-
Notifications
You must be signed in to change notification settings - Fork 2
/
from_data_to_predict.py
489 lines (415 loc) · 15.8 KB
/
from_data_to_predict.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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
from __future__ import print_function
import cv2
import numpy as np
import os
import dicom
import copy
from keras.utils import np_utils
from skimage.transform import rotate, warp_coords
from scipy.ndimage import map_coordinates
import random
from sklearn.cross_validation import train_test_split
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution2D, MaxPooling2D
from keras.utils import np_utils
from keras.optimizers import SGD
from keras.utils import np_utils
from skimage.morphology import remove_small_objects
from scipy import ndimage
from sklearn import metrics, metrics
from scipy.ndimage.measurements import label
from scipy.ndimage.morphology import binary_erosion, binary_dilation
DATA_PATH = os.getcwd()
MODE_TRAIN_PATH = '/data/metROI/train_txt'
MODE_TEST_PATH = '/data/metROI/test_txt'
MODE_SPINE_PATH = '/data/spineRegionROI/txt'
IMG_ROWS = 320
IMG_COLS = 320
RECEP_WEI = 30
RECEP_HEI = 70
nb_classes = 2
batch_size = 128
nb_classes = 2
nb_epoch = 1
# number of convolutional filters to use
nb_filters = 32
# size of pooling area for max pooling
nb_pool = 2
# convolution kernel size
nb_conv = 3
def get_input_and_output(source, spine_source):
file = open(source, 'r')
list = file.readlines()
test_array = np.empty((0, len(list)), int)
for line in list:
# print line
line = line.split(',')
line[len(line) - 1] = line[len(line) - 1].replace('\n', '')
line = map(int, line)
test_array = np.append(test_array, np.array([line]), axis=0)
file2 = open(spine_source, 'r')
list2 = file2.readlines()
test_array2 = np.empty((0, len(list2)), int)
for line in list2:
# print line
line = line.split(',')
line[len(line) - 1] = line[len(line) - 1].replace('\n', '')
line = map(int, line)
test_array2 = np.append(test_array2, np.array([line]), axis=0)
dir_array = source.split('/')
dir_name = dir_array[len(dir_array) - 1].split('_')
dir_image_file = DATA_PATH + '/data/Sagittal-segmentation/' + \
dir_name[0] + '/' + dir_name[1] + '/' + \
str(dir_name[2]).replace('txt', 'dcm')
medical_img = dicom.read_file(dir_image_file)
return medical_img, test_array, test_array2
def getPara(predict, true, threshold, resolution, windowsize):
(TP, FP, TN, FN, class_lable) = perf_measure(true, predict, threshold)
if((TP + FN) == 0):
TPR = 0
else:
TPR = np.float(TP) / (TP + FN)
class_lable = class_lable.astype(bool).reshape(250, 130)
true = true.astype(bool).reshape((250, 130))
num = 2
x = np.arange( -num , num+1, 1)
xx, yy = np.meshgrid( x, x )
struc = (xx * xx + yy * yy)<= num * num
class_lable = binary_dilation(class_lable, struc)
class_lable = binary_erosion(class_lable, struc)
# predict2 = remove_small_objects(class_lable, windowsize * resolution, in_place=False)
predict2 = remove_small_objects(class_lable, windowsize, in_place=False)
labeled_array1, num_features1 = label(predict2)
labeled_array2, num_features2 = label(true)
FP_num = num_features1 - num_features2
if FP_num < 0:
FP_num = 0
return TPR, FP_num
def perf_measure(y_actual, predict, threshold):
TP = 0
FP = 0
TN = 0
FN = 0
predict = transfer_prob(predict, threshold)
for i in range(len(predict)):
if y_actual[i] == predict[i] == 1:
TP += 1
for i in range(len(predict)):
if y_actual[i] == 0 and y_actual[i] != predict[i]:
FP += 1
for i in range(len(predict)):
if y_actual[i] == predict[i] == 0:
TN += 1
for i in range(len(predict)):
if y_actual[i] == 1 and y_actual[i] != predict[i]:
FN += 1
return(TP, FP, TN, FN, predict)
def transfer_prob(y_score, threshold):
y_result = []
for i in range(len(y_score)):
if y_score[i] >= threshold:
y_result.append(1)
else:
y_result.append(0)
return np.asarray(y_result)
def generate_plotnum(windowsize):
y_score = np.load('predicted_image.npy')
y_true = np.load('answer_image.npy')
reso = np.load('resolution.npy')
y_score = y_score[0:y_score.shape[0], 1]
y_true = y_true[0:y_true.shape[0], 1]
scores = []
trues = []
next_start = 0
for i in range(len(reso)):
ysize = 250
xsize = 130
scores.append(y_score[next_start: next_start + np.int(ysize * xsize)])
trues.append(y_true[next_start: next_start + np.int(ysize * xsize)])
next_start = np.int(next_start + ysize * xsize)
y_score = np.asarray(scores)
y_true = np.asarray(trues)
set_size = y_score.shape[0]
count = 1
thresholds = []
tmp = 0
for m in range(1, 10, 1):
tmp += 1
# print(m/np.float(100))
thresholds.append(m / np.float(100))
for i in range(1, 10, 1):
thresholds.append(i / np.float(10))
for m in range(90, 100, 1):
thresholds.append(m / np.float(100))
tmp = 0
for m in range(900, 1000, 1):
tmp += 1
if(tmp % 10 == 0):
thresholds.append(m / np.float(1000))
thresholds = sorted(thresholds, reverse=True)
thresholds = np.asarray(thresholds)
TPR_list = []
FP_num_list = []
for t in range(1, thresholds.size):
tpr_sum = 0
fp_sum = 0
for i in range(set_size):
TPR, FP_num = getPara(y_score[i], y_true[i].astype(np.int), thresholds[t], reso[i], windowsize)
tpr_sum += TPR
fp_sum += FP_num
print(TPR, FP_num)
print(t, thresholds.size, thresholds[t], fp_sum / np.float(set_size ), tpr_sum / np.float(set_size ))
TPR_list.append(tpr_sum / np.float(set_size ))
FP_num_list.append(fp_sum / np.float(set_size ))
#print( t, thresholds.size, fp_sum/np.float(set_size ), tpr_sum/np.float(set_size ))
#TPR_list.append(tpr_sum/np.float(set_size ))
#FP_num_list.append(fp_sum/np.float(set_size ))
# np.save('./resolution/TPR_list_'+ str(windowsize) +'.npy', TPR_list)
# np.save('./resolution/FP_num_list_'+ str(windowsize) +'.npy', FP_num_list)
np.save('TPR_list_'+ str(windowsize) +'.npy', TPR_list)
np.save('FP_num_list_'+ str(windowsize) +'.npy', FP_num_list)
nb_classes = 2
RECEP_HEI = 70
RECEP_WEI = 30
IMG_COLS = 320
IMG_ROWS = 320
ratio_of_outside = 0.2
count_img = 0
count = 0
index = 0
count_pos = 0
class_lable = 2
x_start = RECEP_WEI
y_start = RECEP_HEI
y_max = IMG_COLS
x_max = IMG_ROWS / 2
TRAIN_DATA_TXT_PATH = DATA_PATH + MODE_TRAIN_PATH
TRAIN_SPINE_TXT_PATH = DATA_PATH + MODE_SPINE_PATH
IMG_LIST = os.listdir(TRAIN_DATA_TXT_PATH)
TRAIN_NUM = len(IMG_LIST)
X = []
Y = []
TRAIN_IN = []
TRAIN_OUT = []
for name in IMG_LIST:
med, test_array, spine_region = get_input_and_output(
TRAIN_DATA_TXT_PATH + '/' + name, TRAIN_SPINE_TXT_PATH + '/' + name)
med_img = med.pixel_array.astype('float')
test_array = test_array.astype('float')
spine_region = spine_region.astype('float')
print(med_img.shape)
X.append(med_img)
Y.append([test_array, spine_region, med.PixelSpacing[0]])
X.append(cv2.flip(med_img, 1))
Y.append([cv2.flip(test_array, 1), cv2.flip(
spine_region, 1), med.PixelSpacing[0]])
X.append(cv2.flip(med_img, 0))
Y.append([cv2.flip(test_array, 0), cv2.flip(
spine_region, 0), med.PixelSpacing[0]])
X.append(cv2.flip(med_img, -1))
Y.append([cv2.flip(test_array, -1),
cv2.flip(spine_region, -1), med.PixelSpacing[0]])
X = np.asarray(X)
Y = np.asarray(Y)
# Xtrain, Xtest, Ytrain, Ytest = train_test_split(
# X, Y, test_size=0.25, random_state=42)
# Xtrain, Xtest, Ytrain, Ytest = train_test_split(
# X, Y, test_size=0.25)
# Xtrain = X[np.int((1./4) * len(X)) : len(X)]
# Ytrain = Y[np.int((1./4) * len(X)) : len(Y)]
# Xtest = X[0: np.int((1./4) * len(X)) ]
# Ytest = Y[0: np.int((1./4) * len(Y)) ]
Xtrain = X
Ytrain = Y
# shuffle training data to get better performance
for i in range(Xtrain.shape[0]):
med_img, test_array, spine_region = Xtrain[
i], Ytrain[i][0], Ytrain[i][1]
test_array = (test_array > 0.5).astype('int')
spine_region = (spine_region > 0.5).astype('int')
test_array = test_array + 0.001
spine_region = spine_region + 0.001
med_img = cv2.resize(med_img, (IMG_ROWS, IMG_COLS),
interpolation=cv2.INTER_AREA)
test_array = cv2.resize(
test_array, (IMG_ROWS, IMG_COLS), interpolation=cv2.INTER_AREA)
spine_region = cv2.resize(
spine_region, (IMG_ROWS, IMG_COLS), interpolation=cv2.INTER_AREA)
cropImg1 = med_img[0:IMG_ROWS, IMG_COLS / 4: 3 * IMG_ROWS / 4]
cropImg2 = test_array[0:IMG_ROWS, IMG_COLS / 4: 3 * IMG_ROWS / 4]
cropImg3 = spine_region[0:IMG_ROWS, IMG_COLS / 4: 3 * IMG_ROWS / 4]
sum_pix = cropImg1.shape[0] * cropImg1.shape[1]
spine_pix = 0
pos_pix = 0
count_inside = 0
count_outside = 0
for line in cropImg3:
for ele in line:
if(ele > 0.02):
spine_pix += 1
for line in cropImg2:
for ele in line:
if(ele > 0.02):
pos_pix += 1
inside_num = (1 - ratio_of_outside) * pos_pix
outside_num = ratio_of_outside * pos_pix
inside_len = (int(round(spine_pix / inside_num)))
outside_len = int(round((sum_pix - spine_pix) / outside_num))
for i in range(y_start, y_max):
for m in range(x_start, x_max):
class_lable = cropImg2[i - RECEP_HEI / 2][m - RECEP_WEI / 2]
inside_lable = cropImg3[i - RECEP_HEI / 2][m - RECEP_WEI / 2]
if (class_lable > 0.02):
region = cropImg1[i - RECEP_HEI: i, m - RECEP_WEI: m]
TRAIN_IN.append(region.reshape(1, RECEP_HEI, RECEP_WEI))
TRAIN_OUT.append(
np_utils.to_categorical([1], nb_classes)[0])
count_pos += 1
index += 1
# print(index)
elif(inside_lable > 0.02 and class_lable <= 0.02):
count_inside += 1
if(count_inside % inside_len == 0):
region = cropImg1[i - RECEP_HEI: i, m - RECEP_WEI: m]
TRAIN_IN.append(region.reshape(
1, RECEP_HEI, RECEP_WEI))
TRAIN_OUT.append(
np_utils.to_categorical([0], nb_classes)[0])
index += 1
elif(inside_lable < 0.02 and class_lable <= 0.02):
count_outside += 1
if(count_outside % outside_len == 0):
region = cropImg1[i - RECEP_HEI: i, m - RECEP_WEI: m]
TRAIN_IN.append(region.reshape(
1, RECEP_HEI, RECEP_WEI))
TRAIN_OUT.append(
np_utils.to_categorical([0], nb_classes)[0])
index += 1
count += 1
if(index % 1000 == 0):
print('Processed training set:' + str(index))
count_img += 1
print('Processed images:' + str(count_img) + '.Positive num:' +
str(count_pos) + '.Total size:' + str(index))
X_train = np.asarray(TRAIN_IN, dtype=np.float)
Y_train = np.asarray(TRAIN_OUT, dtype=np.float)
print('Normalizing training data...')
for i in range(X_train.shape[0]):
for m in range(X_train.shape[1]):
mean = np.mean(X_train[i][m]) # mean for data centering
std = np.std(X_train[i][m]) # std for data normalization
X_train[i][m] = (X_train[i][m] - mean) / std
#print('Normalizing testing data...')
# for i in range(X_test.shape[0]):
# for m in range(X_test.shape[1]):
# mean = np.mean(X_test[i][m]) # mean for data centering
# std = np.std(X_test[i][m]) # std for data normalization
# X_test[i][m] = (X_test[i][m] -mean)/std
print('X_train shape:', X_train.shape)
#training and predicting part
model = Sequential()
model.add(Convolution2D(32, 3, 3, border_mode='same',
input_shape=(1, RECEP_HEI, RECEP_WEI)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Convolution2D(32, 3, 3))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Convolution2D(32, 3, 3))
model.add(Activation('relu'))
model.add(Flatten())
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(nb_classes))
model.add(Activation('softmax'))
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy',
optimizer=sgd,
metrics=['accuracy'])
model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
verbose=1, shuffle=True, validation_split=0.25)
model.save_weights('weights.hdf5', overwrite=True)
#predicted_result = model.predict_classes(X_test,verbose = 1)
#np.save('result.npy', predicted_result)
print('Model trained, saving weights and ready to prediction.')
#generate testing data
predict_train_in = []
predict_train_out = []
count_pos = 0
count_img = 0
index = 0
TEST_DATA_TXT_PATH = DATA_PATH + MODE_TEST_PATH
IMG_LIST = os.listdir(TEST_DATA_TXT_PATH)
Xtest = []
Ytest = []
Resolution = []
for name in IMG_LIST:
med, test_array, spine_region = get_input_and_output(
TEST_DATA_TXT_PATH + '/' + name, TRAIN_SPINE_TXT_PATH + '/' + name)
med_img = med.pixel_array.astype('float')
test_array = test_array.astype('float')
print(med_img.shape)
Xtest.append(med_img)
Ytest.append([test_array, med.PixelSpacing[0]])
Xtest = np.asarray(Xtest)
Ytest = np.asarray(Ytest)
Resolution.append(med.PixelSpacing[0])
Resolution = np.asarray(Resolution)
for i in range(Xtest.shape[0]):
med_img, test_array = Xtest[i], Ytest[i][0]
med_img = cv2.resize(med_img, (320, 320),
interpolation=cv2.INTER_AREA)
test_array = test_array + 0.001
test_array = cv2.resize(
test_array, (320, 320), interpolation=cv2.INTER_AREA)
cropImg1 = med_img[0:320, 320 / 4: 3 * 320 / 4]
cropImg2 = test_array[0:320, 320 / 4: 3 * 320 / 4]
y_max = 320
x_max = 320 / 2
for i in range(y_start, y_max):
for m in range(x_start, x_max):
region = cropImg1[i - RECEP_HEI: i, m - RECEP_WEI: m]
class_lable = cropImg2[
i - RECEP_HEI / 2][m - RECEP_WEI / 2]
if(class_lable <= 0.002):
predict_train_in.append(region.reshape(
1, RECEP_HEI, RECEP_WEI))
predict_train_out.append(
np_utils.to_categorical([0], nb_classes)[0])
index += 1
# print(index)
else:
predict_train_in.append(region.reshape(
1, RECEP_HEI, RECEP_WEI))
predict_train_out.append(
np_utils.to_categorical([1], nb_classes)[0])
count_pos += 1
index += 1
# print(index)
count += 1
if(index % 1000 == 0):
print('Processed training set:' + str(index))
count_img += 1
print('Processed images:' + str(count_img) + '.Positive num:' +
str(count_pos) + '.Total size:' + str(index))
predict_train_in = np.asarray(predict_train_in, dtype=np.float)
predict_train_out = np.asarray(predict_train_out, dtype=np.float)
print('Normalizing testing data...')
for i in range(predict_train_in.shape[0]):
for m in range(predict_train_in.shape[1]):
mean = np.mean(predict_train_in[i][m]) # mean for data centering
std = np.std(predict_train_in[i][m]) # std for data normalization
predict_train_in[i][m] = (predict_train_in[i][m] - mean) / std
print('Predicting...')
predict_result = model.predict_proba(predict_train_in)
np.save('answer_image' + '.npy', predict_train_out)
np.save('predicted_image' + '.npy', predict_result)
np.save('input_image'+'.npy',Xtest)
np.save('resolution' + '.npy', Resolution)
#evaluate part
generate_plotnum(300)
generate_plotnum(400)
generate_plotnum(500)
generate_plotnum(600)