def load_train_data(count=2000, start=10000):
"""Loads training and/or evaluation data"""
print("Loading data...")
biffolder = '..\\..\\data\\bifs'
nobiffolder = '..\\..\\data\\nobifs'
# 4 dimensional data
data = np.zeros([2 * count, 17, 17, 17], dtype=np.float32)
# print(np.shape(data))
# First load the bif data.
for i in utility.my_range(start, start + count, 1):
currentfile = biffolder + '\\cropped' + str(i) + ".nii.gz"
data[i - start, :, :, :] = np.array(get_itk_array(currentfile))
print("Loaded bif data.")
# Now load the no bif data.
for i in utility.my_range(start, start + count, 1):
currentfile = nobiffolder + '\\cropped' + str(i) + ".nii.gz"
data[i + count - start, :, :, :] = np.array(get_itk_array(currentfile))
print("Loaded no bif data.")
print("All data loaded.")
return data
def load_test_data_folder(folder_name, number_files=0):
# Read info file.
with open(folder_name + "info.txt") as f:
content = f.readlines()
content = [x.strip() for x in content]
num_files = int(content[4])
print("Total Number of Files: ", num_files)
print("Loading testing data...")
if number_files != 0:
num_files = number_files
print("Loading only the first ", num_files, " of data.")
data = np.zeros([num_files, 64, 64, 64], dtype='float32')
for i in utility.my_range(0, num_files, 1):
data_file_name = folder_name + "cropped" + str(i) + ".nii.gz"
data[i] = np.asarray(get_itk_array(data_file_name), dtype='float32')
print("All data loaded.")
return data
def main(unused_args):
test_data = load_test_data_folder('../input/norm2/', )
test_data = np.reshape(test_data, [-1, 64, 64])
print("Input Data...")
print("Shape: ", np.shape(test_data))
print("Max: ", np.max(test_data))
print("Min: ", np.min(test_data))
print("Mean: ", np.mean(test_data))
# Create the Estimator
mnist_classifier = tf.estimator.Estimator(
model_fn=cnn_model_fn, model_dir="/home/joseph/Projects/vascularnetworks/models/giles2d_deep")
predict_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": test_data},
batch_size=1,
num_epochs=1,
shuffle=False)
# Predict data
predictions = mnist_classifier.predict(input_fn=predict_input_fn)
predictions = np.array(list(predictions))
print(np.shape(predictions))
print(predictions[0]['classes'])
print(np.shape(predictions[0]['classes']))
result = np.zeros([64, 64, 64, 64], dtype='uint8')
for i in utility.my_range(0, 64, 1):
for j in utility.my_range(0, 64, 1):
result[i, j, :, :] = predictions[i * 64 + j]['classes']
# print(result)
# print(np.shape(result))
for i in utility.my_range(0, 64, 1):
write_itk_imageArray(result[i], '../input/myseg2/cropped' + str(i) + '.nii.gz')
def load_test_data(filename, blocksize=8):
"""Loads testing data"""
print("Started loading.")
inputimage = np.array(get_itk_array(filename))
inputsize = np.shape(inputimage)
xsize = inputsize[0]
ysize = inputsize[1]
zsize = inputsize[2]
count = 128 * 128 * 128
data = np.zeros([count, 17, 17, 17], dtype=np.float32)
index = 0
for i in utility.my_range(blocksize, blocksize + 128, 1):
# print("Step", i)
for j in utility.my_range(blocksize, blocksize + 128, 1):
for k in utility.my_range(blocksize, blocksize + 128, 1):
data[index, :, :, :] = inputimage[(i -
blocksize):(i + blocksize +
1),
(j -
blocksize):(j + blocksize +
1),
(k -
blocksize):(k + blocksize +
1)]
index = index + 1
print("Loaded data.")
return data
def main(unused_args):
# Load training and eval data
# mnist = tf.contrib.learn.datasets.load_dataset("mnist")
# train_data = mnist.train.images # Returns np.array (55000, 784)
# train_labels = np.asarray(mnist.train.labels, dtype=np.int32) # (55000, 1)
test_data = load_test_data('..\\input\\test.mhd')
# print("Input Data...")
# print("Shape: ", np.shape(test_data))
# print("Max: ", np.max(test_data))
# print("Min: ", np.min(test_data))
# print("Mean: ", np.mean(test_data))
# Create the Estimator
mnist_classifier = tf.estimator.Estimator(
model_fn=cnn_model_fn, model_dir="/tmp/mnist_convnet_model")
predict_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": test_data},
batch_size=100,
num_epochs=1,
shuffle=False)
# Predict data
predictions = mnist_classifier.predict(input_fn=predict_input_fn)
predictions = np.array(list(predictions))
result = np.zeros(np.shape(predictions))
for index in utility.my_range(0, np.shape(predictions)[0], 1):
result[index] = predictions[index]['classes']
# print(result)
write_itk_imageArray(np.reshape(result, [128, 128, 128]), '..\\output\\detected.nii.gz')
ptindex = 0
Points = vtk.vtkPoints()
line = vtk.vtkCellArray()
radius = vtk.vtkIntArray()
radius.SetNumberOfComponents(1)
radius.SetName("radius")
# line.Allocate(8)
print("started")
data = dict()
edges = dict()
for i in utility.my_range(1, dims[0] - 1, 1):
print("Step: ", i)
for j in utility.my_range(1, dims[1] - 1, 1):
for k in utility.my_range(1, dims[2] - 1, 1):
# If pixel is 1
if inputimg[i, j, k] == 1:
# Get index at image
ind = i * dims[0] * dims[1] + j * dims[1] + k
# Check if already added to list. If not, add.
if ind not in data:
Points.InsertNextPoint(i, j, k)
content = [x.strip() for x in content]
stepsize = content[0]
xSize = content[1]
ySize = content[2]
zSize = content[3]
numfiles = content[4]
print("Step Size is: ", stepsize)
print("Total Image Size is: ", xSize, ySize, zSize)
print("Total Number of Files: ", numfiles)
# Segment every file.
index = 0
for i in utility.my_range(0, int(numfiles), 1):
print("Segmenting file: ", "cropped" + str(index) + ".nii.gz")
datafilename = workfolder + "/cropped" + str(index) + ".nii.gz"
writefilename = segfolder + "/cropped" + str(index) + ".nii.gz"
# Apply thresholding to it.
os.system("python " + executable + " " + datafilename + " " +
writefilename + " " + thresholdvalue)
# Update indices
index = index + 1
print("Done.")
xSize = inputSize[0]
ySize = inputSize[1]
zSize = inputSize[2]
file = open(workfolder + "info.txt", "w")
file.write(str(blocksize) + "\n")
file.write(str(xSize) + "\n")
file.write(str(ySize) + "\n")
file.write(str(zSize) + "\n")
print(blocksize, xSize, ySize, zSize)
# Index for bifurcations
index = 0
for i in utility.my_range(blocksize + stepsize, xSize - blocksize, 1):
print("Step", i)
for j in utility.my_range(blocksize + stepsize, ySize - blocksize, 1):
for k in utility.my_range(blocksize + stepsize, zSize - blocksize, 1):
# print("Step at: (", index, ")", i, j, k)
# If bifurcation exists.
if bifimg[i, j, k] != 0:
# Crop image and write it.
# Checks surrounding. If no bifurc. Add it.
surrounding = bifimg[(i - 2 * stepsize - 1):i,
(j - 2 * stepsize):j,
xSize = int(content[1])
ySize = int(content[2])
zSize = int(content[3])
numfiles = int(content[4])
startX = 0
startY = 0
startZ = 0
# Create a file with that size.
joinedfile = np.zeros([xSize, ySize, zSize], dtype="uint8")
# Join the small files into one big file.
index = 0
for i in utility.my_range(0, xSize, stepsize):
for j in utility.my_range(0, ySize, stepsize):
for k in utility.my_range(0, zSize, stepsize):
print("Step at: (", index, ")", startX, startY, startZ)
endX = startX + stepsize
endY = startY + stepsize
endZ = startZ + stepsize
currentfilename = splitfolder + "/cropped" + str(index) + ".nii.gz"
currentfile = np.array(get_itk_array(currentfilename),
dtype='uint8')
joinedfile[startX:endX, startY:endY, startZ:endZ] = currentfile
# Update indices
