def test_generate_props_from_mask(spot_and_mask):
im_spot, mask_spot = spot_and_mask
prop = regionprop.SpotRegionprop(row_idx=2, col_idx=1, label=4)
bbox = [5, 10, 40, 45]
bg_spot = np.zeros_like(im_spot) + 1.
prop.generate_props_from_mask(im_spot, bg_spot, mask_spot, bbox)
assert prop.label == 4
# Stats for dataframe
assert 29 <= prop.spot_dict['centroid_row'] <= 30
assert 34 <= prop.spot_dict['centroid_col'] <= 35
assert prop.spot_dict['bbox_row_min'] == 18
assert prop.spot_dict['bbox_col_min'] == 23
assert prop.spot_dict['bbox_row_max'] == 42
assert prop.spot_dict['bbox_col_max'] == 47
assert prop.spot_dict['intensity_mean'] == np.mean(im_spot[mask_spot > 0])
assert prop.spot_dict['intensity_median'] == np.median(im_spot[mask_spot > 0])
assert prop.spot_dict['bg_mean'] == 1.
assert prop.spot_dict['bg_median'] == 1.
assert prop.spot_dict['od_norm'] == np.log10(
1. / prop.spot_dict['intensity_median'],
)
# ROIs
assert prop.image.all() == im_spot.all()
assert prop.mask.all() == mask_spot.all()
assert prop.masked_image.all() == (im_spot * mask_spot).all()
def test_generate_props_from_disk(spot_and_mask):
im_spot, _ = spot_and_mask
bg_spot = np.zeros_like(im_spot) + 0.5
prop = regionprop.SpotRegionprop(row_idx=0, col_idx=2, label=3)
bbox = [5, 10, 35, 45]
centroid = [25, 26]
prop.generate_props_from_disk(im_spot, bg_spot, bbox, centroid)
assert prop.label == 3
# Stats for dataframe
assert prop.spot_dict['centroid_row'] == centroid[0]
assert prop.spot_dict['centroid_col'] == centroid[1]
assert prop.spot_dict['bbox_row_min'] == bbox[0]
assert prop.spot_dict['bbox_col_min'] == bbox[1]
assert prop.spot_dict['bbox_row_max'] == bbox[2]
assert prop.spot_dict['bbox_col_max'] == bbox[3]
assert prop.spot_dict['intensity_mean'] > 0
assert prop.spot_dict['intensity_median'] > 0.8
assert prop.spot_dict['intensity_median'] < 0.81
assert prop.spot_dict['bg_mean'] == 0.5
assert prop.spot_dict['bg_median'] == 0.5
assert prop.spot_dict['od_norm'] == np.log10(
0.5 / prop.spot_dict['intensity_median'],
)
# ROIs
assert prop.image.all() == im_spot.all()
assert prop.background.all() == bg_spot.all()
mask = prop.make_mask(im_spot.shape[0])
assert prop.mask.all() == mask.all()
assert prop.masked_image.all() == (im_spot * mask).all()
def test_regionprop_init():
prop = regionprop.SpotRegionprop(row_idx=2, col_idx=3, label=5)
assert prop.label == 5
assert prop.spot_dict['grid_row'] == 2
assert prop.spot_dict['grid_col'] == 3
assert prop.image is None
assert prop.background is None
assert prop.mask is None
assert prop.masked_image is None
def test_compute_stats(spot_and_mask):
im_spot, mask_spot = spot_and_mask
prop = regionprop.SpotRegionprop(row_idx=4, col_idx=5, label=1)
prop.image = im_spot
prop.mask = mask_spot
prop.background = np.zeros_like(im_spot) + 0.5
prop.compute_stats()
assert prop.label == 1
assert prop.spot_dict['intensity_median'] < 0.3
assert prop.spot_dict['bg_median'] == 0.5
assert prop.spot_dict['od_norm'] == np.log10(
0.5 / prop.spot_dict['intensity_median'],
)
def get_spot_intensity(coords, im, background, search_range=3):
"""
Extract signal and background intensity at each spot given the spot coordinate
with the following steps:
1. crop image around each grid point
2. Segment 1 single spot from each image
3. Get median intensity, background and OD within the spot mask
4. If segmentation in 2. returns no mask, use a circular mask with average spot size as the spot mask and do 3.
:param coords: list or tuple
[row, col] coordinates of spots
:param im: ndarray
intensity image of the spots (signals)
:param background: ndarray
background image without spots
:param float search_range: Factor of bounding box size in which to search for
spots. E.g. 2 searches 2 * 2 * bbox width * bbox height
:return pd.DataFrame spots_df: Dataframe containing metrics for
all spots in the grid
:return np.array spot_props: A SpotRegionprop object with ROIs for
each spot in the grid
"""
# values in mm
spot_width = constants.params['spot_width']
pix_size = constants.params['pixel_size']
n_rows = constants.params['rows']
n_cols = constants.params['columns']
# make spot size always odd
spot_size = 2 * int(0.3 * spot_width / pix_size) + 1
bbox_width = bbox_height = spot_size
# Strel disk size for spot segmentation
disk_size = int(np.rint(spot_size / 2.5))
# Array of SpotRegionprop objects to hold ROIs
spot_props = txt_parser.create_array(n_rows, n_cols, dtype=object)
# Dataframe to hold spot metrics for the well
spots_df = pd.DataFrame(columns=constants.SPOT_DF_COLS)
row_col_iter = itertools.product(np.arange(n_rows), np.arange(n_cols))
for count, (row_idx, col_idx) in enumerate(row_col_iter):
coord = coords[count, :]
# make bounding boxes larger to account for interpolation errors
spot_height = int(np.round(search_range * bbox_height))
spot_width = int(np.round(search_range * bbox_width))
# Create large bounding box around spot and segment
im_spot_lg, bbox_lg = img_processing.crop_image_at_center(
im=im,
center=coord,
height=spot_height,
width=spot_width,
)
mask_spot = img_processing.thresh_and_binarize(
image=im_spot_lg,
method='bright_spots',
disk_size=disk_size,
thr_percent=75,
get_lcc=True,
)
# Create spot and background instance
spot_prop = regionprop.SpotRegionprop(
row_idx=row_idx,
col_idx=col_idx,
label=count,
)
# Mask spot should cover a certain percentage of ROI
if np.mean(mask_spot) > constants.SPOT_MIN_PERCENT_AREA:
# Mask detected
bg_spot_lg, _ = img_processing.crop_image_at_center(
im=background,
center=coord,
height=spot_height,
width=spot_width,
)
spot_prop.generate_props_from_mask(
image=im_spot_lg,
background=bg_spot_lg,
mask=mask_spot,
bbox=bbox_lg,
)
else:
# Crop around assumed spot size
im_spot, bbox = img_processing.crop_image_at_center(
im,
coord,
bbox_height,
bbox_width,
)
bg_spot, _ = img_processing.crop_image_at_center(
background,
coord,
bbox_height,
bbox_width,
)
spot_prop.generate_props_from_disk(
image=im_spot,
background=bg_spot,
bbox=bbox,
centroid=coord,
)
spots_df = spots_df.append(spot_prop.spot_dict, ignore_index=True)
spot_props[row_idx, col_idx] = spot_prop
return spots_df, spot_props
def test_make_mask():
prop = regionprop.SpotRegionprop(row_idx=1, col_idx=2, label=0)
mask = prop.make_mask(51)
assert mask.shape == (51, 51)
assert mask[0, 0] == 0
assert mask[25, 25] == 1