-
Notifications
You must be signed in to change notification settings - Fork 11
/
watershed.py
203 lines (172 loc) · 6.52 KB
/
watershed.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
"""
## Functions for computing watershed segmentation
--------------------------------------------------
## Author: Callum J. Court.
## Email: cc889@cam.ac.uk
## Version: 1.0.0
--------------------------------------------------
## License: MIT
## Copyright: Copyright Callum Court & Batuhan Yildirim 2020, ICSG3D
-------------------------------------------------
"""
from itertools import product
import matplotlib.pyplot as plt
import numpy as np
from scipy.ndimage.morphology import distance_transform_edt
from skimage import filters, measure, morphology, segmentation
from viz import plot_points_3d
def get_background(S, kernel_size=1):
kernel = morphology.ball(kernel_size)
return morphology.dilation(S, kernel)
def get_foreground(S, kernel_size=1, erode=True):
if not erode:
return S
else:
kernel = morphology.ball(kernel_size)
return morphology.erosion(S, kernel)
def crop(a, bbox):
return a[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]]
def segment_nuclei(
binary,
species,
intensity,
wmin=8,
it=1,
max_iters=5,
min_convexity=0.8,
verbose=False,
):
""" Computes segmented form of species matrix using recursive watershed segmentation """
# Matrix for storing result
R = np.zeros(binary.shape)
binary = binary.astype(int)
# 1. Label the connected components
labels = measure.label(binary, connectivity=1)
seg_classes, seg_counts = np.unique(labels, return_counts=True)
seg_classes = np.array(
[seg_classes[i] for i in range(len(seg_classes)) if seg_counts[i] > 3]
)
seg_classes = seg_classes[seg_classes != 0]
if verbose:
print("\nIteration", it)
print("Classes", seg_classes)
print("Counts", seg_counts)
plot_points_3d(labels)
for cl in seg_classes:
if verbose:
print("Class", cl)
# Crop the images
binary_cl = np.where(labels == cl, labels, 0)
intensity_cl = np.where(labels == cl, intensity, 0)
species_cl = np.where(labels == cl, species, 0)
region = measure.regionprops(binary_cl, intensity_cl)
bbox = region[0].bbox
binary_bbox = crop(binary_cl, bbox)
intensity_bbox = crop(intensity_cl, bbox)
species_bbox = crop(species_cl, bbox)
chull = morphology.convex_hull_image(binary_bbox)
convexity = np.count_nonzero(binary_bbox) / np.count_nonzero(chull)
if verbose:
print("Convexity:", convexity)
if convexity >= min_convexity:
max_class = np.max(R)
R[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]] = np.where(
binary_bbox == cl,
max_class + 1,
R[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]],
)
continue
# Get the foreground, bg etc.
# Determine wether or not to erode
fg = get_foreground(binary_bbox)
bg = get_background(binary_bbox)
unknown = bg - fg
if verbose:
print("Segmenting")
plot_points_3d(fg)
# Markers for ws
markers = measure.label(fg)
markers += 1
markers[unknown == 1] = 0
# WS
wss = segmentation.watershed(binary_bbox, markers)
wss[wss == 1] = 0
max_class = np.max(R)
wss = wss + max_class # sub region with classes relabelled
wss[wss == max_class] = 0
nclasses = len(np.unique(wss)) - 1
if verbose:
print("WS", it, np.unique(wss, return_counts=True))
plot_points_3d(wss)
print(int(np.count_nonzero(wss) / wmin), nclasses)
# Determine wether or not to segment again on the basis of convexity and object counts
if (
int(np.count_nonzero(wss) / wmin) > len(np.unique(wss)) - 1
and it < max_iters
):
if verbose:
print("Segmenting again")
Rp = segment_nuclei(
wss,
species_bbox,
intensity_bbox,
it=it + 1,
verbose=verbose,
max_iters=max_iters,
min_convexity=min_convexity,
)
max_class = np.max(R)
Rp = Rp + max_class # sub region with classes relabelled
Rp[Rp == max_class] = 0
R[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]] = np.where(
Rp != 0, Rp, R[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]]
)
else:
R[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]] = np.where(
wss != 0,
wss,
R[bbox[0] : bbox[3], bbox[1] : bbox[4], bbox[2] : bbox[5]],
)
if verbose:
print(it, np.unique(R, return_counts=True))
return R
def majority_vote(seg_img, R, cl):
""" Majority vote of class cl in a region R in segmented image"""
binary_label_map = np.where(R == cl, seg_img, 0).astype(int)
if np.count_nonzero(binary_label_map) == 0:
return 0
unique, counts = np.unique(binary_label_map, return_counts=True)
unique_counts = sorted(list(zip(unique, counts)), key=lambda x: x[1])
unique_counts = [i for i in unique_counts if i[0] != 0]
specie = unique_counts[-1][0]
return specie
def centroids(seg_img, R):
""" Determine centroid of a region R in segmented image """
classes = np.unique(R)[1:]
atoms = []
means = []
xc = np.linspace(0, R.shape[0], R.shape[0] + 1)[:-1]
yc = np.linspace(0, R.shape[0], R.shape[0] + 1)[:-1]
zc = np.linspace(0, R.shape[0], R.shape[0] + 1)[:-1]
coords = np.array(list(product(xc, yc, zc))).reshape(32, 32, 32, 3)
seg_img_coords = np.concatenate([seg_img.reshape(32, 32, 32, 1), coords], axis=-1)
for cl in classes:
cmask = R == cl
smask = seg_img_coords[cmask]
specie = majority_vote(seg_img, R, cl)
if specie != 0:
means.append(np.mean(smask[:, 1:], axis=0))
atoms.append(specie)
return atoms, means
def watershed_clustering(M, S, Sb, max_iters=5, return_ws=False, verbose=False):
"""Determine centroids and species of atoms in the density/species matrices
Returns the atom z numbers and means in voxel coordinates"""
M = M.squeeze()
S = S.squeeze()
Sb = Sb.squeeze()
R = segment_nuclei(Sb, S, M, max_iters=max_iters, verbose=verbose)
atoms, means = centroids(S, R)
if return_ws:
return np.array(atoms), np.array(means), R
else:
return np.array(atoms), np.array(means)