[1ee192]: / core / util.py

Download this file

502 lines (411 with data), 18.9 kB

  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
489
490
491
492
493
494
495
496
497
498
499
500
501
# -*- coding: utf-8 -*-
"""
Functions and operations for performance visualization and result store,
some of which are not used in the current situation.
@author: Xinzhe Luo
"""
from __future__ import print_function, division, absolute_import, unicode_literals
import numpy as np
import os
import logging
from PIL import Image
import nibabel as nib
import matplotlib.pyplot as plt
from sklearn.metrics import roc_curve, auc
from core import image_util
import tensorflow as tf
def post_process(data):
"""
tf.image operations
:param data: numpy array, expected dim [batch_size, height, width, channels]
:return: data with same dimension
"""
# data = np.squeeze(data, axis=-1).transpose(1, 2, 0)
data = tf.convert_to_tensor(data, dtype=tf.float32)
data = tf.image.random_contrast(data, lower=0.7, upper=1)
# labels = tf.image.random_brightness(labels, 0.2)
return data
def plot_prediction(x_test, y_test, prediction, save=False):
test_size = x_test.shape[0]
fig, ax = plt.subplots(test_size, 3, figsize=(12, 12), sharey=True, sharex=True)
x_test = crop_to_shape(x_test, prediction.shape)
y_test = crop_to_shape(y_test, prediction.shape)
ax = np.atleast_2d(ax)
for i in range(test_size):
cax = ax[i, 0].imshow(x_test[i])
plt.colorbar(cax, ax=ax[i, 0])
cax = ax[i, 1].imshow(y_test[i, ..., 1])
plt.colorbar(cax, ax=ax[i, 1])
pred = prediction[i, ..., 1]
pred -= np.amin(pred)
pred /= np.amax(pred)
cax = ax[i, 2].imshow(pred)
plt.colorbar(cax, ax=ax[i, 2])
if i == 0:
ax[i, 0].set_title("x")
ax[i, 1].set_title("y")
ax[i, 2].set_title("pred")
fig.tight_layout()
if save:
fig.savefig(save)
else:
fig.show()
plt.show()
def to_rgb(img):
"""
Converts the given array into a RGB image. If the number of channels is not
3 the array is tiled such that it has 3 channels. Finally, the values are
rescaled to [0,255)
:param img: the array to convert [n, nx, ny, nz, channels]
:returns img: the rgb image [n, nx, ny, nz, 3]
"""
if len(img.shape) < 5:
img = np.expand_dims(img, axis=-1)
channels = img.shape[-1]
if channels < 3:
img = np.tile(img, 3)
img[np.isnan(img)] = 0
for k in range(np.shape(img)[3]):
st = img[:, :, :, k, ]
if np.amin(st) != np.amax(st):
st -= np.amin(st)
st /= np.amax(st)
st *= 255
return img
def to_rgb_2d(img):
"""
Converts the given array into a RGB image. If the number of channels is not
3 the array is tiled such that it has 3 channels. Finally, the values are
rescaled to [0,255)
:param img: the array to convert [n, nx, ny, channels]
:returns img: the rgb image [n, nx, ny, 3]
"""
if len(img.shape) < 4:
img = np.expand_dims(img, axis=-1)
channels = img.shape[-1]
if channels < 3:
img = np.tile(img, 3)
img[np.isnan(img)] = 0
for k in range(np.shape(img)[3]):
st = img[:, :, :, k]
if np.amin(st) != np.amax(st):
st -= np.amin(st)
st /= np.amax(st)
st *= 255
return img
'''
def crop_to_shape(data, shape):
"""
Crops the array to the given image shape by removing the border
(expects a tensor of shape [batches, nx, ny, channels].
:param data: the array to crop, shape=[n, nx, ny, nz, n_class]
:param shape: the target shape
"""
assert np.all(data.shape>=shape), "The shape of array to be cropped is smaller than the target shape."
offset1 = (data.shape[1] - shape[1])//2
offset2 = (data.shape[2] - shape[2])//2
offset3 = (data.shape[3] - shape[3])//2
return data[:, offset1:(offset1+shape[1]), offset2:(offset2+shape[2]), offset3:(offset3+shape[3]), :]
'''
def crop_to_shape(data, shape):
"""
Crops the array to the given image shape by removing the border
(expects a tensor of shape [batches, nx, ny, channels]).
:param data: the array to crop, shape=[n, nx, ny, nz, n_class]
:param shape: the target shape
"""
assert np.all(data.shape[1:4] >= shape[1:4]), "The shape of array to be cropped is smaller than the target shape."
offset0 = (data.shape[1] - shape[1]) // 2
offset1 = (data.shape[2] - shape[2]) // 2
offset2 = (data.shape[3] - shape[3]) // 2
remainder0 = (data.shape[1] - shape[1]) % 2
remainder1 = (data.shape[2] - shape[2]) % 2
remainder2 = (data.shape[3] - shape[3]) % 2
if (data.shape[1] - shape[1]) == 0 and (data.shape[2] - shape[2]) == 0 and (data.shape[3] - shape[3]) == 0:
return data
elif (data.shape[1] - shape[1]) != 0 and (data.shape[2] - shape[2]) == 0 and (data.shape[3] - shape[3]) == 0:
return data[:, offset0:(-offset0 - remainder0), ]
elif (data.shape[1] - shape[1]) == 0 and (data.shape[2] - shape[2]) != 0 and (data.shape[3] - shape[3]) == 0:
return data[:, :, offset1:(-offset1 - remainder1), ]
elif (data.shape[1] - shape[1]) == 0 and (data.shape[2] - shape[2]) == 0 and (data.shape[3] - shape[3]) != 0:
return data[:, :, :, offset2:(-offset2 - remainder2), ]
elif (data.shape[1] - shape[1]) != 0 and (data.shape[2] - shape[2]) != 0 and (data.shape[3] - shape[3]) == 0:
return data[:, offset0:(-offset0 - remainder0), offset1:(-offset1 - remainder1), ]
elif (data.shape[1] - shape[1]) == 0 and (data.shape[2] - shape[2]) != 0 and (data.shape[3] - shape[3]) != 0:
return data[:, :, offset1:(-offset1 - remainder1), offset2:(-offset2 - remainder2), ]
elif (data.shape[1] - shape[1]) != 0 and (data.shape[2] - shape[2]) == 0 and (data.shape[3] - shape[3]) != 0:
return data[:, offset0:(-offset0 - remainder0), :, offset2:(-offset2 - remainder2), ]
else:
return data[:, offset0:(-offset0 - remainder0), offset1:(-offset1 - remainder1),
offset2:(-offset2 - remainder2), ]
def crop_to_shape_2d(data, shape):
"""
Crops the array to the given image shape by removing the border
(expects a tensor of shape [batches, nx, ny, channels]).
:param data: the array to crop, shape=[n, nx, ny, n_class/channels]
:param shape: the target shape
"""
assert np.all(data.shape[1:3] >= shape[1:3]), "The shape of array to be cropped is smaller than the target shape."
offset0 = (data.shape[1] - shape[1]) // 2
offset1 = (data.shape[2] - shape[2]) // 2
remainder0 = (data.shape[1] - shape[1]) % 2
remainder1 = (data.shape[2] - shape[2]) % 2
if (data.shape[1] - shape[1]) == 0 and (data.shape[2] - shape[2]) == 0:
return data
elif (data.shape[1] - shape[1]) != 0 and (data.shape[2] - shape[2]) == 0:
return data[:, offset0:(-offset0 - remainder0), ]
elif (data.shape[1] - shape[1]) == 0 and (data.shape[2] - shape[2]) != 0:
return data[:, :, offset1:(-offset1 - remainder1), ]
elif (data.shape[1] - shape[1]) != 0 and (data.shape[2] - shape[2]) != 0:
return data[:, offset0:(-offset0 - remainder0), offset1:(-offset1 - remainder1), ]
else:
return data[:, offset0:(-offset0 - remainder0), offset1:(-offset1 - remainder1), ]
def crop_to_label_center(image, label, center_roi):
"""
crop image according to the center of label
:param image: input image, expected shape [height, width, channels=1]
:param label: input label, expected shape [height, width, channels=1]
:param center_roi: an shape array, expected dim [roi_height, roi_width, 1]
:return: desired shape of image and labels
"""
assert np.all(np.array(center_roi) <= np.array(label.shape)), print(
'Patch size exceeds dimensions.')
center = compute_center(label)
where_are_nan = np.isnan(center)
center[where_are_nan] = int(label.shape[0] // 2)
x = np.array([center[i][0] for i in range(label.shape[-1])]).astype(np.int)
y = np.array([center[i][1] for i in range(label.shape[-1])]).astype(np.int)
x = x[0]
y = y[0]
beginx = x - center_roi[0]
beginy = y - center_roi[1]
endx = x + center_roi[0]
endy = y + center_roi[1]
gt = label[beginx:endx, beginy:endy, :]
img = image[beginx:endx, beginy:endy, :]
return img, gt
def compute_center(label):
points = np.where(label > 0)
return np.array([[np.average(points[0][points[2] == j]), np.average(points[1][points[2] == j])] for j in
range(label.shape[-1])])
def pad_to_shape(data, shape):
"""
Pad the array to the given shape by the edge values
"""
assert np.all(data.shape <= shape), "The shape of array to be padded is larger than the target shape."
offset1 = (shape[1] - data.shape[1]) // 2
offset2 = (shape[2] - data.shape[2]) // 2
offset3 = (shape[3] - data.shape[3]) // 2
remainder1 = (shape[1] - data.shape[1]) % 2
remainder2 = (shape[2] - data.shape[2]) % 2
remainder3 = (shape[3] - data.shape[3]) % 2
return np.pad(data, (
(0, 0), (offset1, offset1 + remainder1), (offset2, offset2 + remainder2), (offset3, offset3 + remainder3),
(0, 0)),
'edge')
def combine_img_prediction(data, gt, pred):
"""
Combines the data, gt and the prediction into one rgb image for each class
to save the images for one patient and one class
:param data: the data tensor
:param gt: the ground truth tensor
:param pred: the prediction tensor
:returns img: the concatenated rgb image
"""
# print("Before combine the image and prediction")
# print("data.shape", data.shape) -> [1, 240, 240, 16, 1]
# print("gt.shape", gt.shape) -> [1, 240, 240, 16, 4]
# print("pred.shape", pred.shape) -> [1, 240, 240, 16, 4]
ny = pred.shape[2]
imgs = []
# print(crop_to_shape(data, pred.shape).shape) -> [1, 240, 240, 16, 1]
# print(to_rgb(crop_to_shape(data, pred.shape)).shape) -> [1, 240, 240, 16, 3]
# print(to_rgb(crop_to_shape(data, pred.shape)).transpose((0, 1, 3, 2, 4)).shape) -> [1, 240, 16, 240, 3]
# print(to_rgb(crop_to_shape(data, pred.shape)).transpose((0, 1, 3, 2, 4)).reshape(-1, ny, 3, order='F').shape)
# -> [3840 240, 3]
for k in range(gt.shape[-1]): # circle from classes
img = np.concatenate(
(to_rgb(crop_to_shape(data, pred.shape)).transpose((0, 1, 3, 2, 4)).reshape(-1, ny, 3, order='F'),
to_rgb(crop_to_shape(gt[..., k], pred.shape)).transpose((0, 1, 3, 2, 4)).reshape(-1, ny, 3, order='F'),
to_rgb(pred[..., k]).transpose((0, 1, 3, 2, 4)).reshape(-1, ny, 3, order='F')), axis=1)
imgs.append(img)
# print("len of imgs", len(imgs)) -> 4
# print("element of imgs", imgs[0].shape) -> [3840, 720, 3]
return imgs
def save_image(imgs, path):
"""
Writes the image to disk
:param imgs: the rgb images to save
:param path: the target path
"""
for i in range(len(imgs)): # circle from classes
img_path = os.path.join(os.path.split(path)[0], 'class%d_' % i + os.path.split(path)[1])
Image.fromarray(imgs[i].round().astype(np.uint8)).save(img_path, 'PNG', dpi=[300, 300], quality=95)
def save_prediction(data, gt, predictions, path):
"""
Combine each prediction and the corresponding ground truth as well as input into one image and save as png files.
:param data: list of input raw images
:param gt: list of ground truth images
:param predictions: list of predictions
:param path: has the form of 'directory'
"""
assert len(data) == len(gt) and len(gt) == len(predictions), print('Numbers of images are not equal.')
abs_pred_path = os.path.abspath(path)
if not os.path.exists(abs_pred_path):
logging.info("Allocating '{:}'".format(abs_pred_path))
os.makedirs(abs_pred_path)
for k in range(len(data)): # circle from patients
pred = np.where(np.equal(np.max(predictions[k], -1, keepdims=True), predictions[k]),
np.ones_like(predictions[k]),
np.zeros_like(predictions[k]))
save_image(combine_img_prediction(data[k], gt[k], pred), os.path.join(path, 'sub%s.png' % k))
def save_prediction_1(predictions, affine, path):
"""
Save the predictions into nibabel images.
Predictions are pre-processed by setting the maximum along classes to be a certain intensity specific to its class.
:param predictions: list of predictions
:param affine: list of corresponding coordinates
:param path: has the form of 'directory'
"""
abs_pred_path = os.path.abspath(path)
if not os.path.exists(abs_pred_path):
logging.info("Allocating '{:}'".format(abs_pred_path))
os.makedirs(abs_pred_path)
for k in range(len(predictions)):
pred = np.squeeze(predictions[k])
intensity = np.tile(np.arange(0, 1000, 999 // (pred.shape[-1] - 1)), np.concatenate((pred.shape[:-1], [1])))
mask = np.equal(np.max(pred, -1, keepdims=True), pred)
img = nib.Nifti1Image(np.sum(mask * intensity, axis=-1).astype(np.float32).transpose((1, 0, 2)),
affine=affine[k])
nib.save(img, os.path.join(path, 'sub%s.nii.gz' % k))
def save_prediction_2(predictions, path):
"""
Save the predictions into numpy array.
:param predictions: list of predictions
:param path: has the form of 'directory'
"""
abs_pred_path = os.path.abspath(path)
if not os.path.exists(abs_pred_path):
logging.info("Allocating '{:}'".format(abs_pred_path))
os.makedirs(abs_pred_path)
for k in range(len(predictions)):
pred = predictions[k]
np.save(os.path.join(path, 'sub%s.npy' % k), pred)
def plot_acc_auc_sens_spec(train_acc, train_auc, train_sens, train_spec, save_path):
t = range(1, len(train_acc) + 1)
plt.figure(1)
plt.subplot(221)
plt.plot(t, train_acc, label=save_path)
plt.ylim([0.0, 1.05])
plt.xlabel('Training Epochs')
plt.ylabel('Accuracy')
plt.subplot(222)
plt.plot(t, train_auc, label=save_path)
plt.ylim([0.0, 1.05])
plt.xlabel('Training Epochs')
plt.ylabel('AUC')
plt.subplot(223)
plt.plot(t, train_sens, label=save_path)
plt.ylim([0.0, 1.05])
plt.xlabel('Training Epochs')
plt.ylabel('Sensitivity')
plt.subplot(224)
plt.plot(t, train_spec, label=save_path)
plt.ylim([0.0, 1.05])
plt.xlabel('Training Epochs')
plt.ylabel('Specificity')
plt.legend()
plt.savefig(save_path, dpi=600)
"""
def plot_dice_coefficient(net, N, model_path_prefix, save_path):
test_data, test_label = image_util.ImageDataProvider(search_path="test/*.npy",
data_prefix="org",
label_prefix="scr",
channels=1,
n_class=3,
shuffle_data=False)(20)
path = './' + model_path_prefix + '(0)\model.cpkt'
prediction = [net.predict(path, data) for data in test_data]
for i in range(1, N):
path = './' + model_path_prefix + '(%s)\model.cpkt' % i
prediction += [net.predict(path, data) for data in test_data]
ground_truth = []
for label in test_label:
ground_truth.append(crop_to_shape(label, prediction[test_label.index(label)].shape))
# mask = np.tile(crop_to_shape(test_mask, pred.shape), [N, 1, 1, 1])[..., 1]
t = np.arange(0, 1.01, 0.01)
dc_mean = np.zeros_like(t)
for k in range(0, 101):
prediction_1 = [np.copy(pred) for pred in prediction]
if k == 0:
for pred_1 in prediction_1:
pred_1[pred_1 >= t[k]] = 1
else:
for pred_1 in prediction_1:
pred_1[pred_1 >= t[k]] = 1
pred_1[pred_1 < t[k]] = 0
dc_mean[k] = np.mean([2 * np.sum(pred_1 * ground_truth[prediction_1.index(pred_1)]) /
np.sum(pred_1 + ground_truth[prediction_1.index(pred_1)]) for pred_1 in prediction_1])
plt.figure(2)
plt.plot(t, dc_mean, label=save_path)
plt.ylim([0.0, 1.0])
plt.xlabel('Threshold')
plt.ylabel('Dice Coefficient')
plt.legend()
plt.savefig(save_path + '.png', dpi=600)
t_max = t[np.argmax(dc_mean)]
dc = np.zeros(len(prediction))
for k in range(len(prediction)):
pred_2 = np.copy(prediction[k])
gt_2 = np.copy(ground_truth[k])
if t_max == 0:
pred_2[pred_2 >= t_max] = 1
else:
pred_2[pred_2 >= t_max] = 1
pred_2[pred_2 < t_max] = 0
dc[k] = (2 * np.sum(pred_2 * gt_2)) / np.sum(pred_2 + gt_2)
np.save(save_path + '.npy', dc)
def plot_roc_curve(net, N, model_path_prefix, save_path):
test_data, test_label = image_util.ImageDataProvider(search_path="test/*.npy",
data_prefix="org",
label_prefix="scr",
channels=1,
shuffle_data=False)(20)
path = './' + model_path_prefix + '(0)\model.cpkt'
prediction = [net.predict(path, data) for data in test_data]
for i in range(1, N):
path = './' + model_path_prefix + '(%s)\model.cpkt' % i
prediction += [net.predict(path, data) for data in test_data]
ground_truth = []
for label in test_label:
ground_truth.append(crop_to_shape(label, prediction[test_label.index(label)].shape))
pred = np.array([])
for pred_1 in prediction:
pred = np.hstack((pred, np.reshape(pred_1, [-1])))
gt = np.array([])
for gt_1 in ground_truth:
gt = np.hstack((gt, np.reshape(gt_1, [-1])))
fpr, tpr, thresholds = roc_curve(np.reshape(gt, [-1]), np.reshape(pred, [-1]), pos_label=1)
roc_auc = auc(fpr, tpr)
np.save(save_path, roc_auc)
plt.figure(3)
plt.plot(fpr, tpr, label='AUC= %.3f' % roc_auc)
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.legend(loc='lower right')
plt.savefig(save_path + '.png', dpi=600)
def compare_stats(stat_1, stat_2):
from scipy import stats
n = len(stat_1)
T = np.sqrt(n) * np.mean(stat_1 - stat_2) / np.std(stat_1 - stat_2)
pval = stats.t.sf(np.abs(T), n - 1) * 2
t0 = stats.t.ppf(0.025, n - 1)
if T > -t0:
print('stat_1 is significantly greater than stat_2.')
elif T < t0:
print('stat_2 is significantly greater than stat_1.')
else:
print('no significant difference between two statistics.')
print('t-statistic= %.3f p-value= %.4f' % (T, pval))
"""