# -*- coding: utf-8 -*-
# <nbformat>3.0</nbformat>
# <codecell>
import os
import numpy as np
from scipy.ndimage import maximum_filter, minimum_filter, binary_fill_holes
import skimage as ski
from skimage import io, filter, color, exposure, morphology, feature, draw, measure, transform
figsize(16, 10)
# <codecell>
l = os.listdir("../data")
print l[6]
qstain = np.array([[.26451728, .5205347, .81183386], [.9199094, .29797825, .25489032], [.28947765, .80015373, .5253158]])
# <codecell>
# Original image
A = io.imread("../data/" + l[2])
io.imshow(A)
# <codecell>
# Colour Deconvolution
deconv = ski.img_as_float(color.separate_stains(A, np.linalg.inv(qstain)))
io.imshow(exposure.adjust_sigmoid(
filter.gaussian_filter(
exposure.rescale_intensity(
deconv[:, :, 0] + deconv[:, :, 2],
out_range=(0, 1)),
17),
gain=7, cutoff=0.6))
# <codecell>
# Blood
blood = \
morphology.remove_small_objects(
filter.threshold_adaptive(
exposure.adjust_sigmoid(
filter.gaussian_filter(
exposure.rescale_intensity(
deconv[:, :, 0] + deconv[:, :, 2],
out_range=(0, 1)),
15),
gain=10, cutoff=0.7),
301, offset=-0.15),
500)
io.imshow(blood)
# <codecell>
# Veins
veins = \
binary_fill_holes(
morphology.remove_small_objects(
filter.threshold_adaptive(
filter.gaussian_filter(
exposure.adjust_sigmoid(A[:,:,1]),
31),
501, offset=-0.07),
3000) #### SRSLY ?
)
io.imshow(veins)
# <codecell>
# Inflammation
inflammation = \
morphology.remove_small_objects(
filter.threshold_adaptive(
exposure.adjust_sigmoid(
filter.gaussian_filter(
exposure.rescale_intensity(
deconv[:, :, 1],
out_range = (0, 1)),
25),
gain = 12, cutoff = 0.32),
501, offset = -0.1),
4000)
io.imshow(inflammation)
#io.imshow(filter.threshold_adaptive(exposure.adjust_sigmoid(filter.gaussian_filter(exposure.rescale_intensity(deconv[:, :, 1], out_range=(0, 1)), 19), gain=15, cutoff=0.4), 401))
# <codecell>
# Labelled
io.imshow(veins + 2 * blood + 3 * inflammation)
# <codecell>
# All regions of interest
#%timeit
"""
all = \
minimum_filter(
maximum_filter(
np.logical_or(np.logical_or(blood, veins), inflammation).astype(float),
size = 101),
size = 51)
"""
"""
morphology.binary_erosion(
morphology.binary_dilation(
np.logical_or(np.logical_or(blood, veins), inflammation),
morphology.disk(31)),
morphology.disk(21))
"""
full = (np.logical_or(np.logical_or(blood, veins), inflammation))
#io.imshow(binary_fill_holes(minimum_filter(maximum_filter(full, size=51, footprint=morphology.diamond(31)), size=51)))
# <codecell>
io.imshow(ski.img_as_float(A)*0.5 + full[:, :, newaxis]*0.5)
# <codecell>
io.imsave("__1.gif", A)
io.imsave("__2.gif", ski.img_as_float(full))
# <codecell>
os.system("gifsicle --delay=80 --loop __*.gif > test.gif")
# <codecell>
Datasets
Models
