def load_complete_data(data_path):
"""
Load the complete data containing both i-contour and o-contour.
:param data_path: Path to the folder with the DICOM data and labels.
:type data_path: str
:return: images, o_contours, i_contours
:rtype: np.ndarray, np.ndarray, np.ndarray
"""
def correspondig_i_contour_path(o_contour, i_contour_dir):
"""
Get i-contour path that corresponds to the given o-contour path.
:param o_contour: Path to the o-countour file.
:type o_contour: str
:param i_contour_dir: Path to the i-contour folder.
:type i_contour_dir: str
:return: Path to the i-contour file.
:rtype: str
"""
i_contour_basename = os.path.basename(o_contour).replace(
'ocontour', 'icontour')
return os.path.join(i_contour_dir, i_contour_basename)
images = []
o_contours = []
i_contours = []
ids_mapping = parse_csv_mapping_file(data_path)
for dicom, contour in ids_mapping.iteritems():
o_contour_files_path = os.path.join(data_path, COUNTOURS_FOLDER_NAME,
contour, O_CONTOURS_FOLDER_NAME)
i_contour_files_path = os.path.join(data_path, COUNTOURS_FOLDER_NAME,
contour, I_CONTOURS_FOLDER_NAME)
dicom_files_path = os.path.join(data_path, DICOMS_FOLDER_NAME, dicom)
for contour_file_path in glob.glob(
os.path.join(o_contour_files_path,
CONTOURS_EXTENSION_PATTERN)):
o_contour_list = parse_contour_file(contour_file_path)
i_contour_path = correspondig_i_contour_path(
contour_file_path, i_contour_files_path)
i_contour_list = parse_contour_file(i_contour_path)
dicom_file_name = contour_name_to_dicom_name(
os.path.basename(contour_file_path))
dicom_pixels = parse_dicom_file(
os.path.join(dicom_files_path, dicom_file_name))['pixel_data']
height, width = dicom_pixels.shape
images.append(dicom_pixels)
o_contours.append(poly_to_mask(o_contour_list, width, height))
i_contours.append(poly_to_mask(i_contour_list, width, height))
images = np.array(images)
o_contours = np.array(o_contours)
i_contours = np.array(i_contours)
assert images.shape[0] == o_contours.shape[0], \
'There must be exactly the same number of DICOMS and o-contours.'
assert images.shape[0] == i_contours.shape[0], \
'There must be exactly the same number of DICOMS and i-contours.'
return images, o_contours, i_contours
def test_contour_file_to_mask(self):
"""
Test parse contour file.
"""
file_path = os.path.join(LOCAL_PATH, 'test_input',
'IM-0001-0048-icontour-manual.txt')
result = poly_to_mask(parse_contour_file(file_path), 256, 256)
expected = np.load(
os.path.join(LOCAL_PATH, 'test_output',
'IM-0001-0048-icontour-manual.npy'))
np.testing.assert_equal(result, expected)
def visualize_contour_masks_diff(o_contour_path,
i_contour_path,
width=256,
height=256):
"""
Visualize diff of i_contour and o_contour.
:param o_contour_path: Path to the contour file.
:type o_contour_path: str
:param i_contour_path: Path to the contour file.
:type i_contour_path: str
:param width: Width of the contour.
:type width: int
:param height: Height of the contour.
:type height: int
"""
i_image = poly_to_mask(parse_contour_file(i_contour_path), width, height)
o_image = poly_to_mask(parse_contour_file(o_contour_path), width, height)
image = o_image ^ i_image
plt.imshow(image)
plt.show()
def visualize_contour_mask(contour_path, width=256, height=256):
"""
Visualize the contour converted to mask.
:param contour_path: Path to the contour file.
:type contour_path: str
:param width: Width of the contour.
:type width: int
:param height: Height of the contour.
:type height: int
"""
image = poly_to_mask(parse_contour_file(contour_path), width, height)
plt.imshow(image)
plt.show()
def visualize_dicom_with_contour(dicom_path, contour_path):
"""
Visualize the DICOM and contour mask.
:param dicom_path: Path to the DICOM image.
:type dicom_path: str
:param contour_path: Path to the corresponding contour file.
:type contour_path: str
"""
dicom_image = parse_dicom_file(dicom_path)['pixel_data']
height, width = dicom_image.shape
contour_image = poly_to_mask(parse_contour_file(contour_path), width,
height)
fig = plt.figure(figsize=(8, 8))
fig.add_subplot(1, 2, 1)
plt.imshow(dicom_image)
fig.add_subplot(1, 2, 2)
plt.imshow(contour_image)
plt.show()
def load_data_as_dataset(data_path):
"""
Load the data.
NOTE: assumes that all the data can fit to the RAM memory.
NOTE: for the purpose of the challenge only the i-contour data is used
NOTE: only DICOM files having the corresponding i-contour file are loaded
:param data_path: Path to the folder with the DICOM data and labels.
:type data_path: str
:return: images, labels
:rtype: np.ndarray.Dataset, np.ndarray.Dataset
"""
images = []
labels = []
ids_mapping = parse_csv_mapping_file(data_path)
for dicom, contour in ids_mapping.iteritems():
contour_files_path = os.path.join(data_path, COUNTOURS_FOLDER_NAME,
contour, I_CONTOURS_FOLDER_NAME)
dicom_files_path = os.path.join(data_path, DICOMS_FOLDER_NAME, dicom)
for contour_file_path in glob.glob(
os.path.join(contour_files_path, CONTOURS_EXTENSION_PATTERN)):
contour_list = parse_contour_file(contour_file_path)
dicom_file_name = contour_name_to_dicom_name(
os.path.basename(contour_file_path))
dicom_pixels = parse_dicom_file(
os.path.join(dicom_files_path, dicom_file_name))['pixel_data']
height, width = dicom_pixels.shape
images.append(dicom_pixels)
labels.append(poly_to_mask(contour_list, width, height))
images = np.array(images)
images = np.expand_dims(images, -1)
labels = np.array(labels)
assert images.shape[0] == labels.shape[0], \
'There must be exactly the same number of DICOMS and contours.'
return images, labels