def gen_crypt_masks(exp, input_run_type, run_type):
"""Generate crypt masks based on nuclei density and EdU count
Args:
exp (string): name of experiment
input_run_type (str): cellprofiler pipeline output to use (merged, etc)
run_type (str): name for current process ('crypt_mask')
Returns:
None
"""
# get file paths
nuc_objs_dir = config.get_cp_output_nuc_objs(exp, input_run_type)
for nuc_path in helpers.list_im_files(nuc_objs_dir):
# paths
im_name = config.get_im_name(nuc_path)
csv_path = config.get_extract_csv_nuc_outpath(exp, input_run_type, im_name)
# read in image
nuc_im = cv2.imread(nuc_path, cv2.IMREAD_ANYDEPTH)
# dense nuclei
kn_size = config.get_crypt_finder_dilate_kn_size('crypt_masks')
min_area = config.get_crypt_finder_min_area('crypt_masks')
nuc_dense, label_mask = dense_objects(nuc_im, kernel_size=kn_size, min_area=min_area)
# filter by edu
edu_thresh = expinfo.get_thresh_edu(exp)
min_edu_num = config.get_crypt_finder_min_edu_num(run_type)
im_data = pd.read_csv(csv_path)
edu_pos_objs = thresh_edu_objects(im_data, edu_thresh)
nuc_crypt = filter_crypts(nuc_dense, label_mask, edu_pos_objs, min_edu_num)
# give nuclei in same crypt the same label
nuc_labeled = label_mask.copy()
nuc_labeled[nuc_crypt==0] = 0
# convex hull on crypts
# crypt_mask = chull_groups(nuc_labeled)
# save images
objs_fname = config.get_fpath_crypt_objs(exp, input_run_type, im_name)
cv2.imwrite(objs_fname, nuc_crypt)
label_fname = config.get_fpath_crypt_labels(exp, input_run_type, im_name)
cv2.imwrite(label_fname, nuc_labeled)
def measure_props_no_paneth(exp, run_type, drop=False):
"""Measure properties for all wells in a plate
Args:
exp (str): plate name
run_type (str): analysis output to use for measurements
drop (bool): whether to drop frames
Returns:
None
"""
# measurement names
crypt_measures = ['num_cells', 'num_edu', 'nuc_area', 'eccentricity',
'solidity', 'row', 'col', 'fld']
well_num_measures = ['num_cells', 'num_edu', 'num_crypt_cells', 'num_crypt_edu',
'num_villus_cells', 'num_villus_edu']
well_measures = well_num_measures + ['avg_eccentricity', 'avg_solidity', 'num_crypts']
# output directories
crypt_dir = config.get_cp_output_crypt_measure(exp, run_type)
well_dir = config.get_cp_output_well_measure(exp, run_type)
# create output directories
if not os.path.exists(crypt_dir):
os.makedirs(crypt_dir)
if not os.path.exists(well_dir):
os.makedirs(well_dir)
rows = constants.get_96_rows()
cols = constants.get_96_cols()
if drop:
im_csv_path = config.get_extract_csv_im_drop_inpath(exp, run_type)
# label
drop_label = config.get_csv_drop_label()
# matrices of well-level measurements
well_mats = {k: np.zeros((len(rows), len(cols))) for k in well_measures}
for r, row in enumerate(rows):
for c, col in enumerate(cols):
# store crypt-level measurements
crypt_props = {k: [] for k in crypt_measures}
# store well-level measurements
well_nums = {k: [] for k in well_num_measures}
# iterate over fields
num_flds = expinfo.get_num_fields(exp)
start = expinfo.get_field_start(exp)
for fld in list(range(start, num_flds+start, 1)):
im_name = config.build_im_name(exp, row, col, fld, 'dna')
objs_path = config.get_fpath_crypt_objs(exp, run_type, im_name)
label_path = config.get_fpath_crypt_labels(exp, run_type, im_name)
im_data_path = config.get_extract_csv_nuc_outpath(exp, run_type, im_name)
crypt_objs = cv2.imread(objs_path, cv2.IMREAD_ANYDEPTH)
label_mask = cv2.imread(label_path, cv2.IMREAD_ANYDEPTH)
im_data = pd.read_csv(im_data_path)
edu_thresh = expinfo.get_thresh_edu(exp)
edu_objs = thresh_edu_objects(im_data, edu_thresh)
# crypt-level measurements
crypt_props = measure_crypts_props_no_paneth(crypt_objs, label_mask, edu_objs, crypt_props, row, col, fld)
# well-level measurements
well_props = {
'num_cells': len(im_data),
'num_edu': len(edu_objs),
'num_crypt_cells': len(nonzero_unique(crypt_objs)),
'num_crypt_edu': count_stained_objs(crypt_objs, edu_objs),
'row': row,
'col': col,
'fld': fld
}
well_props.update({
'num_villus_cells': well_props['num_cells'] - well_props['num_crypt_cells'],
'num_villus_edu': well_props['num_edu'] - well_props['num_crypt_edu']
})
for k in list(well_nums.keys()):
well_nums[k].append(well_props[k])
for k in list(well_nums.keys()):
well_nums[k].append(well_props[k])
# save well-level measurements
for k in well_num_measures:
well_mats[k][r][c] = np.sum(well_nums[k])
well_mats['avg_eccentricity'][r][c] = np.mean(crypt_props['eccentricity'])
well_mats['avg_solidity'][r][c] = np.mean(crypt_props['solidity'])
well_mats['num_crypts'][r][c] = len(crypt_props['num_cells'])
# write crypt-level measurements to file
well_name = config.build_well_name(row, col)
out_path = config.get_fpath_crypt_measure(exp, run_type, well_name)
helpers.dict_to_csv(out_path, crypt_props)
# write well-level measurements to file
for k in list(well_mats.keys()):
out_path = config.get_fpath_well_measure(exp, run_type, k)
np.savetxt(out_path, well_mats[k], delimiter=',')
def measure_dispersion(exp, run_type, drop=False):
"""Measure properties for all wells in a plate
Args:
exp (str): plate name
run_type (str): analysis output to use for measurements
drop (bool): whether to drop frames
Returns:
None
"""
# output directories
well_dir = config.get_cp_output_well_measure(exp, run_type)
# create output directories
if not os.path.exists(well_dir):
os.makedirs(well_dir)
rows = constants.get_96_rows()
cols = constants.get_96_cols()
if drop:
im_csv_path = config.get_extract_csv_im_drop_inpath(exp, run_type)
# label
drop_label = config.get_csv_drop_label()
rows = ['A', 'B', 'C', 'D']
cols = [1,2, 5]
# measurement names
well_disp = pd.DataFrame()
for r, row in enumerate(rows):
for c, col in enumerate(cols):
# wellname
well_name = config.build_well_name(row, col)
# store crypt-level measurements
crypt_disp = []
# iterate over fields
num_flds = expinfo.get_num_fields(exp)
start = expinfo.get_field_start(exp)
for fld in list(range(start, num_flds+start, 1)):
im_name = config.build_im_name(exp, row, col, fld, 'dna')
objs_path = config.get_fpath_crypt_objs(exp, run_type, im_name)
label_path = config.get_fpath_crypt_labels(exp, run_type, im_name)
im_data_path = config.get_extract_csv_nuc_outpath(exp, run_type, im_name)
crypt_objs = cv2.imread(objs_path, cv2.IMREAD_ANYDEPTH)
label_mask = cv2.imread(label_path, cv2.IMREAD_ANYDEPTH)
im_data = pd.read_csv(im_data_path)
edu_thresh = expinfo.get_thresh_edu(exp)
edu_objs = thresh_edu_objects(im_data, edu_thresh)
# list of crypt labels
crypt_labels = nonzero_unique(label_mask)
for l in crypt_labels:
# measure properties for one crypt
crypt_mask = get_object(label_mask, l)
objs = mask_objects(crypt_objs, crypt_mask, mask_val=l)
# add properties to dataframe
edu_dist = measure_objs_dispersion(objs, edu_objs)
# add data to dataframe
if edu_dist is not None:
crypt_disp = pd.DataFrame([{'d_nn': d, 'crypt_num': l, 'well': well_name} for d in edu_dist])
well_disp = well_disp.append(crypt_disp)
out_path = config.get_fpath_well_measure(exp, run_type, 'edu_dispersion')
well_disp.to_csv(out_path, index=False)
print(out_path)