-
Notifications
You must be signed in to change notification settings - Fork 5
/
feature_extraction.py
161 lines (139 loc) · 7.08 KB
/
feature_extraction.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
import os
import argparse
import time
import random
import numpy as np
import pandas as pd
from keras import models
import segmentation_models as sm
from segmentation_models import get_preprocessing
from preprocess.prep import Preprocess
from utils import *
from stage2 import *
import staintools
import openslide
from openslide.deepzoom import DeepZoomGenerator
from PIL import Image
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='parser')
parser.add_argument('--seed', type=int, default=42)
parser.add_argument('--patch_size', type=int, default=256)
parser.add_argument('--is_preprocessed',
type=lambda x: True if x == 'True' else False,
default=True)
parser.add_argument('--model_name', type=str, default='fpn_model')
parser.add_argument('--model_weight', type=str, default='2_fold_fpn_best_model.h5')
parser.add_argument('--ckpt_dir', type=str, default='./data/volume/model/')
parser.add_argument('--heatmap_dir', type=str, default='./data/volume/heatmap/')
parser.add_argument('--feature_dir', type=str, default='./data/volume/feature/')
parser.add_argument('--patches_dir', type=str, default='./data/volume/patches/rescale/')
args = parser.parse_args()
TRAIN_DIR, LABEL_PATH = './data/train', './data/train/label.csv'
MODEL_NAME, HEATMAP_DIR, FEATURE_DIR, PATCHES_DIR = args.model_name, args.heatmap_dir, args.feature_dir, args.patches_dir
random.seed(args.seed)
np.random.seed(args.seed)
# check isdir
set_directory(CKPT_DIR, MODEL_NAME)
make_directory(HEATMAP_DIR)
make_directory(FEATURE_DIR)
make_directory(PATCHES_DIR)
#load model
MODEL_PATH = args.ckpt_dir + args.model_name + '/' + args.model_weight
model = models.load_model(
MODEL_PATH,
custom_objects={
'binary_crossentropy_plus_jaccard_loss': sm.losses.bce_jaccard_loss,
'iou_score': sm.metrics.iou_score,
'f1-score': sm.metrics.f1_score
}
)
print(MODEL_PATH,'Model loaded.')
# set preprocess
preprocess_input = get_preprocessing('resnet34')
preprocess = Preprocess(patch_size=PATCH_SIZE, mode='inference', server='kakao')
TARGET_NORM_PATH = './preprocess/target_norm.png'
normalizer = stain_norm_func(TARGET_NORM_PATH)
slide_pathes = sorted(os.listdir(preprocess.slide_dir))
stain_patches_save_path, phase = stain_patch_dir(PATCHES_DIR, slide_pathes)
start_time = time.time()
full_feature_list = []
for i, slide_path in enumerate(slide_pathes):
current_save_dir = stain_patches_save_path + slide_path[:-4] + '/' # ex) '/data/volume/patches/rescale/test1/slide_001/'
if phase == 'test1' and i <= 60: # AMC dataset
full_slide_path = preprocess.slide_dir + slide_path
else : # SNU dataset
full_slide_path = '/data/test/level0/'+ slide_path +'.mrxs'
print(current_save_dir)
if IS_PREPROCESSED :
stain_patches_names = sorted(os.listdir(current_save_dir))
else :
make_directory(current_save_dir)
with openslide.open_slide(full_slide_path) as slide:
if slide.dimensions[1] < 20000:
print('AMC data!')
patch_size = 256
else :
print('SNU data!')
patch_size = 290
slide_tiles = DeepZoomGenerator(slide, tile_size = patch_size, overlap = 0 , limit_bounds = False)
if patch_size == 290:
output_preds = np.zeros((int((slide.dimensions[1] / 8 + 1)/1.13), int((slide.dimensions[0] / 8 + 1)/1.13)))
else: ### snu resolution
output_preds = np.zeros((slide.dimensions[1],slide.dimensions[0]))
print('output_preds shape : ',output_preds.shape)
samples, _ = preprocess.find_patches_from_slide(slide_path = full_slide_path, mask_path = None, patch_size = patch_size)
print(samples.is_tissue.value_counts())
cnt = 0
for idx, batch_sample in samples.iterrows():
is_tissue = batch_sample.is_tissue
x,y = batch_sample.tile_loc[::-1]
if is_tissue :
if patch_size == 290:
img = slide_tiles.get_tile(slide_tiles.level_count-1 -3,(x,y)) # SNU -> level 3
else :
img = slide_tiles.get_tile(slide_tiles.level_count-1,(x,y))
if (img.size == (patch_size, patch_size)):
if IS_PREPROCESSED:
try :
full_stain_patches_path = current_save_dir + str(idx) + '.png'
cnt += 1
img = Image.open(full_stain_patches_path)
X = np.array(img, dtype =np.uint8)
except:
X = np.zeros((256,256,3))
else :
if img.size[0] == 290 :
img = img.resize((256,256))
X = np.array(img, dtype = np.uint8)
try :
X = staintools.LuminosityStandardizer.standardize(X)
X = normalizer.transform(X)
x_img = Image.fromarray(X)
x_img.save(current_save_dir + str(idx) + '.png')
except:
X = np.zeros((256, 256,3))
else :
try :
full_stain_patches_path = current_save_dir + str(idx) + '.png'
cnt += 1
img = Image.open(full_stain_patches_path)
X = np.array(img, dtype =np.uint8)
except :
X = np.zeros((256,256, 3))
X = X.astype(np.float32)
X = preprocess_input(X)
pred_j = predict_from_model(X, model)
'''fill output_preds : full heatmap'''
new_x, new_y = batch_sample.tile_loc[0] * 256, batch_sample.tile_loc[1] * 256
output_preds[new_x:new_x+256, new_y:new_y+256] = pred_j
'''make different level heatmaps / input : full size heatmap / output : different scale heatmap'''
heatmaps_list = make_different_level_heatmaps(output_preds)
'''extract feature from different level heatmaps'''
feature_list, feature_name_list = extract_feature_from_heatmaps(heatmaps_list)
if i == 0:
print(feature_name_list)
print(feature_list)
full_feature_list.append(feature_list)
pd_feature = pd.DataFrame(np.array(full_feature_list), columns=feature_name_list)
save_feature_path = FEATURE_DIR + MODEL_NAME +'_' +phase+'_feature.csv'
pd_feature.to_csv(save_feature_path)