l:c[0]+len_l, c[1]-len_l:c[1]+len_l] = 1 elif self.dim==3: I0[c[0] - len_s:c[0] + len_s, c[1] - len_s:c[1] + len_s, c[2]-len_s:c[2]+len_s] = 1 I1[c[0] - len_l:c[0] + len_l, c[1] - len_l:c[1] + len_l, c[2]-len_l:c[2]+len_l] = 1 else: raise ValueError('Square examples only supported in dimensions 1-3.') # now transform from single-channel to multi-channel image format I0 = I0.reshape([1, 1] + list(I0.shape)) I1 = I1.reshape([1, 1] + list(I1.shape)) sz = np.array(I0.shape) spacing = 1. / (sz[2::] - 1) # the first two dimensions are batch size and number of image channels return I0,I1,spacing
[docs]class CreateRealExampleImages(CreateExample): """ Class to create two example brain images. Currently only supported in 2D """ def __init__(self,dim=2,s_path=None,t_path=None): super(CreateRealExampleImages, self).__init__(dim) if s_path is None: self.s_path = '../mermaid_test_data/brain_slices/ws_slice.nrrd' self.t_path = '../mermaid_test_data/brain_slices/wt_slice.nrrd' else: self.s_path = s_path self.t_path = t_path
[docs] def create_image_pair(self,sz=None,params=None): """ Loads the two brain images using SimpleITK, normalizes them so that the 95-th percentile is as 0.95 and returns them. :param sz: Ignored :param params: Ignored :return: Returns the two brain slices. """ # create small and large squares if self.dim==2: I0,_,_,squeezed_spacing = fileio.ImageIO().read_to_nc_format(filename=self.s_path,intensity_normalize=True,squeeze_image=True) I1,_,_,squeezed_spacing = fileio.ImageIO().read_to_nc_format(filename=self.t_path,intensity_normalize=True,squeeze_image=True) else: raise ValueError('Real examples only supported in dimension 2 at the moment.') return I0,I1,squeezed_spacing