def test_rectangle():
fixture = get_fixture("subpixel_rectangle.roi")
with fixture.open("rb") as f:
rect = ijroi.read_roi(f)
assert (rect == np.array([[5, 4], [5, 8], [10, 8], [10, 4]])).all()
assert rect.dtype == np.float32
fixture = get_fixture("integer_rectangle.roi")
with fixture.open("rb") as f:
rect = ijroi.read_roi(f)
assert (rect == np.array([[5, 4], [5, 8], [10, 8], [10, 4]])).all()
assert rect.dtype == np.int16
def test_rectangle():
fixture = get_fixture("subpixel_rectangle.roi")
with fixture.open("rb") as f:
rect = ijroi.read_roi(f)
assert (rect == np.array([[5, 4], [5, 8], [10, 8], [10, 4]])).all()
assert rect.dtype == np.float32
fixture = get_fixture("integer_rectangle.roi")
with fixture.open("rb") as f:
rect = ijroi.read_roi(f)
assert (rect == np.array([[5, 4], [5, 8], [10, 8], [10, 4]])).all()
assert rect.dtype == np.int16
def test_polygon():
fixture = get_fixture("polygon_circle.roi")
with fixture.open("rb") as f:
circle = ijroi.read_roi(f)
assert len(circle) == 100
assert abs(circle[:, 1].mean()-10) < 0.01
assert abs(circle[:, 0].mean()-15) < 0.01
fixture = get_fixture("polygon_integer.roi")
with fixture.open("rb") as f:
polyint = ijroi.read_roi(f)
assert len(polyint) == 3
assert all(polyint[2, :] == [1, 10])
assert polyint.dtype == np.int16
def test_polygon():
fixture = get_fixture("polygon_circle.roi")
with fixture.open("rb") as f:
circle = ijroi.read_roi(f)
assert len(circle) == 100
assert abs(circle[:, 1].mean()-10) < 0.01
assert abs(circle[:, 0].mean()-15) < 0.01
fixture = get_fixture("polygon_integer.roi")
with fixture.open("rb") as f:
polyint = ijroi.read_roi(f)
assert len(polyint) == 3
assert all(polyint[2, :] == [1, 10])
assert polyint.dtype == np.int16
def test_freehand_circle():
fixture = get_fixture("freehand_circle.roi")
with fixture.open("rb") as f:
circle = ijroi.read_roi(f)
assert len(circle) == 100
assert abs(circle[:, 1].mean()-10) < 0.01
assert abs(circle[:, 0].mean()-15) < 0.01
def test_freehand_circle():
fixture = get_fixture("freehand_circle.roi")
with fixture.open("rb") as f:
circle = ijroi.read_roi(f)
assert len(circle) == 100
assert abs(circle[:, 1].mean()-10) < 0.01
assert abs(circle[:, 0].mean()-15) < 0.01
def test_integer_freehand():
fixture = get_fixture("freehand_integer.roi")
with fixture.open("rb") as f:
freehand = ijroi.read_roi(f)
assert len(freehand) == 3
assert all(freehand[2, :] == [1, 10])
assert freehand.dtype == np.int16
def test_float_point():
fixture = get_fixture("float_point.roi")
with fixture.open("rb") as f:
point = ijroi.read_roi(f)
assert point.ndim == 2
assert abs(point[0,0] - 567.8) < 0.01
assert abs(point[0,1] - 123.4) < 0.01
def test_point():
fixture = get_fixture("int_point.roi")
with fixture.open("rb") as f:
point = ijroi.read_roi(f)
assert point.ndim == 2
assert point[0,0] == 256
assert point[0,1] == 128
def test_point():
fixture = get_fixture("int_point.roi")
with fixture.open("rb") as f:
point = ijroi.read_roi(f)
assert point.ndim == 2
assert point[0, 0] == 256
assert point[0, 1] == 128
def test_float_point():
fixture = get_fixture("float_point.roi")
with fixture.open("rb") as f:
point = ijroi.read_roi(f)
assert point.ndim == 2
assert abs(point[0, 0] - 567.8) < 0.01
assert abs(point[0, 1] - 123.4) < 0.01
def test_integer_freehand():
fixture = get_fixture("freehand_integer.roi")
with fixture.open("rb") as f:
freehand = ijroi.read_roi(f)
assert len(freehand) == 3
assert all(freehand[2, :] == [1, 10])
assert freehand.dtype == np.int16
def test_oval_points_center_diam():
fixture = get_fixture("16x16_centered_spanning_oval.roi")
with fixture.open("rb") as f:
points = ijroi.read_roi(f)
center, diam = ijroi.oval_points_center_diam(points)
assert np.array_equal(center, np.array((8, 8)))
assert diam == 16
def parse_ij_roi(self, path, pixel_size):
with open(path, "rb") as f:
r = ijroi.read_roi(f)
# extract top left and bottom right of roi
# reorganise to make a list of x0, y0, x1, y1, id
roi_df = pd.DataFrame([[r[0,1].astype('float'),r[0,0].astype('float'),\
r[2,1].astype('float'),r[2,0].astype('float')]]) * pixel_size
roi_df.columns = [
'x0 [pixels]', 'y0 [pixels]', 'x1 [pixels]', 'y1 [pixels]'
]
return roi_df
def generate_roi_file():
roi_files = glob.glob(os.path.join(ROI_DIR, '**/*.roi', recursive=True))
with open(ROI_BBOX_FILE, 'w') as out:
for roi_file in roi_files:
with open(roi_file, "rb") as f:
roi = ijroi.read_roi(f)
out.write(
roi_file.replace("roi/", "").replace(".roi", ".jpg") +
", ")
out.write(", ".join(map(str, roi[0][::-1])) + ", ")
out.write(", ".join(map(str, roi[2][::-1])) + ", ")
out.write("cervix\n")
def import_ij_rois(basename, roipath):
rois = []
for filename in os.listdir(roipath):
if ('.roi' in filename) and (basename in filename):
fpath = os.path.join(roipath, filename)
print('roi path: ', fpath)
with open(fpath, "rb") as f:
r = ijroi.read_roi(f)
# extract top left and bottom right of roi
# reorganise to make a list of x0, y0, x1, y1
roi = [r[0,1].astype('float'),r[0,0].astype('float'),\
r[2,1].astype('float'),r[2,0].astype('float')]
rois.append(roi)
localisations
return rois
def _convert_from_roi(fname):
"""Convert a roi file to a numpy array [x, y, h, w].
Parameters
----------
fname : string
If ends with `.roi`, we assume a full path is given
"""
with open(fname, 'rb') as f:
roi = ijroi.read_roi(f)
top, left = roi[0]
bottom, right = roi[2]
height, width = bottom - top, right - left
return np.array([top, left, height, width])
def get_polygons_ijroi(self, ij_rois_fp):
"""Short summary.
Parameters
----------
ij_rois_fp : str
Filepath to an ImageJ ROI file
Returns
-------
list
Python list of polygon verteces as numpy arrays
"""
fn, fe = os.path.splitext(ij_rois_fp)
print(fe)
if fe == '.zip':
rois = ijroi.read_roi_zip(ij_rois_fp)
if fe == '.roi':
rois = ijroi.read_roi(open(ij_rois_fp, "rb"))
polyallcoords = [poly[1] for poly in rois]
self.polygons = polyallcoords
return im.size
if __name__ == '__main__':
import sys
from PIL import Image
import ijroi
TARGET_SIZE = (299, 299)
if len(sys.argv) != 3:
print(
"Take a list of .roi files that were created over the original (non-resized non-cropped) dataset and fix them to fit our 299x299 dataset format."
)
print("Example usage:")
print(
" $ ls path/to/large/.rois | python3 roi.py path/to/folder/with/images target/path"
)
sys.exit()
rois_to_fix = [roi.strip() for roi in sys.stdin]
img_dir, output_dir = sys.argv[1:]
for roi in rois_to_fix:
base_name = roi[roi.rfind('/') + 1:-len('.roi')]
with open(roi, "rb") as f:
bb = ijroi.read_roi(f)
original_dims = get_img_dims(img_dir + '/' + base_name + '.jpg')
bb_resized = downsize_bb(bb, original_dims, TARGET_SIZE)
with open(output_dir + '/' + base_name + '.roi', "wb+") as f:
write_bb(f, bb_resized)
def load_ij_roi(filename):
import ijroi
with open(filename, "rb") as fh:
return ijroi.read_roi(fh).astype(int)
def ROI2binaryimage(input_folder,output_folder,size):
compstruct = scipy.ndimage.generate_binary_structure(2, 2) # Mask for image dilation.
normal_color = 4294901760 # Stroke color (chosen by me) for normal cells.
stressed_color = 4278255360 # Stroke color (chosen by me) for stressed cells.
# Create the result folders if they don't exist.
# In the outputfolder as given to ROI2binaryimage, a folder is created named
# 'Manual_segmentation_results'. In this folder, two subfolders are created;
# 'Original_images' as well as 'Manual_segmentation'.
if not(os.path.isdir(os.path.join(output_folder, "Manual_segmentation_results"))):
os.makedirs(os.path.join(output_folder, "Manual_segmentation_results"))
os.makedirs(os.path.join(output_folder, "Manual_segmentation_results",
"Original_images"))
os.makedirs(os.path.join(output_folder, "Manual_segmentation_results",
"Manual_segmentations"))
# Ignoring hidden folders such as .DS_store in Mac, find folders for images + rois.
for folder in [p for p in os.listdir(input_folder) if not(p.startswith('.'))]:
blank = np.zeros(size, dtype=np.int8) # The image size
canvas = [np.copy(blank), np.copy(blank)] # Prep for background.
overlap = [np.copy(blank), np.copy(blank)] # Prep for the later overlapping images.
# Find the original image and save it to the new location; also save its name.
for image in [i for i in os.listdir(os.path.join(input_folder, folder)) if i
.endswith('.tif')]:
if not (np.all(size == list(np.asarray(misc.imread(os.path.join(input_folder,folder,image))).shape))):
raise ValueError("Given/assumed size doesn't match image size")
original_image_name = image
shutil.copy2((os.path.join(input_folder, folder, image)),
(os.path.join(output_folder, "Manual_segmentation_results",
"Original_images", image)))
notif = original_image_name[0:-4]
# Find all the .roi files in 'Roiset' and work with them.
for file in [f for f in os.listdir(os.path.join(
input_folder, folder, "Roiset")) if f.endswith('.roi')]:
temp_canvas = [np.copy(blank),np.copy(blank)]
with open((os.path.join(input_folder, folder, "Roiset", file)),"rb") as f:
[roi,roi_color] = ijroi.read_roi(f)
# Determine whether the points are valid (the roi has to fit in the given 'size')
valid = np.logical_and(np.min(roi, axis=1) >= 0, roi[:,0] < size[0], roi[:,1] < size[1])
valid_roi = roi[valid]
x = valid_roi[:,0]
y = valid_roi[:,1]
# Differentiate between the normal and stressed cell rois.
rr, cc = skimage.draw.polygon(x,y)
if roi_color == normal_color:
q = 0
if roi_color == stressed_color:
q = 1
# Add roi to the image.
canvas[q][rr,cc] = 1
temp_canvas[q][rr,cc] = 1
# Determine the overlap by adding the dilated version of the cell to the overlap file.
# (This is to ensure the cells do not 'touch' as well as do not 'overlap').
temp_canvas[q] = scipy.ndimage.binary_dilation( # Dilate cell in all directions.
temp_canvas[q], structure=compstruct).astype(temp_canvas[q].dtype)
overlap[q] = overlap[q] + temp_canvas[q]
# If there is an overlap of different cells, that area is set to 0.
for ax in np.arange(size[0]):
for ay in np.arange(size[1]):
for i in range(len(overlap)):
if overlap[i][ax,ay] > 1:
canvas[i][ax,ay] = 0
background = np.array([1 - np.clip((canvas[0]+canvas[1]), 0, 1)], ndmin=3, dtype=np.int8)
image = np.array([canvas[0], canvas[1]])
result = np.concatenate([background, image], axis=0)
result = skimage.util.img_as_ubyte(result)
imsave(os.path.join(output_folder, "Manual_segmentation_results",'Manual_segmentations',
(notif + '.tif')), result)
return None
def ROI2binaryimage(input_folder, output_folder, size):
compstruct = scipy.ndimage.generate_binary_structure(
2, 2) # Mask for image dilation.
normal_color = 4294901760 # Stroke color (chosen by me) for normal cells.
stressed_color = 4278255360 # Stroke color (chosen by me) for stressed cells.
# Create the result folders if they don't exist.
# In the outputfolder as given to ROI2binaryimage, a folder is created named
# 'Manual_segmentation_results'. In this folder, two subfolders are created;
# 'Original_images' as well as 'Manual_segmentation'.
if not (os.path.isdir(
os.path.join(output_folder, "Manual_segmentation_results"))):
os.makedirs(os.path.join(output_folder, "Manual_segmentation_results"))
os.makedirs(
os.path.join(output_folder, "Manual_segmentation_results",
"Original_images"))
os.makedirs(
os.path.join(output_folder, "Manual_segmentation_results",
"Manual_segmentations"))
# Ignoring hidden folders such as .DS_store in Mac, find folders for images + rois.
for folder in [
p for p in os.listdir(input_folder) if not (p.startswith('.'))
]:
blank = np.zeros(size, dtype=np.int8) # The image size
canvas = [np.copy(blank), np.copy(blank)] # Prep for background.
overlap = [np.copy(blank),
np.copy(blank)] # Prep for the later overlapping images.
# Find the original image and save it to the new location; also save its name.
for image in [
i for i in os.listdir(os.path.join(input_folder, folder))
if i.endswith('.tif')
]:
if not (np.all(size == list(
np.asarray(
misc.imread(os.path.join(input_folder, folder,
image))).shape))):
raise ValueError("Given/assumed size doesn't match image size")
original_image_name = image
shutil.copy2(
(os.path.join(input_folder, folder, image)),
(os.path.join(output_folder, "Manual_segmentation_results",
"Original_images", image)))
notif = original_image_name[0:-4]
# Find all the .roi files in 'Roiset' and work with them.
for file in [
f for f in os.listdir(
os.path.join(input_folder, folder, "Roiset"))
if f.endswith('.roi')
]:
temp_canvas = [np.copy(blank), np.copy(blank)]
with open((os.path.join(input_folder, folder, "Roiset", file)),
"rb") as f:
[roi, roi_color] = ijroi.read_roi(f)
# Determine whether the points are valid (the roi has to fit in the given 'size')
valid = np.logical_and(
np.min(roi, axis=1) >= 0, roi[:, 0] < size[0],
roi[:, 1] < size[1])
valid_roi = roi[valid]
x = valid_roi[:, 0]
y = valid_roi[:, 1]
# Differentiate between the normal and stressed cell rois.
rr, cc = skimage.draw.polygon(x, y)
if roi_color == normal_color:
q = 0
if roi_color == stressed_color:
q = 1
# Add roi to the image.
canvas[q][rr, cc] = 1
temp_canvas[q][rr, cc] = 1
# Determine the overlap by adding the dilated version of the cell to the overlap file.
# (This is to ensure the cells do not 'touch' as well as do not 'overlap').
temp_canvas[
q] = scipy.ndimage.binary_dilation( # Dilate cell in all directions.
temp_canvas[q],
structure=compstruct).astype(temp_canvas[q].dtype)
overlap[q] = overlap[q] + temp_canvas[q]
# If there is an overlap of different cells, that area is set to 0.
for ax in np.arange(size[0]):
for ay in np.arange(size[1]):
for i in range(len(overlap)):
if overlap[i][ax, ay] > 1:
canvas[i][ax, ay] = 0
background = np.array([1 - np.clip((canvas[0] + canvas[1]), 0, 1)],
ndmin=3,
dtype=np.int8)
image = np.array([canvas[0], canvas[1]])
result = np.concatenate([background, image], axis=0)
result = skimage.util.img_as_ubyte(result)
imsave(
os.path.join(output_folder, "Manual_segmentation_results",
'Manual_segmentations', (notif + '.tif')), result)
return None
import ijroi
import numpy as np
rootDir = '/Users/yanzhexu/Desktop/Research/GBM/aCGH_whole_tumor_maps_for_Neuro-Onc_dataset/CEFSL_slices_only/slice22/ROI for +C_3D_AXIAL_IRSPGR_Fast_IM-0005-0022.roi'
with open(rootDir, "rb") as file:
roi = ijroi.read_roi(file)
print isinstance(roi, np.ndarray)
