def contour_pm_watershed(contour_pm,
sigma=2,
h=0,
tissue_mask=None,
padding_mask=None,
min_area=None,
max_area=None):
if tissue_mask is None:
tissue_mask = np.ones_like(contour_pm)
padded = None
if padding_mask is not None and np.any(padding_mask == 0):
contour_pm, padded = crop_with_padding_mask(contour_pm,
padding_mask,
return_mask=True)
tissue_mask = crop_with_padding_mask(tissue_mask, padding_mask)
maxima = peak_local_max(extrema.h_maxima(ndi.gaussian_filter(
np.invert(contour_pm), sigma=sigma),
h=h),
indices=False,
footprint=np.ones((3, 3)))
maxima = label(maxima).astype(np.int32)
# Passing mask into the watershed function will exclude seeds outside
# of the mask, which gives fewer and more accurate segments
maxima = watershed(
contour_pm, maxima, watershed_line=True, mask=tissue_mask) > 0
if min_area is not None and max_area is not None:
maxima = label(maxima, connectivity=1).astype(np.int32)
areas = np.bincount(maxima.ravel())
size_passed = np.arange(areas.size)[np.logical_and(
areas > min_area, areas < max_area)]
maxima *= np.isin(maxima, size_passed)
np.greater(maxima, 0, out=maxima)
if padded is None:
return maxima.astype(np.bool)
else:
padded[padded == 1] = maxima.flatten()
return padded.astype(np.bool)
def locate_with_mask(img, padding_mask=None, min_peak_int=None):
if padding_mask is not None and np.any(padding_mask == 0):
img = rowit.crop_with_padding_mask(img, padding_mask)
if min_peak_int is None:
p = 95
else:
p = percentileofscore(img.flatten(), min_peak_int)
with warnings.catch_warnings():
warnings.filterwarnings('ignore',
message='Image contains no local maxima.')
warnings.filterwarnings(
'ignore', message='All local maxima were in the margins.')
spot_table = locate(img,
3,
preprocess=False,
percentile=p,
separation=2)
return spot_table