BCRNet/extraction.py at main · JoeSu666/BCRNet · GitHub

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
import openslide
import matplotlib.pyplot as plt
import numpy as np
import cv2
from skimage.color import rgb2gray
import glob
import os
from os.path import join
import argparse
from tqdm import tqdm


parser = argparse.ArgumentParser(description='save patch coordinates into npy. Coords under certain level 0. \
                                                (IN (row, col) INDEXING, NOT (x, y))')
parser.add_argument('-p', default=128, type=int, help='Patch size')
parser.add_argument('-s', default=128, type=int, help='Stride')
parser.add_argument('-l', default=0, type=int, help='Magnification level')
parser.add_argument('--save', default='data', type=str, help='Saving directory')
parser.add_argument('--data',
                    type=str, help='Data directory')
parser.add_argument('--mask', default='./data/masks',
                    type=str, help='mask directory')
parser.add_argument('--type', default=None,
                    type=str, help='None (default, tissue region), normal or tumor')
parser.add_argument('--code', default='newcases', type=str, help='code')
parser.add_argument('--start', default=0, type=int, help='start')


args = parser.parse_args()

def main(args=args):
    namelist = sorted(glob.glob(join(args.mask, '*', '*.png')))[args.start:]

    save_dir = join(args.save, 'pts', args.code+'l'+str(args.l)+'p'+str(args.p)+'s'+str(args.s))
    if not os.path.exists(save_dir):
        os.mkdir(save_dir)
        os.mkdir(join(save_dir, 'negative'))
        os.mkdir(join(save_dir, 'positive'))

    for name in namelist:
        # read mask
        pid = name.split('/')[-1].split('.')[0]

        dataname = join(args.data, pid+'.tif')


        if name.split('/')[-2] == 'positive':
            label = 1
        elif name.split('/')[-2] == 'negative':
            label = 0

        mask = cv2.imread(name, 0)

        print('***********')
        print(pid)

        # load wsi
        with openslide.OpenSlide(dataname) as fp:
            w, h = fp.level_dimensions[args.l]
            w0, h0 = fp.dimensions

            #extract coords
            pts = extract(fp, w, h, w0, h0, args.p, args.s, mask, args.l)

        # save pts

        np.save(join(save_dir, name.split('/')[-2], pid+'.npy'), pts)


def extract(fp, w, h, w0, h0, ps, stride, mask, level):

    boundaries = cv2.findContours(mask,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
    boundaries = boundaries[0]
    minw, minh, maxw, maxh = mask.shape[1], mask.shape[0], 0, 0
    for i in range(len(boundaries)):
        b = np.squeeze(boundaries[i], 1)
        if np.min(b[:, 0]) < minw:
            minw = np.min(b[:, 0])
        if np.min(b[:, 1]) < minh:
            minh = np.min(b[:, 1])
        if np.max(b[:, 0]) > maxw:
            maxw = np.max(b[:, 0])
        if np.max(b[:, 1]) > maxh:
            maxh = np.max(b[:, 1])

    psy = ps * mask.shape[0] / float(h)
    psx = ps * mask.shape[1] / float(w)

    stride = stride * mask.shape[0] / float(h)

    # Grid of points
    ys = np.arange(minh, maxh, stride)
    xs = np.arange(minw, maxw, stride)

    [ys, xs] = np.meshgrid(ys, xs, indexing='ij')
    ys = ys.reshape((-1, 1))
    xs = xs.reshape((-1, 1))
    pts = np.concatenate([ys, xs], 1)

    # Here's where we put things
    bag = np.zeros((ps, ps, 3, pts.shape[0]), 'uint8')
    keep = np.zeros((pts.shape[0],), dtype=bool)

    for p in range(pts.shape[0]):
        # Query pts
        rx = pts[p, 1]
        ry = pts[p, 0]

        # Pass first one
        if p == 0:
            continue

        # Checks if inside the mask image
        if mask.shape[1] > rx + psx and mask.shape[0] > ry + psy and \
        mask[round(ry+psy/2), round(rx+psx/2)] == 255 and \
        mask[round(ry), round(rx)] == 255 and \
        mask[round(ry+psy), round(rx)] == 255 and \
        mask[round(ry), round(rx+psx)] == 255:


            im = np.array(fp.read_region((round(rx*float(w0)/mask.shape[1]), round(ry*float(h0)/mask.shape[0])), level, (ps, ps)).convert('RGB'))

            if np.mean(cv2.cvtColor(im, cv2.COLOR_RGB2GRAY) < 240) > 0.75:
                pts[p] = np.array([round(ry*float(h0)/mask.shape[0]), round(rx*float(w0)/mask.shape[1])])

                keep[p] = True

    print('{} patches in bounding box'.format(pts.shape[0]))
    pts = pts[keep]
    print('Found {} tissue patches'.format(pts.shape[0]))

    return np.around(pts).astype('int')

if __name__ == '__main__':
    main()