def Segment():
patient = readindex()
patient = np.array(patient)
for i in range(25):
data = DataProcess(i + 1)
print('data shape:' + str(data.shape))
OUTMHA = np.zeros([data.shape[0], data.shape[1], data.shape[2]])
for x in range(data.shape[0]):
for y in range(data.shape[1] - 64):
x1, x2 = Getbatch(data, x, y + 32)
print('x1,x2:' + str(x1.shape) + str(x2.shape))
X1 = tf.placeholder(dtype=tf.float32,
shape=[None, 65, 65, 4],
name='big_patch')
X2 = tf.placeholder(dtype=tf.float32,
shape=[None, 33, 33, 4],
name='small_patch')
OUT = Network(X1, X2, data.shape[2] - 64)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
saver = tf.train.Saver()
saver.restore(
sess=sess,
save_path=
'E:/zhangjinjing/brain2D/brats2013/brain_session_2013_kongdong.ckpt'
)
OUT = sess.run(OUT, feed_dict={X1: x1, X2: x2})
OUTMHA[x, y, 32:data.shape[2] - 32] = np.argmax(OUT, 1)
print(OUTMHA.shape)
util.save_nuarray_as_mha(
'E://zhangjinjing/brain2D/2013test/VSD.Seg_HG_001.' +
str(patient[i]) + '.mha', OUTMHA)
def segment_unparalel(patient_parent_file, description, model, extra_file):
before = int(round(time.time() * 1000))
print("unparalel segmentation")
print(patient_parent_file)
p = Patient()
p.set_parent_file(patient_parent_file)
p.set_window_size(model_setup.window_size)
p.start_iteration()
print(p.file_FLAIR)
image_name = re.findall(r'\d+', p.file_FLAIR)
image_name = image_name[len(image_name) - 1]
path_ttt, file_nnn = os.path.split(patient_parent_file)
image_name = "VSD." + description + "_(" + file_nnn + ")." + image_name + ".mha"
print("result name : %s" % image_name)
file_result_name = os.path.join(patient_parent_file, image_name)
extra_file = os.path.join(extra_file, image_name)
if (os.path.isfile(file_result_name)):
return
segmentation = numpy.zeros(shape=p.label.shape)
z = 0
utility.save_nuarray_as_mha(file_result_name, segmentation)
utility.save_nuarray_as_mha(extra_file, segmentation)
while z < p.limit_z:
for y in range(0, p.limit_y):
if (y % 10 == 0):
print("%s %s" % (z, y))
for x in range(0, p.limit_x):
if (p.is_back_ground(z, y, x)):
segmentation[z][y][x] = 0
else:
t, g = p.get_batch_at(z, y, x)
features = []
features.append(t)
features = numpy.asarray(features)
features = features.astype('float32')
r = model.predict_classes(features,
batch_size=1,
verbose=False)
segmentation[z][y][x] = r
z += 1
utility.save_nuarray_as_mha(file_result_name, segmentation)
utility.save_nuarray_as_mha(extra_file, segmentation)
p.stop_iteration()
print("segmentation done")
totoal_time = int(round(time.time() * 1000)) - before
print("total time : %s minutes" % (totoal_time / 1000.0 / 60))
def Segment():
X1 = tf.placeholder(dtype=tf.float32,
shape=[None, 65, 65, 4],
name='big_patch')
X2 = tf.placeholder(dtype=tf.float32,
shape=[None, 33, 33, 4],
name='small_patch')
OUT = Network(X1, X2)
patient = readindex()
patient = np.array(patient)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
saver = tf.train.Saver()
saver.restore(
sess, 'E:/zhangjinjing/brain2D/brats2013/brain_session_2013_8.ckpt')
for i in range(25):
data = DataProcess(i)
print('data shape:' + str(data.shape))
OUTMHA = np.zeros([data.shape[0], data.shape[1], data.shape[2]])
print('MHA:', OUTMHA.shape)
index = Getcoord(data)
print('index:', index.shape)
n = 0
while (n < index.shape[0] - 175):
subindex = np.array(index[n:n + 175])
print('sunindex:', subindex.shape)
print('n：', n)
x1, x2 = Getbatch(data, subindex)
n = n + 175
OUTLINE = sess.run(OUT, feed_dict={X1: x1, X2: x2})
OUTLINE = np.array(OUTLINE)
#print('OUTLINE:',OUTLINE.shape)
for s in range(175):
OUTMHA[subindex[s, 0], subindex[s, 1],
subindex[s, 2]] = np.argmax(OUTLINE, 1)[s]
print('subindex:', subindex[s, 0], subindex[s, 1], subindex[s,
2])
print('max:', np.argmax(OUTLINE, 1)[s])
print('finish!')
util.save_nuarray_as_mha(
'E://zhangjinjing/brain2D/2013test/VSD.Seg_HG_001.' +
str(patient[i]) + '.mha', OUTMHA)
def preprocess(self):
self.set_parent_file(self.file_parent)
if (self._find_file("back_ground") == None):
back_ground = self.BFS()
utility.save_nuarray_as_mha(
os.path.join(self.file_parent, "back_ground.mha"), back_ground)
self.set_parent_file(self.file_parent)
self.back_ground = utility.read_mha_image_as_nuarray(
self._find_file("back_ground\.mha"))
balanced_data_file = os.path.join(self.file_parent,
"balanced_data.pickle")
if (not os.path.isfile(balanced_data_file)):
self.balance_data()
ff = open(balanced_data_file, "rb")
self.balanced_data_indices = pickle.load(ff)
ff.close()
def preprocess(
self
): # creates a background.mha file ,which displays black brain part
self.set_parent_file(self.file_parent)
if (self._find_file("back_ground") == None):
back_ground = self.BFS()
utility.save_nuarray_as_mha(
os.path.join(self.file_parent, "back_ground.mha"),
back_ground) #getting image from array
self.set_parent_file(self.file_parent)
self.back_ground = utility.read_mha_image_as_nuarray(
self._find_file("back_ground\.mha")) # getting array from img
balanced_data_file = os.path.join(
self.file_parent, "balanced_data.pickle") #file pickling
if (not os.path.isfile(balanced_data_file)):
self.balance_data()
ff = open(balanced_data_file, "rb")
self.balanced_data_indices = pickle.load(ff)
ff.close()
import numpy as np
import utility as util
modi210=util.read_mha_image_as_nuarray('test210.mha')
temp=np.zeros([155,240,240])
temp[:,32:207,32:207]=modi210
util.save_nuarray_as_mha('test210_2.mha', temp)
import scipy.io
import numpy as np
import utility as util
mat = scipy.io.loadmat('brain_1125.mat')
print(mat)
files = mat['img']
# file=mat['V']
y = np.zeros([files.shape[2], files.shape[1], files.shape[0]])
for i in range(files.shape[2]):
x = files[:, :, i]
for j in range(files.shape[1]):
y[i, j] = x[:, j]
print(y.shape)
util.save_nuarray_as_mha('ktest.mha', y)
import numpy as np
import os
import utility as util
seg="E://zhangjinjing/brain2D/test210.mha"
label="E://zhangjinjing/brain2D/BrainTrain/210/0005/0005.mha"
def compare(path1,path2):
data1=util.read_mha_image_as_nuarray(path1)
data2=util.read_mha_image_as_nuarray(path2)
for i in range(data1.shape[0]):
for j in range(data1.shape[1]):
for k in range (data1.shape[2]):
if data1[i,j,k]==data2[i,j,k]:
data1[i,j,k]=0
return data1
if __name__=="__main__":
data=compare(seg,label)
util.save_nuarray_as_mha("E://zhangjinjing/brain2D/error.mha",data)
# print(OUTLINE.shape)
# util.save_nuarray_as_mha('VSD.Seg_LG_011.'+str(dict[zip_num+85])+'.mha', OUTMHA)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
saver = tf.train.Saver()
saver.restore(
sess,
'E:/zhangjinjing/brain2D/brain_session_HLGG/brain_session_resize_8.ckpt')
print('read session ok')
iter = 0
OUTMHA = np.zeros([ALLtest.shape[0], ALLtest.shape[1], ALLtest.shape[2]])
for pos_x in range(ALLtest.shape[0]):
for pos_y in range(ALLtest.shape[1] - patch_size):
iter += 1
print('\n', iter)
print('finish in: ', iter * 700. / 189875., '%')
#Input_batch1,Input_batch2,Label_batch=getBatch2D(pos_x,pos_y)
Input_batch1, Input_batch2 = getBatch2D(pos_x, pos_y)
#OUTLINE=sess.run([OUT],feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch})
OUTLINE = sess.run([OUT],
feed_dict={
Xp1: Input_batch1,
Xp2: Input_batch2
})
OUTLINE = np.array(OUTLINE)
OUTMHA[pos_x, pos_y + halfpz,
halfpz:175 + halfpz] = np.argmax(OUTLINE[0, :, :], 1)
util.save_nuarray_as_mha("test/8/VSD.Seg_HG_001.40767.mha", OUTMHA)
print(OUTLINE.shape)
loss=tf.reduce_mean(-tf.reduce_sum(Yp*tf.log(OUT),reduction_indices=[1]))
correct_prediction = tf.equal(tf.argmax(OUT,1), tf.argmax(Yp,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess=tf.Session(config=config)
saver = tf.train.Saver()
saver.restore(sess,'E:/zhangjinjing/brain2D/brain_session_HLGG/brain_session_resize.ckpt')
print('read sessionOK')
iter=0
OUTMHA=np.zeros([ALLtest.shape[0],ALLtest.shape[1],ALLtest.shape[2]])
for pos_x in range(ALLtest.shape[0]):
for pos_y in range(ALLtest.shape[1]-patch_size):
iter+=1
print('\n',iter)
print('finish in: ',iter*700./189875.,'%')
Input_batch1,Input_batch2,Label_batch=getBatch2D(pos_x,pos_y)
#Input_batch1, Input_batch2 =getBatch2D(pos_x,pos_y)
OUTLINE=sess.run([OUT],feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch})
#OUTLINE=sess.run([OUT],feed_dict={Xp1:Input_batch1,Xp2:Input_batch2})
OUTLINE=np.array(OUTLINE)
# print(tf.Session().run(tf.arg_max(OUTLINE,1),feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch}))
# print(tf.Session().run(tf.arg_max(Label_batch,1),feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch}))
OUTMHA[pos_x,pos_y+halfpz,halfpz:175+halfpz]=np.argmax(OUTLINE[0,:,:],1)
print(OUTLINE.shape)
util.save_nuarray_as_mha('test40461_resize.mha', OUTMHA)
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
saver = tf.train.Saver()
saver.restore(
sess, 'E:/zhangjinjing/brain2D/brain_session_HLGG/brain_session.ckpt')
print('read sessionOK')
iter = 0
OUTMHA = np.zeros([ALLtest.shape[0], ALLtest.shape[1], ALLtest.shape[2]])
for pos_x in range(ALLtest.shape[0]):
for pos_y in range(ALLtest.shape[1] - patch_size):
iter += 1
print('\n', iter)
print('finish in: ', iter * 700. / 189875., '%')
#Input_batch1,Input_batch2,Label_batch=getBatch 2D(pos_x,pos_y)
Input_batch1, Input_batch2 = getBatch2D(pos_x, pos_y)
#OUTLINE=sess.run([OUT],feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch})
OUTLINE = sess.run([OUT],
feed_dict={
Xp1: Input_batch1,
Xp2: Input_batch2
})
OUTLINE = np.array(OUTLINE)
# print(tf.Session().run(tf.arg_max(OUTLINE,1),feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch}))
# print(tf.Session().run(tf.arg_max(Label_batch,1),feed_dict={Xp1:Input_batch1,Xp2:Input_batch2,Yp:Label_batch}))
OUTMHA[pos_x, pos_y + halfpz,
halfpz:175 + halfpz] = np.argmax(OUTLINE[0, :, :], 1)
print(OUTLINE.shape)
util.save_nuarray_as_mha("VSD.Seg_HG_001." + index[zip_num] + ".mha",
OUTMHA)
convout = tf.nn.conv2d(output,
weight_variable([21, 21, 224, 5]),
strides=[1, 1, 1, 1],
padding='VALID') + bias_variable([5])
convout = tf.reshape(convout, [175, 5])
OUT = tf.nn.softmax(convout)
OUTMHA = np.zeros([ALLTest.shape[0], ALLTest.shape[1], ALLTest.shape[2]])
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
saver = tf.train.Saver()
saver.restore(
sess,
'E:/zhangjinjing/brain2D/brain_session_HLGG/brain_session_resize2.ckpt')
print('read sessionOK')
iter = 1
for pos_x in range(ALLTest.shape[0]):
for pos_y in range(ALLTest.shape[1] - patch_size):
input_batch1, input_batch2 = get_2Dbatch(pos_x, pos_y)
out = sess.run([OUT], feed_dict={XP1: input_batch1, XP2: input_batch2})
out = np.array(out)
Endpatch = half_patch + 175
OUTMHA[pos_x, pos_y + half_patch,
half_patch:Endpatch] = np.argmax(out[0, :, :], 1)
print("inter=", iter / 26250)
iter = iter + 1
print(out.shape)
util.save_nuarray_as_mha('VSD.Seg_HG_001.41163.mha', OUTMHA)