def prepocess():
images = get_files(PRE_IMAGE_PATH)
labels = get_files(PRE_LABEL_PATH)
files = get_files(PRE_IMAGE_PATH, prefix=False)
def normlize_data(image):
vaild_area = (image >= 0)
invaild_area = (image < 0)
image[vaild_area] = np.log(image[vaild_area] + 1)
i_mean = image[vaild_area].mean()
i_std = image[vaild_area].std()
low = i_mean - 3.5 * i_std
high = i_mean + 3.5 * i_std
image = (image - low) / (high - low)
image[invaild_area] = -1
return image
for f_i, f_l, f in tqdm(zip(images, labels, files)):
im = load_image(f_i).get_data()
label = load_image(f_l).get_data()
im = normlize_data(im)
label[label == 2] = 1
np.save(IMAGE_PATH + f[:-7] + '.npy', im)
np.save(LABEL_PATH + f[:-7] + '.npy', label)
def predict_test_data():
model_files = get_files(MODEL_PATH, prefix=False)
models = []
for file_name in model_files:
m_setting = file_name.split('_')
model_setting = {}
model_setting['modelname'] = m_setting[0]
model_setting[
'axis'] = None if m_setting[2] == 'None' else m_setting[2]
model_setting['loss'] = diceLoss
if m_setting[3] == 'ostu':
model_setting['postPocess'] = ostu
elif m_setting[3] == 'around':
model_setting['postPocess'] = threshold_filter
else:
raise NameError("postPocess error")
model_setting['path'] = MODEL_PATH + file_name
models.append(model_setting)
test_data = get_files(TEST_DATA, prefix=False)
seg_recovery(test_data, models)
def EDA_warp():
files = get_files(PRE_IMAGE_PATH, prefix=False)
labels = []
images = []
for i in range(len(files)):
im = load_image(PRE_IMAGE_PATH + files[i]).get_data()
if im.shape == (256, 256, 180, 1):
labels.append(PRE_LABEL_PATH + files[i])
images.append(PRE_IMAGE_PATH + files[i])
EDA(labels, images, '180')
labels = []
images = []
for i in range(len(files)):
im = load_image(PRE_IMAGE_PATH + files[i]).get_data()
if im.shape == (256, 256, 166, 1):
labels.append(PRE_LABEL_PATH + files[i])
images.append(PRE_IMAGE_PATH + files[i])
EDA(labels, images, '256_166')
labels = []
images = []
for i in range(len(files)):
im = load_image(PRE_IMAGE_PATH + files[i]).get_data()
if im.shape == (192, 192, 160, 1):
labels.append(PRE_LABEL_PATH + files[i])
images.append(PRE_IMAGE_PATH + files[i])
EDA(labels, images, '192')
def prepocess():
images = get_files(PRE_IMAGE_PATH)
labels = get_files(PRE_LABEL_PATH)
files = get_files(PRE_IMAGE_PATH, prefix=False)
for f_i, f_l, f in tqdm(zip(images, labels, files)):
im = load_image(f_i).get_data()
label = load_image(f_l).get_data()
label[label == 2] = 1
# im = cv2.resize(im,(192,192,160,1),interpolation=cv2.INTER_LINEAR)
# print(im.shape)
# im = normlize_data(im)
np.save(IMAGE_PATH + f[:-7] + '.npy', im)
np.save(LABEL_PATH + f[:-7] + '.npy', label)
def test():
files = get_files(PRE_LABEL_PATH)
h1 = []
h2 = []
for f in files:
label = load_image(f).get_data()
h1.append(np.sum(label==1))
h2.append(np.sum(label==2))
print(sorted(h1))
print(sorted(h2))
def get_shape():
images = get_files(PRE_IMAGE_PATH, prefix=False)
im_shapes = {}
for f in images:
im = load_image(PRE_IMAGE_PATH + f)
print(im.__array__)
im = im.get_data()
im_shapes[f] = str(im.shape[2])
with open(INFO + 'shape.json', 'w') as f:
f.write(json.dumps(im_shapes, indent=4, separators=(',', ': ')))
def main():
from sklearn.cross_validation import KFold
files = get_files(LABEL_PATH, prefix=False)
kf = KFold(len(files), n_folds=K, shuffle=True)
for i, (train_index, valid_index) in enumerate(kf):
print('第{}次训练...'.format(i))
trainset = files[train_index]
validset = files[valid_index]
model = segment_model(valfiles=validset)
train_model(model, trainset, batchsize=3)
break
def main(loss,
modelname='Unet',
axis=None,
metric=dice_metric,
postPocess=None,
postPocess_str=None):
from sklearn.cross_validation import KFold
files = get_files(LABEL_PATH, prefix=False)
K = 10
kf = KFold(len(files), n_folds=K, shuffle=True)
for i, (train_index, valid_index) in enumerate(kf):
print('第{}次训练...'.format(i))
trainset = files[train_index]
validset = files[valid_index]
model = segment_model(valfiles=validset,
modelname=modelname,
axis=axis,
metric=metric,
loss=loss,
postPocess=postPocess)
if axis == None:
model_id = modelname + '_' + str(
i) + '_None_' + postPocess_str + '_'
else:
model_id = modelname + '_' + str(
i) * 2 + '_' + axis + '_' + postPocess_str + '_'
best_score = train_model(model,
trainset,
batchsize=8,
model_name=model_id,
axis=axis)
print(str(i), ': ', best_score)
def run(self):
done = 0
fail = 0
ignored = 0
not_allowed = 0
total_on_folders = 0
for folder in self.main.folders:
files = get_files(folder)
if self.main.code == 3:
files = self.main.sort_file_list(folder, files,
self.main.arg['i3'],
self.main.arg['i4'])
total_on_a_folder = 0
for a_file in files:
filename, ext = get_filename_extension(a_file)
status = self.main.check_for_allow_and_ignore(
a_file, filename, ext)
if status == 1:
ignored += 1
self.outSignal.emit(
f"<b>[+] - Ignored :</b> {join(folder, a_file)}")
elif status == 2:
not_allowed += 1
self.outSignal.emit(
f"<b>[+] - Not Allowed :</b> {join(folder, a_file)}")
else:
total_on_a_folder += 1
total_on_folders += 1
try:
new_name = self.main.get_new_name(
a_file, filename, ext, folder, total_on_a_folder,
total_on_folders)
except Exception as e:
print(e)
self.outSignal.emit(
f"<b>[+] - Error :</b> {join(folder, a_file)} <b>>>></b> Error on making new name"
)
self.outSignal.emit(f"<b>[+] - Stopping Renaming</b>")
self.main.do_log(False, join(folder, a_file), '')
break
if rename_file(folder, a_file, new_name):
done += 1
self.outSignal.emit(
f"<b>[+] - Done :</b> {join(folder, a_file)} <b>>></b> {new_name}"
)
self.main.do_log(True, join(folder, a_file), new_name)
else:
fail += 1
self.main.do_log(False, join(folder, a_file), new_name)
self.outSignal.emit(
f"<b>[+] - Failed :</b> {join(folder, a_file)} <b>>></b> {new_name}"
)
self.outSignal.emit("")
self.outSignal.emit(f"<b>[+] - Renaming Finished</b>")
self.outSignal.emit("")
total = done + fail + ignored + not_allowed
self.outSignal.emit(
f"<table> <tr> <td><b>[+] - Total</b></td> <td>:</td> <td>{total}</td> </tr> <tr> <td><b>[+] - Renamed</b></td> <td>:</td> <td>{done}</td> </tr> <tr> <td><b>[+] - Failed</b></td> <td>:</td> <td>{fail}</td> </tr> <tr> <td><b>[+] - Not Allowed</b></td> <td>:</td> <td>{not_allowed}</td> </tr> <tr> <td><b>[+] - Ignored</b></td> <td>:</td> <td>{ignored}</td> </tr> </table>"
)
self.outSignal.emit("")
if self.main.log_file is not None and self.main.make_log:
self.main.log_file.close()
if self.main.make_log_error:
self.outSignal.emit(f"[+] - <b>Log Failed</b>")
else:
self.outSignal.emit(
f"[+] - Log path : <b>{str(Path('logs.txt').absolute())}</b>"
)
self.outSignal.emit(f"[+] - Press Finish to quit")
def get_images_info():
def get_info(image, label):
def fune_range(info, target, value):
for i, t in enumerate(['x', 'y', 'z', 'voxel']):
if value[i] < info[target + '_' + t][0]:
info[target + '_' + t][0] = value[i]
if value[i] > info[target + '_' + t][1]:
info[target + '_' + t][1] = value[i]
return info
info = {
'brain_x': [256, 0],
'brain_y': [256, 0],
'brain_z': [256, 0],
'brain_voxel': [10000, -10000],
'h1_x': [256, 0],
'h1_y': [256, 0],
'h1_z': [256, 0],
'h1_voxel': [10000, -10000],
'h2_x': [256, 0],
'h2_y': [256, 0],
'h2_z': [256, 0],
'h2_voxel': [10000, -10000],
}
info['shape'] = image.shape
shape = image.shape
for i in range(shape[0]):
for j in range(shape[1]):
for k in range(shape[2]):
value = (i, j, k, int(image[i, j, k, 0]))
if label[i, j, k, 0] == 1:
info = fune_range(info, 'h1', value)
elif label[i, j, k, 0] == 2:
info = fune_range(info, 'h2', value)
elif image[i, j, k, 0] != -1:
info = fune_range(info, 'brain', value)
return info
images_info = {}
files = get_files(PRE_IMAGE_PATH, prefix=False)
image_files = [PRE_IMAGE_PATH + f for f in files]
label_files = [PRE_LABEL_PATH + f for f in files]
for l_f, im_f, name in tqdm(zip(label_files, image_files, files)):
# label, image = load_image(l_f).get_data(), load_image(im_f).get_data()
label, image = sitk.ReadImage(l_f), sitk.ReadImage(im_f)
label, image = sitk.GetArrayFromImage(label), sitk.GetArrayFromImage(
image)
label, image = label.transpose((1, 2, 0)), image.transpose((1, 2, 0))
label, image = label[:, :, :, np.newaxis], image[:, :, :, np.newaxis]
images_info[name] = get_info(image, label)
variables = [
'brain_x',
'brain_y',
'brain_z',
'brain_voxel',
'h1_x',
'h1_y',
'h1_z',
'h1_voxel',
'h2_x',
'h2_y',
'h2_z',
'h2_voxel',
]
for file in images_info.keys():
for vari in variables:
images_info[file][vari+'_range'] = \
images_info[file][vari][1] - images_info[file][vari][0]
for file in images_info.keys():
images_info[file]['h_z_range'] = \
images_info[file]['h2_z'][1] - images_info[file]['h1_z'][0]
for file in images_info.keys():
for axis in ['x', 'y', 'z']:
for h in ['h1_', 'h2_']:
images_info[file]['effec_' + h + axis] = [0, 0]
images_info[file]['effec_' + h + axis][0] = images_info[file][
h + axis][0] - images_info[file]['brain_' + axis][0]
images_info[file]['effec_' + h + axis][1] = images_info[file][
h + axis][1] - images_info[file]['brain_' + axis][0]
with open(INFO + 'image_info.json', 'w') as f:
f.write(json.dumps(images_info, indent=4, separators=(',', ': ')))