def test_macenko_normalization(self):
"""Test macenko_pca normalization."""
stain_unmixing_routine_params = {
'stains': ['hematoxylin', 'eosin'],
'stain_unmixing_method': 'macenko_pca',
}
print("Macenko - Unmasked, using default, 'idealized' W_target")
tissue_rgb_normalized = deconvolution_based_normalization(
tissue_rgb,
stain_unmixing_routine_params=stain_unmixing_routine_params)
self.assertTupleEqual(tuple(
[int(tissue_rgb_normalized[..., i].mean()) for i in range(3)]),
(192, 161, 222)
)
print("Macenko - Unmasked, using W_target from good image")
tissue_rgb_normalized = deconvolution_based_normalization(
tissue_rgb, W_target=W_target,
stain_unmixing_routine_params=stain_unmixing_routine_params)
self.assertTupleEqual(tuple(
[int(tissue_rgb_normalized[..., i].mean()) for i in range(3)]),
(198, 163, 197)
)
print("Macenko - Masked, using W_target from good image")
tissue_rgb_normalized = deconvolution_based_normalization(
tissue_rgb, W_target=W_target, mask_out=mask_out,
stain_unmixing_routine_params=stain_unmixing_routine_params)
self.assertTupleEqual(tuple(
[int(tissue_rgb_normalized[..., i].mean()) for i in range(3)]),
(194, 172, 201)
)
def return_norm_image(img, mask, W_source=None, W_target=None):
img = deconvolution_based_normalization(
img,
W_source=W_source,
W_target=W_target,
im_target=None,
stains=['hematoxylin', 'eosin'],
mask_out=~mask,
stain_unmixing_routine_params={"I_0": 215})
return img
def test_macenko_normalization(self):
"""Test macenko_pca normalization."""
stain_unmixing_routine_params = {
'stains': ['hematoxylin', 'eosin'],
'stain_unmixing_method': 'macenko_pca',
}
# TCGA-A2-A3XS-DX1_xmin21421_ymin37486_.png, Amgad et al, 2019)
# for macenco (obtained using rgb_separate_stains_macenko_pca()
# and using reordered such that columns are the order:
# Hamtoxylin, Eosin, Null
W_target = np.array([[0.5807549, 0.08314027, 0.08213795],
[0.71681094, 0.90081588, 0.41999816],
[0.38588316, 0.42616716, -0.90380025]])
# Macenko - Unmasked, using default, 'idealized' W_target"
tissue_rgb_normalized = deconvolution_based_normalization(
cfg.tissue_rgb,
stain_unmixing_routine_params=stain_unmixing_routine_params)
assert tuple([
int(tissue_rgb_normalized[..., i].mean()) for i in range(3)
]) == (182, 121, 211)
# Macenko - Unmasked, using W_target from good image
tissue_rgb_normalized = deconvolution_based_normalization(
cfg.tissue_rgb,
W_target=W_target,
stain_unmixing_routine_params=stain_unmixing_routine_params)
assert tuple([
int(tissue_rgb_normalized[..., i].mean()) for i in range(3)
]) == (187, 124, 175)
# Macenko - Masked, using W_target from good image
tissue_rgb_normalized = deconvolution_based_normalization(
cfg.tissue_rgb,
W_target=W_target,
mask_out=cfg.mask_out,
stain_unmixing_routine_params=stain_unmixing_routine_params)
assert tuple([
int(tissue_rgb_normalized[..., i].mean()) for i in range(3)
]) == (187, 125, 174)
def colornorm(self, image_location, save_location, img_file):
# color norm. standard (from TCGA-A2-A3XS-DX1, Amgad et al, 2019)
cnorm = {
'mu': np.array([8.74108109, -0.12440419, 0.0444982]),
'sigma': np.array([0.6135447, 0.10989545, 0.0286032]),
}
# TCGA-A2-A3XS-DX1_xmin21421_ymin37486_.png, Amgad et al, 2019)
# for macenco (obtained using rgb_separate_stains_macenko_pca()
# and reordered such that columns are the order:
# Hamtoxylin, Eosin, Null
W_target = np.array([[0.5807549, 0.08314027, 0.08213795],
[0.71681094, 0.90081588, 0.41999816],
[0.38588316, 0.42616716, -0.90380025]])
# specify stains of input image
stains = [
'hematoxylin', # nuclei stain
'eosin', # cytoplasm stain
'null'
] # set to null if input contains only two stains
imInput = skimage.io.imread(os.path.join(image_location,
img_file))[:, :, :3]
stain_unmixing_routine_params = {
'stains': ['hematoxylin', 'eosin'],
'stain_unmixing_method': 'macenko_pca',
}
im_norm = deconvolution_based_normalization(
imInput,
W_target=W_target,
stain_unmixing_routine_params=stain_unmixing_routine_params)
stain_color_map = htk.preprocessing.color_deconvolution.stain_color_map
# create stain matrix
W = np.array([stain_color_map[st] for st in stains]).T
# perform standard color deconvolution
imDeconvolved = htk.preprocessing.color_deconvolution.color_deconvolution(
im_norm, W)
norm = Image.fromarray(im_norm)
norm_bw = norm.convert('L')
imDeconvolved.Stains[:, :, 2] = np.array(norm_bw)
decon = Image.fromarray(imDeconvolved.Stains)
norm.save(os.path.join(save_location + '_norm', img_file))
norm_bw.save(os.path.join(save_location + '_bw', img_file))
decon.save(os.path.join(save_location + '_decon', img_file))