import numpy as np
import os
from tifffile import imread, imwrite
from skimage.segmentation import relabel_sequential
import skimage.segmentation as sksegmentation
import skimage.color as skcolor
import seaborn as sns
import copy
[docs]
class merge_segmentations():
"""
This class merges the segmentation of the binary mask which is ideal for macrophages that do not touch and the
cellpose based segmentation that is required to distinguish touching macrophages
"""
[docs]
def __init__(self):
"""
Init function and parameters
"""
self.debug = 1
self.segmentedimagename_cellpose = "labels_xy-integrated_gradients-correct_noexpand.tif"
self.segmentedimagename_conn = "labels_xy-connectedcomponents.tif"
self.maximalcellsize = 40000
self.mincellsize=5000
[docs]
def process_timepoint(self, segmenteddata_connected, segmenteddata_cellpose, savefolder):
"""
Merge segmentations of one timepoint and save resulting image files to disk.
:param segmenteddata_connected: imagefile with segmented data of connected component
:param segmenteddata_cellpose: imagefile with segmented data of cellpose segmentation
:param savefolder: folder where to save merged, segmented data
"""
print(segmenteddata_connected)
print(segmenteddata_cellpose)
print(savefolder)
# =============================================================================
# open images
# =============================================================================
# segmenteddata_connected = os.path.join(segmenteddata_connected_folder, "high_stack_002", "t00008", "labels_xy-connectedcomponents.tif")
# segmenteddata_cellpose = os.path.join(segmenteddata_cellpose_folder, "high_stack_002", "t00008", "labels_xy-integrated_gradients-correct_noexpand.tif")
# savefolder = os.path.join(segmenteddata_cellpose_folder, "high_stack_002", "t00008", "test.tif")
# open images
segmentedimage_connected = imread(segmenteddata_connected)
segmentedimage_cellpose = imread(segmenteddata_cellpose)
merged_image = copy.deepcopy(segmentedimage_connected)
# =============================================================================
# replace labels that are too large in size in multi-otsu thresholded image with Cellpose based segmentation
# =============================================================================
#make binary 0 or 1
im_binary_cellpose = segmentedimage_cellpose > 0.5
#get maximum label in connected image
maximumlabel = np.max(segmentedimage_connected)
#process all labels
for currentlabel in range(1, maximumlabel+1):
#currentlabel = 2
#get indices for specific label
currentlabel_indices = np.where(segmentedimage_connected == currentlabel)
#process all labels that are bigger than the maximalcellsize
if currentlabel_indices[0].shape[0] > self.maximalcellsize:
print("divide:" + str(currentlabel))
merged_image[currentlabel_indices] = im_binary_cellpose[currentlabel_indices] * (segmentedimage_cellpose[currentlabel_indices]+maximumlabel+1)
#relabel image so that labels go from 1 to x
relab, fw, inv = relabel_sequential(merged_image)
# =============================================================================
# merge small labels with close by macrophages
# =============================================================================
maximumlabel_relab = np.max(relab)
# process all labels
for currentlabel in range(1, maximumlabel_relab + 1):
# currentlabel = 13
#get indices for specific label
currentlabel_indices = np.where(relab == currentlabel)
if currentlabel_indices[0].shape[0] < self.mincellsize:
currentcrop = relab[
np.min(currentlabel_indices[0]):np.max(currentlabel_indices[0]),
np.min(currentlabel_indices[1]):np.max(currentlabel_indices[1]),
np.min(currentlabel_indices[2]):np.max(currentlabel_indices[2])]
unique_label, unique_counts = np.unique(currentcrop, return_counts=True)
sorted_counts = -np.sort(-unique_counts)
newlabel=0
for count_iter in range(len(sorted_counts)):
label = unique_label[unique_counts==(sorted_counts[count_iter])][0]
if label !=0 and label!=currentlabel:
newlabel = label
break
print(newlabel)
relab = np.where(relab == currentlabel,newlabel, relab)
# =============================================================================
# save merged image
# =============================================================================
# relabel image so that labels go from 1 to x
relab2, fw2, inv2 = relabel_sequential(relab)
imwrite(os.path.join(savefolder, 'labels_xy-merged.tif'), relab2)
relab_color = np.uint8(255 * skcolor.label2rgb(relab2.copy(),
colors=sns.color_palette('hls', n_colors=16),
bg_label=0))
imwrite(os.path.join(savefolder, 'labels_xy-connectedcomponentsRGB.tif'),
np.uint8(relab_color))
[docs]
def iterate_throughfolder(self, segmenteddata_connected_folder, segmenteddata_cellpose_folder, parentsavefolder):
"""
Iterate through parentfolder to process individual timepoints.
:param segmenteddata_connected_folder: folder with the data of the connected-components segmentation
:param segmenteddata_cellpose_folder: folder with the data of the cellpose-based segmentation
:param parentsavefolder: parent folder to save the image output
"""
# get all timepoints from folder
dir_list = os.listdir(segmenteddata_connected_folder)
regionlist = []
for path in dir_list:
if path.startswith('high'):
regionlist.append(path)
regionlist.sort()
print(regionlist)
for region in regionlist:
parentfolder_segmented = os.path.join(segmenteddata_connected_folder, region)
# get all timepoints from folder
dir_list = os.listdir(parentfolder_segmented)
timepointlist = []
for path in dir_list:
if path.startswith('t'):
timepointlist.append(path)
timepointlist.sort()
print(timepointlist)
# if you establish parameters, only open first timepoint
if self.debug == 1:
timepointlist = ["t00000"]
for i_time in timepointlist:
# if i_time<"t00069":
# continue
# construct filepath
segmenteddata_connectedcomponents = os.path.join(parentfolder_segmented, i_time, self.segmentedimagename_conn)
segmenteddata_cellpose = os.path.join(segmenteddata_cellpose_folder, region, i_time, self.segmentedimagename_cellpose)
savefolder = os.path.join(parentsavefolder, region, i_time)
try:
os.makedirs(savefolder)
except OSError as error:
pass
# process timepoint in region
self.process_timepoint(segmenteddata_connectedcomponents, segmenteddata_cellpose, savefolder)
if __name__ == '__main__':
#generate class function
merge_it = merge_segmentations()
parentfolder = "/archive/bioinformatics/Danuser_lab/Fiolka/LabMembers/Stephan/multiscale_data/xenograft_experiments/U2OS_WT/20220729_Daetwyler_U2OS"
segmenteddata_cellpose_folder = os.path.join(parentfolder, "Experiment0001_highresSeg_run_again", "processed_segmentation")
segmenteddata_connected_folder = os.path.join(parentfolder, "Experiment0001_highresSeg_connectedComp_multiOtsu",
"processed_segmentation_thres130")
parentsavefolder = os.path.join(parentfolder, "Experiment0001_highresSeg_connectedComp_multiOtsu",
"processed_segmentation_merged130")
merge_it.debug = 0 #if 1 only process timepoint zero
merge_it.iterate_throughfolder(segmenteddata_connected_folder, segmenteddata_cellpose_folder, parentsavefolder)