def shapeindex_preprocess(self,im):
''' apply shap index map at three scales'''
sh = np.zeros((im.shape[0],im.shape[1],3))
if np.max(im) ==0:
return sh
# pad to minimize edge artifacts
sh[:,:,0] = shape_index(im,1, mode='reflect')
sh[:,:,1] = shape_index(im,1.5, mode='reflect')
sh[:,:,2] = shape_index(im,2, mode='reflect')
#sh = 0.5*(sh+1.0)
return sh
def run(params):
image_location = params['inputImagePath']
result_location = params['resultPath']
sigma = float(params['sigma'])
tCount = int(params['TCount'])
zCount = int(params['ZCount'])
if not os.path.exists(image_location):
print(f"Error: {image_location} does not exist")
return
image_data = imread(image_location)
dims = image_data.shape
shape_image = np.empty(image_data.shape, dtype=np.float32)
# 3D+T
if tCount > 1 and zCount > 1:
print(f"Applying to 3D+T case with dims: {image_data.shape}")
for t in range(0, dims[0]):
for z in range(0, dims[1]):
shape_image[t, z, :, :] = shape_index(image_data[t, z, :, :],
sigma=sigma,
mode='reflect').astype(
np.float32)
axes = 'YXZT'
# 2D+T or 3D
elif (tCount > 1 and zCount == 1) or (tCount == 1 and zCount > 1):
print(f"Applying to 2D+T or 3D case with dims: {image_data.shape}")
for d in range(0, dims[0]):
shape_image[d, :, :] = shape_index(image_data[d, :, :],
sigma=sigma,
mode='reflect').astype(
np.float32)
if tCount > 1:
axes = 'YXT'
else:
axes = 'YXZ'
# 2D
else:
print(f"Applying to 2D case with dims: {image_data.shape}")
shape_image = shape_index(image_data, sigma=sigma, mode='reflect')
axes = 'YX'
# NaNs are usually returned - convert these to possible pixel values
shape_image = np.nan_to_num(shape_image)
if image_data.dtype == np.uint16:
shape_image = img_as_uint(shape_image)
else:
shape_image = img_as_ubyte(shape_image)
imsave(result_location, shape_image, metadata={'axes': axes})
def get_shape_index_features(data, size=10):
# http://scikit-image.org/docs/dev/auto_examples/features_detection/plot_shape_index.html#sphx-glr-auto-examples-features-detection-plot-shape-index-py
# -- filterbank1 on original image
result = []
for i in range(len(data)):
img = data[i]
image_gray = color.rgb2gray(img)
s = shape_index(image_gray)
# In this example we want to detect 'spherical caps',
# so we threshold the shape index map to
# find points which are 'spherical caps' (~1)
target = 1
delta = 0.05
point_y, point_x = np.where(np.abs(s - target) < delta)
point_z = image_gray[point_y, point_x]
# The shape index map relentlessly produces the shape, even that of noise.
# In order to reduce the impact of noise, we apply a Gaussian filter to it,
# and show the results once in
# s_smooth = ndi.gaussian_filter(s, sigma=0.5)
# point_y_s, point_x_s = np.where(np.abs(s_smooth - target) < delta)
# point_z_s = image_gray[point_y_s, point_x_s]
# plt.imshow(hog_image)
# plt.show()
point_z.sort()
result.append(point_z[-size:])
result = normalize_features(result, v_max=1.0, v_min=0.0)
return result
def get_hand_crafted(one_image):
""" Extracts various features out of the given image
:param array one_image: the image from which features are to be extracted
:return: the features associated with this image
:rtype: Numpy array of size (38, 1)
"""
#Select wavelet decomposition level so as to have the
#same number of approximation coefficients
if (one_image.shape[0] == 1000):
wavedec_level = 9
elif (one_image.shape[0] == 64):
wavedec_level = 5
hist = histogram(one_image, nbins=20, normalize=True)
features = hist[0]
blob_lo = blob_log(one_image,
max_sigma=2.5,
min_sigma=1.5,
num_sigma=5,
threshold=0.05)
shape_ind = shape_index(one_image)
shape_hist = np.histogram(shape_ind, range=(-1, 1), bins=9)
shan_ent = shannon_entropy(one_image)
max_val = one_image.max()
min_val = one_image.min()
variance_val = np.var(one_image)
wavelet_approx = pywt.wavedec2(one_image, 'haar',
level=wavedec_level)[0].flatten()
features = np.concatenate([
features, [blob_lo.shape[0]], shape_hist[0], [shan_ent], [max_val],
[min_val], [variance_val], wavelet_approx
])
return features
def calc_shape_index_scores(dose_dataset):
shape_key = [
-1, -7 / 8, -5 / 8, -3 / 8, -1 / 8, 1 / 8, 3 / 8, 5 / 8, 7 / 8, 1
]
scores_list = []
arr_size = len(dose_dataset.pixel_array)
for j in range(0, len(shape_key) - 1):
tick = 0
delta = 0.05
for i in range(0, arr_size):
im = dose_dataset.pixel_array[i]
sh_im = shape_index(im)
s_smooth = ndi.gaussian_filter(sh_im, sigma=0.875)
point_y_s, point_x_s = np.where((sh_im < shape_key[j + 1])
& (sh_im > shape_key[j]))
point_z_s = im[point_y_s, point_x_s]
tick += len(point_z_s)
scores_list.append(np.around(tick / arr_size, 3))
return scores_list
def process_single_image(filename, image_format, scale_metadata_path,
threshold_radius, smooth_radius, brightness_offset,
crop_radius, smooth_method):
image = imageio.imread(filename, format=image_format)
scale = _get_scale(image, scale_metadata_path)
if crop_radius > 0:
c = crop_radius
image = image[c:-c, c:-c]
pixel_threshold_radius = int(np.ceil(threshold_radius / scale))
pixel_smoothing_radius = smooth_radius * pixel_threshold_radius
thresholded = pre.threshold(image,
sigma=pixel_smoothing_radius,
radius=pixel_threshold_radius,
offset=brightness_offset,
smooth_method=smooth_method)
quality = shape_index(image, sigma=pixel_smoothing_radius, mode='reflect')
skeleton = morphology.skeletonize(thresholded) * quality
framedata = csr.summarise(skeleton, spacing=scale)
framedata['squiggle'] = np.log2(framedata['branch-distance'] /
framedata['euclidean-distance'])
framedata['scale'] = scale
framedata.rename(columns={'mean pixel value': 'mean shape index'},
inplace=True)
framedata['filename'] = filename
return image, thresholded, skeleton, framedata
def experimental_thresholding(image, mask=None, window_size=15,
gaussian_sigma=3.0, shift=0.2, target=-0.5, quotient=1.2,
return_threshold=False, **kwargs):
"""
A novel thresholding method basing upon the shape index as defined by [Koenderink1992]_, and [Bataineh2011]_
automatic adaptive thresholding. The method is due to be explained in detail in the future.
.. [Koenderink1992] Koenderink and van Doorn (1992) Image Vision Comput.
DOI: `10.1016/0262-8856(92)90076-F <https://dx.doi.org/10.1016/0262-8856(92)90076-F>`_
.. [Bataineh2011] Bataineh et al. (2011) Pattern Recognit. Lett.
DOI: `10.1016/j.patrec.2011.08.001 <https://dx.doi.org/10.1016/j.patrec.2011.08.001>`_
:param image: Input image
:param mask: Possible mask denoting a ROI
:param window_size: Window size
:param gaussian_sigma: Sigma of the Gaussian used for smoothing
:param shift: Shift parameter
:param target: Target shape index parameter
:param quotient: Quotient parameter
:param return_threshold: Whether to return a binarization, or the actual threshold values
:param kwargs: For compatibility
:return:
"""
# novel method based upon shape index and Bataineh thresholding
means, stddev = mean_and_std(image, window_size)
with np.errstate(invalid='ignore'):
sim = shape_index(image, gaussian_sigma)
if mask is not None:
adaptive_stddev = (stddev - stddev[mask].min()) / (stddev[mask].max() - stddev[mask].min())
image_mean = image[mask].mean()
else:
adaptive_stddev = (stddev - stddev.min()) / (stddev.max() - stddev.min())
image_mean = image.mean()
if numexpr:
threshold = numexpr.evaluate(
"(exp((-(sim - target)**2)/quotient) + shift)*"
"means*"
"((image_mean + stddev) * (stddev + adaptive_stddev))/(means**2 - stddev)"
)
else:
threshold = (
(np.exp((-(sim - target)**2)/quotient) + shift) *
means *
((image_mean + stddev) * (stddev + adaptive_stddev))/(means**2 - stddev)
)
if return_threshold:
return threshold
else:
with warnings.catch_warnings():
warnings.simplefilter('ignore')
return image < threshold
def test_shape_index():
square = np.zeros((5, 5))
square[2, 2] = 4
s = shape_index(square, sigma=0.1)
assert_almost_equal(
s, np.array([[ np.nan, np.nan, -0.5, np.nan, np.nan],
[ np.nan, 0, np.nan, 0, np.nan],
[ -0.5, np.nan, -1, np.nan, -0.5],
[ np.nan, 0, np.nan, 0, np.nan],
[ np.nan, np.nan, -0.5, np.nan, np.nan]])
)
def test_shape_index():
square = np.zeros((5, 5))
square[2, 2] = 4
s = shape_index(square, sigma=0.1)
assert_almost_equal(
s, np.array([[ np.nan, np.nan, -0.5, np.nan, np.nan],
[ np.nan, 0, np.nan, 0, np.nan],
[ -0.5, np.nan, -1, np.nan, -0.5],
[ np.nan, 0, np.nan, 0, np.nan],
[ np.nan, np.nan, -0.5, np.nan, np.nan]])
)
def test_shape_index():
square = np.zeros((5, 5))
square[2, 2] = 4
with expected_warnings(['divide by zero', 'invalid value']):
s = shape_index(square, sigma=0.1)
assert_almost_equal(
s,
np.array([[np.nan, np.nan, -0.5, np.nan, np.nan],
[np.nan, 0, np.nan, 0, np.nan],
[-0.5, np.nan, -1, np.nan, -0.5],
[np.nan, 0, np.nan, 0, np.nan],
[np.nan, np.nan, -0.5, np.nan, np.nan]]))
def test_shape_index():
square = np.zeros((5, 5))
square[2, 2] = 4
with expected_warnings(['divide by zero', 'invalid value']):
s = shape_index(square, sigma=0.1)
assert_almost_equal(
s, np.array([[ np.nan, np.nan, -0.5, np.nan, np.nan],
[ np.nan, 0, np.nan, 0, np.nan],
[ -0.5, np.nan, -1, np.nan, -0.5],
[ np.nan, 0, np.nan, 0, np.nan],
[ np.nan, np.nan, -0.5, np.nan, np.nan]])
)
def postprocess_ws(im, yp):
'''Watershed based postprocessing using image and its pixel probabilities'''
# mask dilated
mask = (yp[:, :, 0] > 0.4)
# watershed potential
d = shape_index(im, 1.5, mode='reflect')
# markers
# get poles from contour predictions as markers
sh = shape_index(yp[:, :, 1], 1, mode='reflect')
markers, c = label(yp[:, :, 0] > 0.95)
# ther markers should be unique to each cell
markers = markers * (sh < -0.5) # only poles
ws = watershed(d,
markers=markers,
watershed_line=True,
mask=mask,
compactness=1,
connectivity=1)
return ws
def add_noise(im, sensitivity=0.1, invert=False, seed=42):
'''preprocessing by adding random noise to image only where the shape index map is around -0.4 for phase contrast images'''
t0, s = -0.4, 0.025
sim = -shape_index(im, 1)
if invert:
sim = -sim
ed = np.logical_and(sim > t0 - s, sim > t0 + s) * 1.0
#ed = erosion(ed)
np.random.seed(seed)
noise = np.random.rand(ed.shape[0], ed.shape[1])
noise = noise * (ed + 0.5 * np.random.rand(ed.shape[0], ed.shape[1]))
img = normalize2max(im) + sensitivity * noise
return img
def preprocess(self,im):
n = len(self.scalesvals)
sh = np.zeros((im.shape[0],im.shape[1],n))
if np.max(im) ==0:
return sh
pw = 15
im = pad(im,pw,'reflect')
sh = np.zeros((im.shape[0],im.shape[1],n))
for i in range(n):
sh[:,:,i] = shape_index(im,self.scalesvals[i])
return sh[pw:-pw,pw:-pw,:]
def test_shape_index():
# software floating point arm doesn't raise a warning on divide by zero
# https://github.com/scikit-image/scikit-image/issues/3335
square = np.zeros((5, 5))
square[2, 2] = 4
with expected_warnings([r'divide by zero|\A\Z', r'invalid value|\A\Z']):
s = shape_index(square, sigma=0.1)
assert_almost_equal(
s, np.array([[ np.nan, np.nan, -0.5, np.nan, np.nan],
[ np.nan, 0, np.nan, 0, np.nan],
[ -0.5, np.nan, -1, np.nan, -0.5],
[ np.nan, 0, np.nan, 0, np.nan],
[ np.nan, np.nan, -0.5, np.nan, np.nan]])
)
def test_shape_index():
# software floating point arm doesn't raise a warning on divide by zero
# https://github.com/scikit-image/scikit-image/issues/3335
square = np.zeros((5, 5))
square[2, 2] = 4
with expected_warnings([r'divide by zero|\A\Z', r'invalid value|\A\Z']):
s = shape_index(square, sigma=0.1)
assert_almost_equal(
s,
np.array([[np.nan, np.nan, -0.5, np.nan, np.nan],
[np.nan, 0, np.nan, 0, np.nan],
[-0.5, np.nan, -1, np.nan, -0.5],
[np.nan, 0, np.nan, 0, np.nan],
[np.nan, np.nan, -0.5, np.nan, np.nan]]))
def calc_shape_diff_scores(dose_dataset):
shape_key = [-1,-7/8,-5/8,-3/8,-1/8,1/8,3/8,5/8,7/8,1]
arr_list = []
scores_list = []
arr_size = len(dose_dataset.pixel_array)
for i in range (0,arr_size):
im = dose_dataset.pixel_array[i]
sh_im = shape_index(im)
arr_list.append(sh_im)
score = np.abs(np.diff( np.nan_to_num(arr_list), axis=0).sum())/arr_size
#scores_list.append(np.around(tick/arr_size,3))
return score
def main_loop(pps_list, mirror, params_dict):
virago_dir = '{}/v3-analysis'.format(os.getcwd())
vcount_dir = '{}/vcounts'.format(virago_dir)
img_dir = '{}/processed_images'.format(virago_dir)
histo_dir = '{}/histograms'.format(virago_dir)
overlay_dir = '{}/overlays'.format(virago_dir)
filo_dir = '{}/filo'.format(virago_dir)
fluor_dir = '{}/fluor'.format(virago_dir)
if not os.path.exists(virago_dir):
os.makedirs(virago_dir)
if not os.path.exists(img_dir): os.makedirs(img_dir)
if not os.path.exists(histo_dir): os.makedirs(histo_dir)
if not os.path.exists(fluor_dir): os.makedirs(fluor_dir)
if not os.path.exists(filo_dir): os.makedirs(filo_dir)
if not os.path.exists(overlay_dir): os.makedirs(overlay_dir)
if not os.path.exists(vcount_dir): os.makedirs(vcount_dir)
cam_micron_per_pix = params_dict['cam_micron_per_pix']
mag = params_dict['mag']
pix_per_um = mag / cam_micron_per_pix
spacing = 1 / pix_per_um
conv_factor = (cam_micron_per_pix / mag)**2
exo_toggle = params_dict['exo_toggle']
cv_cutoff = params_dict['cv_cutoff']
perc_range = params_dict['perc_range']
# IRISmarker = imread('/usr3/bustaff/ajdevaux/virago/images/IRISmarker_new.tif')
IRISmarker = params_dict['IRISmarker']
# IRISmarker_exo = skio.imread('images/IRISmarker_v4_topstack.tif')
# pps_list, mirror = vpipes.mirror_finder(pps_list)
passes_per_spot = len(pps_list)
spot_ID = pps_list[0][:-8]
scans_counted = [int(file.split(".")[2]) for file in pps_list]
first_scan = min(scans_counted)
circle_dict, marker_dict, overlay_dict, shift_dict = {}, {}, {}, {}
vdata_dict = vpipes.get_vdata_dict(exo_toggle, version="3.x")
# missing_data = set(range(1,pass_counter+1)).difference(scans_counted)
#
# if missing_data != set():
# print("Missing image files... fixing...\n")
# for scan in missing_data:
# bad_scan = '{0}.{1}'.format(*(spot_ID, str(scan).zfill(3)))
# vdata_dict.update({'img_name':bad_scan})
#
# with open('{}/{}.vdata.txt'.format(vcount_dir,bad_scan),'w') as f:
# for k,v in vdata_dict.items():
# f.write('{}: {}\n'.format(k,v))
# print("Writing blank data files for {}".format(bad_scan))
total_shape_df = pd.DataFrame()
for scan in range(
0, passes_per_spot): ##Main Loop for image processing begins here.
img_stack = tuple(file for file in pps_list
if file.startswith(pps_list[scan]))
fluor_files = [
file for file in img_stack if file.split(".")[-2] in 'ABC'
]
if fluor_files:
img_stack = tuple(file for file in img_stack
if file not in fluor_files)
print(
"\nFluorescent channel(s) detected: {}\n".format(fluor_files))
topstack_img = img_stack[0]
name_split = topstack_img.split('.')
img_name = '.'.join(name_split[:-1])
spot_num, pass_num = map(int, name_split[1:3])
if name_split[-1] == 'tif':
tiff_toggle = True
else:
tiff_toggle = False
pic3D = vpipes.load_image(img_stack, tiff_toggle)
print("{} Loaded\n".format(img_name))
validity = True
# if convert_tiff == True:
# vpipes.pgm_to_tiff(pic3D, img_name, img_stack,
# tiff_compression=1, archive_pgm=True)
#
zslice_count, nrows, ncols = pic3D.shape
total_pixels = nrows * ncols
if mirror.size == total_pixels:
pic3D = pic3D / mirror
print("Applying mirror to image stack...\n")
pic3D_norm = np.uint8(normalize(pic3D, None, 0, 255, NORM_MINMAX))
pic3D_clahe = vimage.cv2_clahe_3D(pic3D_norm,
kernel_size=(1, 1),
cliplim=4)
pic3D_rescale = vimage.rescale_3D(pic3D_clahe, perc_range=perc_range)
print("Contrast adjusted\n")
#Many operations are on the Z-stack compressed image.
#Several methods to choose, but Standard Deviation works well.
# maxmin_proj_rescale = np.max(pic3D_rescale, axis = 0) - np.min(pic3D_rescale, axis = 0)
sd_proj_rescale = np.std(pic3D_rescale, axis=0)
#Convert to 8-bit for OpenCV comptibility
sd_proj_rescale = np.uint8(
normalize(sd_proj_rescale, None, 0, 255, NORM_MINMAX))
if pass_num == 1:
marker = IRISmarker
else:
marker = found_markers
if img_name not in marker_dict:
marker_locs = vimage.marker_finder(pic3D_rescale[0],
marker=marker,
thresh=0.6)
marker_dict[img_name] = marker_locs
else:
marker_locs = marker_dict[img_name]
pos_plane_list = vquant.measure_focal_plane(
pic3D_norm, marker_locs, exo_toggle, marker_shape=IRISmarker.shape)
if pos_plane_list != []:
pos_plane = max(pos_plane_list)
else:
pos_plane = zslice_count // 3
pic_rescale_pos = pic3D_rescale[pos_plane]
overlay_dict['.'.join(img_name.split('.')[1:])] = sd_proj_rescale
print("Using image {} from stack\n".format(
str(pos_plane + 1).zfill(3)))
# if pass_counter <= 15:
# overlay_mode = 'series'
# else:
overlay_mode = 'baseline'
if pass_num == first_scan:
print("First Valid Scan\n")
valid_shift = (0, 0)
overlay_toggle = False
else:
prescan_img, postscan_img = vimage._dict_matcher(overlay_dict,
spot_num,
pass_num,
mode=overlay_mode)
overlay_toggle = True
# if img_name in shift_dict:
# valid_shift = shift_dict[img_name]
# else:
ORB_shift = vimage.measure_shift_ORB(prescan_img,
postscan_img,
ham_thresh=10,
show=False)
# for coord in ORB_shift:
# if abs(coord) < 75:
#
valid_shift = ORB_shift
# else: ##In case ORB fails to give a good value
# # overlay_toggle = False
# print("Using alternative shift measurement...\n")
# mean_shift, overlay_toggle = vimage.measure_shift(marker_dict,pass_num,
# spot_num,mode=overlay_mode
# )
# valid_shift = mean_shift
print("Valid Shift: {}\n".format(valid_shift))
img_overlay = vimage.overlayer(prescan_img, postscan_img,
valid_shift)
shape_mask = vimage.shape_mask_shift(shape_mask, valid_shift)
img_overlay_difference = np.int16(img_overlay[:, :, 1]) - np.int16(
img_overlay[:, :, 0])
median_overlay = np.median(img_overlay_difference)
sd_overlay = np.std(img_overlay_difference)
print(median_overlay, sd_overlay)
# overlay_name = "{}_overlay_{}".format(img_name, overlay_mode)
# vimage.gen_img_deets(img_overlay_difference, name=overlay_name, savedir=overlay_dir)
if (overlay_toggle == False) & (pass_num != first_scan):
validity = False
print("No compatible markers, cannot compute shift")
#
# else:
# print("Cannot overlay images\n")
if spot_num in circle_dict: #Get the location of the Antibody spot
spot_coords = circle_dict[spot_num]
shift_x = spot_coords[0] + valid_shift[1]
shift_y = spot_coords[1] + valid_shift[0]
spot_coords = (shift_x, shift_y, spot_coords[2])
else: #Find the Antibody spot if it has not already been determined
circles = None
cannyMax = 200
cannyMin = 100
while type(circles) == type(None):
circles = HoughCircles(sd_proj_rescale,
HOUGH_GRADIENT,
1,
minDist=500,
param1=cannyMax,
param2=cannyMin,
minRadius=300,
maxRadius=600)
cannyMax -= 50
cannyMin -= 25
spot_coords = tuple(map(lambda x: round(x, 0), circles[0][0]))
print("Spot center coordinates (row, column, radius): {}\n".format(
spot_coords))
circle_dict[spot_num] = spot_coords
row, col = np.ogrid[:nrows, :ncols]
width = col - spot_coords[0]
height = row - spot_coords[1]
rad = spot_coords[2] - 25
disk_mask = (width**2 + height**2 > rad**2)
marker_mask, found_markers = vimage.marker_masker(
pic3D_rescale[0], marker_locs, marker)
full_mask = disk_mask + marker_mask
# vimage.image_details(pic3D_norm[pos_plane], pic3D_clahe[pos_plane], pic3D_rescale[pos_plane].copy(), pic_canny)
# maxmin_proj_rescale_masked = np.ma.array(maxmin_proj_rescale, mask=full_mask).filled(fill_value=np.nan)
# maxmin_median = np.nanmedian(maxmin_proj_rescale_masked)
#*********************************************************************************************#
# if pass_num > first_scan:
# if overlay_toggle == True:
# img_overlay = np.ma.array(img_overlay, mask=full_mask).filled(fill_value=np.nan)
# if Ab_spot_mode == False:
# pic_to_show = sd_proj_rescale
# elif exo_toggle == True:
# pic_to_show = sd_proj_rescale
# else:
if pass_num == 1:
pic_to_show = sd_proj_rescale
else:
pic_to_show = img_overlay_difference
#*********************************************************************************************#
with warnings.catch_warnings():
##RuntimeWarning ignored: invalid values are expected
warnings.simplefilter("ignore")
warnings.warn(RuntimeWarning)
shapedex = shape_index(pic_rescale_pos)
shapedex = np.ma.array(shapedex,
mask=full_mask).filled(fill_value=np.nan)
if pass_num > first_scan:
shapedex = np.ma.array(shapedex,
mask=shape_mask).filled(fill_value=-1)
shapedex_gauss = gaussian_filter(shapedex, sigma=1)
pix_area = np.count_nonzero(np.invert(np.isnan(shapedex)))
area_sqmm = round((pix_area * conv_factor) * 1e-6, 6)
##Pixel topology classifications
background = 0
ridge = 0.5
sphere = 1
bg_rows, bg_cols = zip(*vquant.classify_shape(
shapedex, sd_proj_rescale, background, delta=0.25, intensity=0))
sd_proj_bg = sd_proj_rescale[bg_rows, bg_cols]
sd_proj_bg_median = np.median(sd_proj_bg) ##Important
sd_proj_bg_stdev = np.std(sd_proj_bg)
print("Median intensity of spot background={}, SD={}".format(
round(sd_proj_bg_median, 4), round(sd_proj_bg_stdev, 4)))
# if Ab_spot_mode == True:
# if exo_toggle == True:
ridge_thresh = sd_proj_bg_median * 3.5
sphere_thresh = sd_proj_bg_median * 2.5
ridge_thresh_s = sd_proj_bg_median * 3.5
# else:
# ridge_thresh = sd_proj_bg_median+sd_proj_bg_stdev*2
# sphere_thresh = sd_proj_bg_median+sd_proj_bg_stdev*2
# ridge_thresh_s = sd_proj_bg_median+sd_proj_bg_stdev*3
# else:
# ridge_thresh = sd_proj_bg_median+sd_proj_bg_stdev*2.75
# sphere_thresh = sd_proj_bg_median+sd_proj_bg_stdev*2.75
# ridge_thresh_s = sd_proj_bg_median+sd_proj_bg_stdev*2.75
ridge_list = vquant.classify_shape(shapedex,
sd_proj_rescale,
ridge,
delta=0.25,
intensity=ridge_thresh)
sphere_list = vquant.classify_shape(shapedex,
sd_proj_rescale,
sphere,
delta=0.2,
intensity=sphere_thresh)
ridge_list_s = vquant.classify_shape(shapedex_gauss,
sd_proj_rescale,
ridge,
delta=0.3,
intensity=ridge_thresh_s)
pix_list = ridge_list + sphere_list
ridge_list_s = list(set(pix_list) - set(ridge_list_s))
pix_list = pix_list + ridge_list_s
pic_binary = np.zeros_like(sd_proj_rescale, dtype=int)
if not pix_list == []:
rows, cols = zip(*pix_list)
pic_binary[rows, cols] = 1
pic_binary = binary_fill_holes(pic_binary)
#*********************************************************************************************#
vdata_dict.update({
'img_name': img_name,
# 'spot_type': spot_type,
'area_sqmm': area_sqmm,
'valid_shift': valid_shift,
'overlay_mode': overlay_mode,
'pos_plane': pos_plane,
'spot_coords': spot_coords,
'marker_locs': marker_locs,
'classifier_median': round(sd_proj_bg_median, 4)
})
#*********************************************************************************************#
##EXTRACT DATA FROM THE BINARY IMAGE
prop_list = [
'label', 'coords', 'area', 'centroid', 'moments_central', 'bbox',
'filled_image', 'major_axis_length', 'minor_axis_length'
]
shape_df = vquant.binary_data_extraction(pic_binary,
pic3D[pos_plane],
prop_list,
pix_range=(3, 500))
print('S')
if not shape_df.empty:
particle_mask = vquant.particle_masker(pic_binary, shape_df,
pass_num, first_scan)
if pass_num == first_scan:
shape_mask = binary_dilation(particle_mask, iterations=3)
else:
shape_mask = np.add(
shape_mask, binary_dilation(particle_mask, iterations=2))
shape_df['pass_number'] = [pass_num] * len(shape_df.index)
shape_df['coords'] = shape_df.coords.apply(
lambda a: [tuple(x) for x in a])
shape_df['bbox'] = shape_df.bbox.map(vquant.bbox_verts)
print('R')
else:
print("----No valid particle shapes----\n")
vdata_dict.update({'total_valid_particles': 0, 'validity': False})
print(shape_df)
with open('{}/{}.vdata.txt'.format(vcount_dir, img_name),
'w') as f:
for k, v in vdata_dict.items():
f.write('{}: {}\n'.format(k, v))
continue
#*********************************************************************************************#
filo_pts_tot, round_pts_tot = [], []
z_intensity_list, max_z_slice_list, max_z_stacks, shape_validity = [],[],[],[]
greatest_max_list, sd_above_med_difference = [], []
intensity_increase_list = []
print('Measuring particle intensities...\n')
for coord_array in shape_df.coords:
coord_set = set(coord_array)
filo_pts = len(coord_set.intersection(ridge_list))
filo_pts = filo_pts + (len(coord_set.intersection(ridge_list_s)) *
0.15)
round_pts = len(coord_set.intersection(sphere_list))
filo_pts_tot.append(filo_pts)
round_pts_tot.append(round_pts)
# if pic3D.ndim > 2:
all_z_stacks = np.array(
[pic3D[:, coords[0], coords[1]] for coords in coord_array])
greatest_max = np.max(all_z_stacks)
max_z_stack = all_z_stacks[np.where(
all_z_stacks == np.max(all_z_stacks))[0][0]].tolist()
if max_z_stack[0] >= max_z_stack[-1]:
shape_validity.append(True)
else:
shape_validity.append(False)
maxmax_z = max(max_z_stack)
max_z_slice = max_z_stack.index(maxmax_z)
z_intensity = (maxmax_z - min(max_z_stack)) * 100
std_z_stack = list(np.round(np.std(all_z_stacks, axis=0), 4))
max_z_slice_list.append(max_z_slice)
max_z_stacks.append(max_z_stack)
z_intensity_list.append(z_intensity)
greatest_max_list.append(greatest_max)
if (pass_num > first_scan) & (overlay_toggle == True):
intensity_increase = max([
img_overlay_difference[coords[0], coords[1]]
for coords in coord_array
])
intensity_increase_list.append(intensity_increase)
else:
intensity_increase_list = [np.nan] * len(shape_df)
print('N')
shape_df['max_z_slice'] = max_z_slice_list
shape_df['max_z_stack'] = max_z_stacks
shape_df['z_intensity'] = z_intensity_list
shape_df['greatest_max'] = greatest_max_list
shape_df['validity'] = shape_validity
shape_df['filo_points'] = filo_pts_tot
shape_df['round_points'] = round_pts_tot
shape_df['intensity_increase'] = intensity_increase_list
bbox_pixels = [
vquant.get_bbox_pixels(bbox, pic3D[z])
for i, z, bbox in shape_df[['max_z_slice', 'bbox']].itertuples()
]
median_bg_list, shape_df['cv_bg'] = zip(*map(
lambda x: (np.median(x), np.std(x) / np.mean(x)), bbox_pixels))
shape_df['perc_contrast'] = (
(shape_df['greatest_max'] - median_bg_list) * 100 / median_bg_list)
shape_df.loc[shape_df.perc_contrast <= 0, 'validity'] = False
shape_df.loc[shape_df.cv_bg > cv_cutoff, 'validity'] = False
shape_df.loc[shape_df.intensity_increase < 40, 'validity'] = False
# if len(shape_df) > 1:
# regression = smapi.OLS(shape_df.z_intensity, shape_df.perc_contrast).fit()
# outlier_df = regression.outlier_test()
# shape_df.loc[outlier_df['bonf(p)'] < 0.5, 'validity'] = False
shape_df = vquant.remove_overlapping_objs(shape_df, radius=10)
#---------------------------------------------------------------------------------------------#
##Filament Measurements
shape_df['circularity'] = list(
map(lambda A, P: round((4 * np.pi * A) / (perimeter(P)**2), 4),
shape_df.area, shape_df.filled_image))
shape_df['ellipticity'] = round(shape_df.major_axis_length /
shape_df.minor_axis_length,
4) #max val = 1
shape_df['eccentricity'] = shape_df.moments_central.map(
vquant.eccentricity)
shape_df = shape_df[(shape_df['filo_points'] +
shape_df['round_points']) >= 1]
shape_df['filo_score'] = (
(shape_df['filo_points'] / shape_df['area']) -
(shape_df['round_points'] / shape_df['area']))
shape_df['roundness_score'] = ((shape_df['round_points'] /
shape_df['area']))
#---------------------------------------------------------------------------------------------#
filolen_df = pd.DataFrame([
vfilo.measure_fiber_length(coords, spacing=spacing)
for coords in shape_df.coords
],
columns=['fiber_length', 'vertices'],
index=shape_df.index)
shape_df = pd.concat([shape_df, filolen_df], axis=1)
shape_df['curl'] = (shape_df['major_axis_length'] *
spacing) / shape_df['fiber_length']
total_particles = len(shape_df)
shape_df['channel'] = ['V'] * total_particles
valid_shape_df = shape_df[(shape_df['cv_bg'] < cv_cutoff)
& (shape_df['validity'] == True)]
total_valid_particles = len(valid_shape_df)
#---------------------------------------------------------------------------------------------#
#---------------------------------------------------------------------------------------------#
kparticle_density = round(total_valid_particles / area_sqmm * 0.001, 2)
if pass_num != first_scan:
print("Particle density in {}: {} kp/sq.mm\n".format(
img_name, kparticle_density))
else:
print("Background density in {}: {} kp/sq.mm\n".format(
img_name, kparticle_density))
# shape_df.reset_index(drop=True, inplace=True)
total_shape_df = pd.concat([total_shape_df, shape_df],
axis=0,
sort=False)
# total_shape_df.reset_index(drop=True, inplace=True)
keep_data = [
'label', 'area', 'centroid', 'pass_number', 'max_z_slice',
'eccentricity', 'ellipticity', 'curl', 'circularity', 'validity',
'z_intensity', 'perc_contrast', 'cv_bg', 'sd_above_med_difference',
'fiber_length', 'filo_score', 'roundness_score', 'channel',
'fl_intensity'
]
vdata_dict.update({
'total_valid_particles': total_valid_particles,
'validity': validity
})
with open('{}/{}.vdata.txt'.format(vcount_dir, img_name), 'w') as f:
for k, v in vdata_dict.items():
f.write('{}: {}\n'.format(k, v))
#---------------------------------------------------------------------------------------------#
vgraph.gen_particle_image(pic_to_show,
shape_df,
spot_coords,
pix_per_um=pix_per_um,
show_particles=True,
cv_cutoff=cv_cutoff,
r2_cutoff=0,
scalebar=15,
markers=marker_locs,
exo_toggle=exo_toggle)
savefig('{}/{}.png'.format(img_dir, img_name), dpi=96)
clf()
close('all')
print(
"#******************PNG generated for {}************************#\n\n"
.format(img_name))
#---------------------------------------------------------------------------------------------#
total_shape_df.to_csv('{}/{}.particle_data.csv'.format(
vcount_dir, spot_ID),
columns=keep_data)
greycghg=feature.greycoprops(glcm, prop='homogeneity')/(256*256*9)#size1*5
greycgcl=feature.greycoprops(glcm, prop='correlation')/(256*256*9)
greycgeg=feature.greycoprops(glcm, prop='energy')/(256*256*9)
greycgasm=feature.greycoprops(glcm, prop='ASM')/(256*256*9)
greycgctt=feature.greycoprops(glcm, prop='contrast')/(256*256*9)
lbp=feature.local_binary_pattern(imgrey, 8, np.pi/4)#size同图片
plm=feature.peak_local_max(imgrey, min_distance=1)#多个坐标对
st=feature.structure_tensor(imgrey, sigma=1, mode='constant', cval=0)#三个同图片一样大的矩阵
ste=feature.structure_tensor_eigvals(st[0],st[1],st[2])#两个个同图片一样大的矩阵
# hmf=feature.hessian_matrix(image, sigma=1, mode='constant',
# cval=0, order=None)#6个三通道的原图大小矩阵
hmd=feature.hessian_matrix_det(imgrey, sigma=1)#原图大小矩阵
# hme=feature.hessian_matrix_eigvals(hmf, Hxy=None, Hyy=None)
si=feature.shape_index(imgrey, sigma=1, mode='constant', cval=0)#原图大小矩阵
# ckr=feature.corner_kitchen_rosenfeld(image, mode='constant', cval=0) ##原图大小矩阵 三通道
# ch=feature.corner_harris(imgrey, method='k', k=0.05, eps=1e-06, sigma=1)#原图大小矩阵
# cht=feature.corner_shi_tomasi(imgrey, sigma=1)#原图大小矩阵
# cfs=feature.corner_foerstner(imgrey, sigma=1)#2个 #原图大小矩阵
# csb=feature.corner_subpix(image, ch, window_size=11, alpha=0.99)
cps=feature.corner_peaks(imgrey, min_distance=1, threshold_abs=None,
threshold_rel=0.1, exclude_border=True, indices=True,
footprint=None, labels=None)#一堆坐标值
# cmr=feature.corner_moravec(imgrey, window_size=1)#原图大小矩阵
# cft=feature.corner_fast(imgrey, n=12, threshold=0.15)#原图大小矩阵
corners = feature.corner_peaks(feature.corner_fast(imgrey, 9), min_distance=1)#一堆坐标
corts=feature.corner_orientations(imgrey, corners, octagon(3, 2))#一维矩阵长度不定
# mtem=feature.match_template(image, template, pad_input=False,
# mode='constant', constant_values=0)
# bldg=feature.blob_dog(imgrey, min_sigma=1, max_sigma=50,
def get_shape_index(img):
img = io.imread(img, as_gray=True)
shape_index = feature.shape_index(img)
shape_index_1D = np.ravel(shape_index)
avg_squares_shape_index = np.average(np.square(shape_index_1D))
return avg_squares_shape_index
loc=1.0,
scale=0.5,
size=(cloud_noise_size, cloud_noise_size)
),
image_size / cloud_noise_size
)
np.random.set_state(state)
return ndi.gaussian_filter(image, sigma=2.0)
# First create the test image and its shape index
image = create_test_image()
s = shape_index(image)
# In this example we want to detect 'spherical caps',
# so we threshold the shape index map to
# find points which are 'spherical caps' (~1)
target = 1
delta = 0.05
point_y, point_x = np.where(np.abs(s - target) < delta)
point_z = image[point_y, point_x]
# The shape index map relentlessly produces the shape, even that of noise.
# In order to reduce the impact of noise, we apply a Gaussian filter to it,
# and show the results once in
def get_shape_index(img):
shape_index = feature.shape_index(img)
shape_index_1D = np.ravel(shape_index)
avg_squares_shape_index = np.average(np.square(shape_index_1D))
return avg_squares_shape_index
def Shape_indexing_normalization(img, shape_indexing_sigma=2):
surface = feature.shape_index(img, sigma=shape_indexing_sigma)
surface = np.nan_to_num(surface, copy=True)
surface = (exposure.equalize_hist(surface) * 255).astype(np.uint8)
return (surface)
row, col = np.ogrid[:nrows,:ncols]
width = col - xyr[0]
height = row - xyr[1]
rad = xyr[2] - 50
disk_mask = (width**2 + height**2 > rad**2)
full_mask = disk_mask# + marker_mask
#*********************************************************************************************#
pic_maxmin_masked = np.ma.array(pic_maxmin, mask = full_mask).filled(fill_value = np.nan)
with warnings.catch_warnings():
##RuntimeWarning ignored: invalid values are expected
warnings.simplefilter("ignore")
warnings.warn(RuntimeWarning)
shapedex = shape_index(pic3D_rescale[5])
shapedex = np.ma.array(shapedex,mask = full_mask).filled(fill_value = np.nan)
# if pass_num > 1:
# shapedex= np.ma.array(shapedex,mask = particle_mask).filled(fill_value = -1)
# vimage.gen_img(shapedex)
shapedex_gauss = ndi.gaussian_filter(shapedex, sigma=1)
pix_area = np.count_nonzero(np.invert(np.isnan(shapedex)))
pix_per_um = mag/cam_micron_per_pix
conv_factor = (cam_micron_per_pix / mag)**2
area_sqmm = round((pix_area * conv_factor) * 1e-6, 6)
# def classify_shape(shapedex, pic2D, shape, delta, intensity = 0.55, operator = 'greater'):
image *= ndi.zoom(
np.random.normal(loc=1.0,
scale=0.5,
size=(cloud_noise_size, cloud_noise_size)),
image_size / cloud_noise_size)
np.random.set_state(state)
return ndi.gaussian_filter(image, sigma=2.0)
# First create the test image and its shape index
image = create_test_image()
s = shape_index(image)
# In this example we want to detect 'spherical caps',
# so we threshold the shape index map to
# find points which are 'spherical caps' (~1)
target = 1
delta = 0.05
point_y, point_x = np.where(np.abs(s - target) < delta)
point_z = image[point_y, point_x]
# The shape index map relentlessly produces the shape, even that of noise.
# In order to reduce the impact of noise, we apply a Gaussian filter to it,
# and show the results once in
well = all_wells[well_id]
well_imgs = [f for f in fnames if well in f and 'ch2' not in f]
gamma = 0.5
pad = 5
imgdata = []
for i in range(1,5):
wellpos = [f for f in well_imgs if 'f0' + str(i)+ 'p' in f]
imgseries = load_image_series(path=platedir, imgfiles=wellpos)
imgseries = imgseries.reshape((8, 4, 2160,2160))
mipseries = np.amax(imgseries, axis=0)
hoechst = mipseries[0]
img_th = threshold_img(hoechst**gamma, method='otsu', binary=False)
img_s = shape_index(img_th)
img_enh = nantonum(img_s, pad=-1)
# run blob detection on the shape-index enhanced image
blobs = blob_log(img_enh,
min_sigma=10,
max_sigma=14,
threshold=0.05)
if len(blobs):
bbox = np.stack([np.array([bl[1] - bl[2] - pad,
bl[1] + bl[2] + pad,
bl[0] - bl[2] - pad,
bl[0] + bl[2] + pad]) for bl in blobs])
imax = hoechst.shape[0] - 1
bbox[bbox < 0] = 0
bbox[bbox > imax ] = imax
bbox = bbox.astype(int)
