def run(
self, stack: ImageStack,
) -> Tuple[IntensityTable, ConnectedComponentDecodingResult]:
"""decode pixels and combine them into spots using connected component labeling
Parameters
----------
stack : ImageStack
ImageStack containing spots
Returns
-------
IntensityTable :
IntensityTable containing decoded spots
ConnectedComponentDecodingResult :
Results of connected component labeling
"""
pixel_intensities = IntensityTable.from_image_stack(
stack, crop_x=self.crop_x, crop_y=self.crop_y, crop_z=self.crop_z)
decoded_intensities = self.codebook.metric_decode(
pixel_intensities,
max_distance=self.distance_threshold,
min_intensity=self.magnitude_threshold,
norm_order=self.norm_order,
metric=self.metric
)
caf = CombineAdjacentFeatures(
min_area=self.min_area,
max_area=self.max_area,
mask_filtered_features=True
)
decoded_spots, image_decoding_results = caf.run(intensities=decoded_intensities)
return decoded_spots, image_decoding_results
def compute_magnitudes(stack, norm_order=2):
pixel_intensities = IntensityTable.from_image_stack(zero_norm_stack)
feature_traces = pixel_intensities.stack(traces=(Indices.CH.value,
Indices.ROUND.value))
norm = np.linalg.norm(feature_traces.values, ord=norm_order, axis=1)
return norm
def test_run_pipline(self):
tempdir = exec.stages(self.stages, self.subdirs, keep_data=True)
intensities = IntensityTable.load(
os.path.join(tempdir, "results", self.spots_file))
self.verify_results(intensities)
if os.getenv("TEST_KEEP_DATA") is None:
shutil.rmtree(tempdir)
def intensities(self, codebook=None) -> IntensityTable:
if codebook is None:
codebook = self.codebook()
intensities = IntensityTable.synthetic_intensities(
codebook, self.n_z, self.height, self.width, self.n_spots,
self.mean_fluor_per_spot, self.mean_photons_per_fluor)
assert intensities.dtype == np.float32 and intensities.max() <= 1
return intensities
def test_synthetic_intensities_generates_correct_number_of_features(
loaded_codebook):
n_spots = 2
intensities = IntensityTable.synthetic_intensities(loaded_codebook,
n_spots=n_spots)
assert isinstance(intensities, IntensityTable)
# shape should have n_spots and channels and hybridization rounds equal to the codebook's shape
assert intensities.shape == (n_spots, *loaded_codebook.shape[1:])
def test_empty_intensity_table():
x = [1, 2]
y = [2, 3]
z = [1, 1]
r = [1, 1]
spot_attributes = pd.MultiIndex.from_arrays([x, y, z, r],
names=('x', 'y', 'z', 'r'))
empty = IntensityTable.empty_intensity_table(spot_attributes, 2, 2)
assert empty.shape == (2, 2, 2)
assert np.sum(empty.values) == 0
def test_synthetic_intensities_have_correct_number_of_on_features(
loaded_codebook):
n_spots = 2
intensities = IntensityTable.synthetic_intensities(loaded_codebook,
n_spots=n_spots)
on_features = np.sum(intensities.values != 0)
# this asserts that the number of features "on" in intensities should be equal to the
# number of "on" features in the codewords, times the total number of spots in intensities.
assert on_features == loaded_codebook.sum(
(Indices.CH, Indices.HYB)).values.mean() * n_spots
def test_reshaping_between_stack_and_intensities():
"""
transform an pixels of an ImageStack into an IntensityTable and back again, then verify that
the created Imagestack is the same as the original
"""
np.random.seed(777)
image = ImageStack.from_numpy_array(np.random.rand(1, 2, 3, 4, 5).astype(np.float32))
pixel_intensities = IntensityTable.from_image_stack(image, 0, 0, 0)
image_shape = (image.shape['z'], image.shape['y'], image.shape['x'])
image_from_pixels = pixel_intensities_to_imagestack(pixel_intensities, image_shape)
assert np.array_equal(image.numpy_array, image_from_pixels.numpy_array)
def measure_spot_intensities(
data_image: ImageStack,
spot_attributes: SpotAttributes,
measurement_function: Callable[[Sequence], Number],
radius_is_gyration: bool = False,
) -> IntensityTable:
"""given spots found from a reference image, find those spots across a data_image
Parameters
----------
data_image : ImageStack
ImageStack containing multiple volumes for which spots' intensities must be calculated
spot_attributes : pd.Dataframe
Locations and radii of spots
measurement_function : Callable[[Sequence], Number])
Function to apply over the spot volumes to identify the intensity (e.g. max, mean, ...)
radius_is_gyration : bool
if True, indicates that the radius corresponds to radius of gyration, which is a function of
spot intensity, but typically is a smaller unit than the sigma generated by blob_log.
In this case, the spot's bounding box is rounded up instead of down when measuring
intensity. (default False)
Returns
-------
IntensityTable :
3d tensor of (spot, channel, round) information for each coded spot
"""
# determine the shape of the intensity table
n_ch = data_image.shape[Indices.CH]
n_round = data_image.shape[Indices.ROUND]
# construct the empty intensity table
intensity_table = IntensityTable.empty_intensity_table(
spot_attributes=spot_attributes,
n_ch=n_ch,
n_round=n_round,
)
# fill the intensity table
indices = product(range(n_ch), range(n_round))
for c, r in indices:
image, _ = data_image.get_slice({Indices.CH: c, Indices.ROUND: r})
blob_intensities: pd.Series = measure_spot_intensity(
image,
spot_attributes,
measurement_function,
radius_is_gyration=radius_is_gyration)
intensity_table[:, c, r] = blob_intensities
return intensity_table
def small_intensity_table():
intensities = np.array([[[0, 1], [1, 0]], [[1, 0], [0, 1]], [[0, 0],
[1, 1]]])
spot_attributes = dataframe_to_multiindex(
pd.DataFrame(
data={
IntensityTable.SpotAttributes.X: [0, 1, 2],
IntensityTable.SpotAttributes.Y: [3, 4, 5],
IntensityTable.SpotAttributes.Z: [0, 0, 0],
IntensityTable.SpotAttributes.RADIUS: [0.1, 0.2, 0.3]
}))
return IntensityTable.from_spot_data(intensities, spot_attributes)
def _cli(cls, args, print_help=False):
"""Runs the decoder component based on parsed arguments."""
if args.decoder_algorithm_class is None or print_help:
cls.decoder_group.print_help()
cls.decoder_group.exit(status=2)
instance = args.decoder_algorithm_class(**vars(args))
# load intensities and codebook
intensities = IntensityTable.load(args.input)
codebook = Codebook.from_json(args.codebook)
# decode and save output
intensities = instance.decode(intensities, codebook)
intensities.save(args.output)
def _measure_spot_intensities(
self, stack: ImageStack,
spot_attributes: pd.DataFrame) -> IntensityTable:
n_ch = stack.shape[Indices.CH]
n_hyb = stack.shape[Indices.HYB]
spot_attribute_index = dataframe_to_multiindex(spot_attributes)
intensity_table = IntensityTable.empty_intensity_table(
spot_attribute_index, n_ch, n_hyb)
indices = product(range(n_ch), range(n_hyb))
for c, h in indices:
image, _ = stack.get_slice({Indices.CH: c, Indices.HYB: h})
blob_intensities: pd.Series = self._measure_blob_intensity(
image, spot_attributes, self.measurement_function)
intensity_table[:, c, h] = blob_intensities
return intensity_table
def test_run_pipeline(self):
tempdir = tempfile.mkdtemp()
coverage_enabled = "STARFISH_COVERAGE" in os.environ
def callback(interval):
print(" ".join(stage[:2]), " ==> {} seconds".format(interval))
try:
for subdir in TestWithIssData.SUBDIRS:
os.makedirs(
"{tempdir}".format(tempdir=os.path.join(tempdir, subdir)))
for stage in TestWithIssData.STAGES:
cmdline = [
element(tempdir=tempdir) if callable(element) else element
for element in stage
]
if cmdline[0] == "starfish" and coverage_enabled:
coverage_cmdline = [
"coverage",
"run",
"-p",
"--source",
"starfish",
"-m",
"starfish",
]
coverage_cmdline.extend(cmdline[1:])
cmdline = coverage_cmdline
with clock.timeit(callback):
subprocess.check_call(cmdline)
intensities = IntensityTable.load(
os.path.join(tempdir, "results", "spots.nc"))
genes, counts = np.unique(
intensities.coords[Codebook.Constants.GENE.value],
return_counts=True)
gene_counts = pd.Series(counts, genes)
assert gene_counts['ACTB_human'] > gene_counts['ACTB_mouse']
finally:
if os.getenv("TEST_ISS_KEEP_DATA") is None:
shutil.rmtree(tempdir)
def _cli(cls, args, print_help=False):
"""Runs the gene_assignment component based on parsed arguments."""
from starfish import munge
if args.gene_assignment_algorithm_class is None or print_help:
cls.gene_assignment_group.print_help()
cls.gene_assignment_group.exit(status=2)
with open(args.coordinates_geojson, "r") as fh:
coordinates = json.load(fh)
regions = munge.geojson_to_region(coordinates)
print('Assigning genes to cells...')
intensity_table = IntensityTable.load(args.intensities)
instance = args.gene_assignment_algorithm_class(**vars(args))
result = instance.assign_genes(intensity_table, regions)
print("Writing | cell_id | spot_id to: {}".format(args.output))
result.to_json(args.output, orient="records")
def concatenate_spot_attributes_to_intensities(
spot_attributes: Sequence[Tuple[SpotAttributes, Dict[Indices, int]]]
) -> IntensityTable:
"""
Merge multiple spot attributes frames into a single IntensityTable without merging across
channels and imaging rounds
Parameters
----------
spot_attributes : Sequence[Tuple[SpotAttributes, Dict[Indices, int]]]
A sequence of SpotAttribute objects and the Indices (channel, round) that each object is
associated with.
Returns
-------
IntensityTable :
concatenated input SpotAttributes, converted to an IntensityTable object
"""
n_ch: int = max(inds[Indices.CH] for _, inds in spot_attributes) + 1
n_round: int = max(inds[Indices.ROUND] for _, inds in spot_attributes) + 1
all_spots = pd.concat([sa.data for sa, inds in spot_attributes])
# this drop call ensures only x, y, z, radius, and quality, are passed to the IntensityTable
features_coordinates = all_spots.drop(['spot_id', 'intensity'], axis=1)
intensity_table = IntensityTable.empty_intensity_table(
SpotAttributes(features_coordinates),
n_ch,
n_round,
)
i = 0
for attrs, inds in spot_attributes:
for _, row in attrs.data.iterrows():
intensity_table[i, inds[Indices.CH],
inds[Indices.ROUND]] = row['intensity']
i += 1
return intensity_table
def small_intensity_table():
intensities = np.array([
[[0, 1], [1, 0]],
[[1, 0], [0, 1]],
[[0, 0], [1, 1]],
[
[0.5, 0.5], # this one should fail decoding
[0.5, 0.5]
],
[[0.1, 0], [0,
0.1]], # this one is a candidate for intensity filtering
])
spot_attributes = pd.DataFrame(
data={
Indices.X.value: [0, 1, 2, 3, 4],
Indices.Y.value: [3, 4, 5, 6, 7],
Indices.Z.value: [0, 0, 0, 0, 0],
Features.SPOT_RADIUS: [0.1, 2, 3, 2, 1]
})
return IntensityTable.from_spot_data(intensities, spot_attributes)
def _calculate_mean_pixel_traces(
label_image: np.ndarray,
intensities: IntensityTable,
) -> IntensityTable:
"""
For all pixels that contribute to a connected component, calculate the mean value for
each (ch, round), producing an average "trace" of a feature across the imaging experiment
Parameters
----------
label_image : np.ndarray
An image where all pixels of a connected component share the same integer ID
intensities : IntensityTable
decoded intensities
Returns
-------
IntensityTable :
an IntensityTable where the number of features equals the number of connected components
and the intensities of each each feature is its mean trace.
"""
pixel_labels = label_image.reshape(-1)
intensities['spot_id'] = (Features.AXIS, pixel_labels)
mean_pixel_traces = intensities.groupby('spot_id').mean(Features.AXIS)
mean_distances = intensities[Features.DISTANCE].groupby(
'spot_id').mean(Features.AXIS)
mean_pixel_traces[Features.DISTANCE] = ('spot_id',
np.ravel(mean_distances))
# the 0th pixel trace corresponds to background. If present, drop it.
try:
mean_pixel_traces = mean_pixel_traces.drop(0, dim='spot_id')
except KeyError:
pass
return mean_pixel_traces
def metric_decode(self,
intensities: IntensityTable,
max_distance: Number,
min_intensity: Number,
norm_order: int,
metric: str = 'euclidean') -> IntensityTable:
"""Assign the closest target by euclidean distance to each feature in an intensity table
Normalizes both the codes and the features to be unit vectors and finds the closest code
for each feature
Parameters
----------
intensities : IntensityTable
features to be decoded
max_distance : Number
maximum distance between a feature and its closest code for which the coded target will
be assigned.
min_intensity : Number
minimum intensity for a feature to receive a target annotation
norm_order : int
the scipy.linalg norm to apply to normalize codes and intensities
metric : str
the sklearn metric string to pass to NearestNeighbors
See Also
--------
The available norms for this function can be found at the following link:
https://docs.scipy.org/doc/numpy-1.14.0/reference/generated/numpy.linalg.norm.html
Returns
-------
IntensityTable :
Intensity table containing normalized intensities, target assignments, distances to
the nearest code, and the filtering status of each feature.
"""
self._validate_decode_intensity_input_matches_codebook_shape(
intensities)
# normalize both the intensities and the codebook
norm_intensities, norms = self._normalize_features(
intensities, norm_order=norm_order)
norm_codes, _ = self._normalize_features(self, norm_order=norm_order)
metric_outputs, targets = self._approximate_nearest_code(
norm_codes, norm_intensities, metric=metric)
# only targets with low distances and high intensities should be retained
passes_filters = np.logical_and(norms >= min_intensity,
metric_outputs <= max_distance,
dtype=np.bool)
# set targets, distances, and filtering results
norm_intensities[Features.TARGET] = (Features.AXIS, targets)
norm_intensities[Features.DISTANCE] = (Features.AXIS, metric_outputs)
norm_intensities[Features.PASSES_THRESHOLDS] = (Features.AXIS,
passes_filters)
# norm_intensities is a DataArray, make it back into an IntensityTable
return IntensityTable(norm_intensities)
def decode_per_round_max(self,
intensities: IntensityTable) -> IntensityTable:
"""decode each feature by selecting the per-imaging-round max-valued channel
Notes
-----
- If no code matches the per-round maximum for a feature, it will be assigned 'nan' instead
of a target value
- Numpy's argmax breaks ties by picking the first channel -- this can lead to
unexpected results where some features with "tied" channels will decode, but others will
be assigned 'nan'.
Parameters
----------
intensities : IntensityTable
features to be decoded
Returns
-------
IntensityTable :
intensity table containing additional data variables for target assignments
"""
def _view_row_as_element(array: np.ndarray) -> np.ndarray:
"""view an entire code as a single element
This view allows vectors (codes) to be compared for equality without need for multiple
comparisons by casting the data in each code to a structured dtype that registers as
a single value
Parameters
----------
array : np.ndarray
2-dimensional numpy array of shape (n_observations, (n_ch * n_round)) where
observations may be either features or codes.
Returns
-------
np.ndarray :
1-dimensional vector of shape n_observations
"""
nrows, ncols = array.shape
dtype = {
'names': ['f{}'.format(i) for i in range(ncols)],
'formats': ncols * [array.dtype]
}
return array.view(dtype)
self._validate_decode_intensity_input_matches_codebook_shape(
intensities)
max_channels = intensities.argmax(Indices.CH.value)
codes = self.argmax(Indices.CH.value)
# TODO ambrosejcarr, dganguli: explore this quality score further
# calculate distance scores by evaluating the fraction of signal in each round that is
# found in the non-maximal channels.
max_intensities = intensities.max(Indices.CH.value)
round_intensities = intensities.sum(Indices.CH.value)
distance = 1 - (max_intensities / round_intensities).mean(
Indices.ROUND.value)
a = _view_row_as_element(codes.values.reshape(self.shape[0], -1))
b = _view_row_as_element(
max_channels.values.reshape(intensities.shape[0], -1))
targets = np.full(intensities.shape[0],
fill_value=np.nan,
dtype=object)
# decode the intensities
for i in np.arange(codes.shape[0]):
targets[np.where(a[i] == b)[0]] = codes[Features.TARGET][i]
# a code passes filters if it decodes successfully
passes_filters = ~pd.isnull(targets)
intensities[Features.TARGET] = (Features.AXIS, targets.astype('U'))
intensities[Features.DISTANCE] = (Features.AXIS, distance)
intensities[Features.PASSES_THRESHOLDS] = (Features.AXIS,
passes_filters)
return intensities
def intensities(self, codebook=None) -> IntensityTable:
if codebook is None:
codebook = self.codebook()
return IntensityTable.synthetic_intensities(
codebook, self.n_z, self.height, self.width, self.n_spots,
self.mean_fluor_per_spot, self.mean_photons_per_fluor)
def test_imagestack_to_intensity_table():
codebook, intensity_table, image = codebook_intensities_image_for_single_synthetic_spot()
pixel_intensities = IntensityTable.from_image_stack(image)
pixel_intensities = codebook.metric_decode(
pixel_intensities, max_distance=0, min_intensity=1000, norm_order=2)
assert isinstance(pixel_intensities, IntensityTable)
def decode_per_hyb_max(self,
intensities: IntensityTable) -> IntensityTable:
"""decode each feature by selecting the per-hybridization round max-valued channel
Notes
-----
If no code matches the per-channel max of a feature, it will be assigned np.nan instead
of a gene value
Parameters
----------
intensities : IntensityTable
features to be decoded
Returns
-------
IntensityTable :
intensity table containing additional data variables for gene assignments
"""
def _view_row_as_element(array: np.ndarray) -> np.ndarray:
"""view an entire code as a single element
This view allows vectors (codes) to be compared for equality without need for multiple
comparisons by casting the data in each code to a structured dtype that registers as
a single value
Parameters
----------
array : np.ndarray
2-dimensional numpy array of shape (n_observations, (n_ch * n_hyb)) where
observations may be either features or codes.
Returns
-------
np.ndarray :
1-dimensional vector of shape n_observations
"""
nrows, ncols = array.shape
dtype = {
'names': ['f{}'.format(i) for i in range(ncols)],
'formats': ncols * [array.dtype]
}
return array.view(dtype)
max_channels = intensities.argmax(Indices.CH.value)
codes = self.argmax(Indices.CH.value)
a = _view_row_as_element(codes.values.reshape(self.shape[0], -1))
b = _view_row_as_element(
max_channels.values.reshape(intensities.shape[0], -1))
genes = np.empty(intensities.shape[0], dtype=object)
genes.fill(np.nan)
for i in np.arange(a.shape[0]):
genes[np.where(a[i] == b)[0]] = codes['gene_name'][i]
gene_index = pd.Index(genes.astype('U'))
intensities[IntensityTable.Constants.GENE.value] = (
IntensityTable.Constants.FEATURES.value, gene_index)
return intensities
def decode_euclidean(self, intensities: IntensityTable) -> IntensityTable:
"""Assign the closest gene by euclidean distance to each feature in an intensity table
Parameters
----------
intensities : IntensityTable
features to be decoded
Returns
-------
IntensityTable :
intensity table containing additional data variables for gene assignments and feature
qualities
"""
def _min_euclidean_distance(observation: xr.DataArray,
codes: Codebook) -> np.ndarray:
"""find the code with the closest euclidean distance to observation
Parameters
----------
observation : xr.DataArray
2-dimensional DataArray of shape (n_ch, n_hyb)
codes :
Codebook containing codes to compare to observation
Returns
-------
np.ndarray :
1-d vector containing the distance of each code to observation
"""
squared_diff = (codes - observation)**2
code_distances = np.sqrt(
squared_diff.sum((Indices.CH, Indices.HYB)))
# order of codes changes here (automated sorting on the reshaping?)
return code_distances
# normalize both the intensities and the codebook
norm_intensities = intensities.groupby(
IntensityTable.Constants.FEATURES.value).apply(
lambda x: x / x.sum())
norm_codes = self.groupby(
Codebook.Constants.GENE.value).apply(lambda x: x / x.sum())
# calculate pairwise euclidean distance between codes and features
func = functools.partial(_min_euclidean_distance, codes=norm_codes)
distances = norm_intensities.groupby(
IntensityTable.Constants.FEATURES.value).apply(func)
# calculate quality of each decoded spot
qualities = 1 - distances.min(Codebook.Constants.GENE.value)
qualities_index = pd.Index(qualities)
# identify genes associated with closest codes
closest_code_index = distances.argmin(Codebook.Constants.GENE.value)
gene_ids = distances.indexes[Codebook.Constants.GENE.value].values[
closest_code_index.values]
gene_index = pd.Index(gene_ids)
# set new values on the intensity table in-place
intensities[IntensityTable.Constants.GENE.value] = (
IntensityTable.Constants.FEATURES.value, gene_index)
intensities[IntensityTable.Constants.QUALITY.value] = (
IntensityTable.Constants.FEATURES.value, qualities_index)
return intensities
##### visualization of rolonies
# EPY: END markdown
# EPY: START code
distance_threshold = min_dist
psd = SpotFinder.PixelSpotDetector(codebook=exp.codebook,
metric='euclidean',
distance_threshold=distance_threshold,
magnitude_threshold=magnitude_threshold,
min_area=area_threshold[0],
max_area=area_threshold[1])
spot_intensities, results = psd.run(zero_norm_stack)
spot_intensities = IntensityTable(
spot_intensities.where(spot_intensities[Features.PASSES_THRESHOLDS],
drop=True))
# EPY: END code
# EPY: START code
# exclude spots that don't meet our area thresholds
area_lookup = lambda x: 0 if x == 0 else results.region_properties[x - 1].area
vfunc = np.vectorize(area_lookup)
mask = np.squeeze(vfunc(results.label_image))
new_image = np.squeeze(results.decoded_image) * (mask > area_threshold[0]) * (
mask < area_threshold[1])
plt.figure(figsize=(10, 10))
plt.imshow(new_image, cmap='nipy_spectral')
plt.axis('off')
plt.title('Coded rolonies')
def run(
self, intensities: IntensityTable
) -> Tuple[IntensityTable, ConnectedComponentDecodingResult]:
"""
Execute the combine_adjacent_features method on an IntensityTable containing pixel
intensities
Parameters
----------
intensities : IntensityTable
Pixel intensities of an imaging experiment
Returns
-------
IntensityTable :
Table whose features comprise sets of adjacent pixels that decoded to the same target
ConnectedComponentDecodingResult :
NamedTuple containing :
region_properties :
the properties of each connected component, in the same order as the
IntensityTable
label_image : np.ndarray
An image where all pixels of a connected component share the same integer ID
decoded_image : np.ndarray
Image whose pixels correspond to the targets that the given position in the
ImageStack decodes to.
"""
# map target molecules to integers so they can be reshaped into an image that can
# be subjected to a connected-component algorithm to find adjacent pixels with the
# same targets
targets = intensities[Features.TARGET].values
target_map = TargetsMap(targets)
# create the decoded_image
decoded_image = self._intensities_to_decoded_image(
intensities,
target_map,
self._mask_filtered,
)
# label the decoded image to extract connected component features
label_image: np.ndarray = label(decoded_image,
connectivity=self._connectivity)
# calculate properties of each feature
props: List = regionprops(np.squeeze(label_image))
# calculate mean intensities across the pixels of each feature
mean_pixel_traces = self._calculate_mean_pixel_traces(
label_image,
intensities,
)
# Create SpotAttributes and determine feature filtering outcomes
spot_attributes, passes_filter = self._create_spot_attributes(
props,
decoded_image,
target_map,
)
# augment the SpotAttributes with filtering results and distances from nearest codes
spot_attributes.data[Features.DISTANCE] = mean_pixel_traces[
Features.DISTANCE]
spot_attributes.data[Features.PASSES_THRESHOLDS] = passes_filter
# create new indexes for the output IntensityTable
channel_index = mean_pixel_traces.indexes[Indices.CH]
round_index = mean_pixel_traces.indexes[Indices.ROUND]
coords = IntensityTable._build_xarray_coords(spot_attributes,
channel_index,
round_index)
# create the output IntensityTable
dims = (Features.AXIS, Indices.CH.value, Indices.ROUND.value)
intensity_table = IntensityTable(data=mean_pixel_traces,
coords=coords,
dims=dims)
# combine the various non-IntensityTable results into a NamedTuple before returning
ccdr = ConnectedComponentDecodingResult(props, label_image,
decoded_image)
return intensity_table, ccdr
def synthetic_intensity_table(loaded_codebook) -> IntensityTable:
return IntensityTable.synthetic_intensities(loaded_codebook, n_spots=2)
def test_imagestack_to_intensity_table_no_noise(synthetic_spot_pass_through_stack):
codebook, intensity_table, image = synthetic_spot_pass_through_stack
pixel_intensities = IntensityTable.from_image_stack(image)
pixel_intensities = codebook.metric_decode(
pixel_intensities, max_distance=0, min_intensity=1000, norm_order=2)
assert isinstance(pixel_intensities, IntensityTable)
def synthetic_spots(
cls,
intensities: IntensityTable,
num_z: int,
height: int,
width: int,
n_photons_background=1000,
point_spread_function=(4, 2, 2),
camera_detection_efficiency=0.25,
background_electrons=1,
graylevel: float = 37000.0 / 2**16,
ad_conversion_bits=16,
) -> "ImageStack":
"""Generate a synthetic ImageStack from a set of Features stored in an IntensityTable
Parameters
----------
intensities : IntensityTable
IntensityTable containing coordinates of fluorophores. Used to position and generate
spots in the output ImageStack
num_z : int
Number of z-planes in the ImageStack
height : int
Height in pixels of the ImageStack
width : int
Width in pixels of the ImageStack
n_photons_background : int
Poisson rate for the number of background photons to add to each pixel of the image.
Set this parameter to 0 to eliminate background.
(default 1000)
point_spread_function : Tuple[int]
The width of the gaussian density wherein photons spread around their light source.
Set to zero to eliminate this (default (4, 2, 2))
camera_detection_efficiency : float
The efficiency of the camera to detect light. Set to 1 to remove this filter (default
0.25)
background_electrons : int
Poisson rate for the number of spurious electrons detected per pixel during image
capture by the camera (default 1)
graylevel : float
The number of shades of gray displayable by the synthetic camera. Larger numbers will
produce higher resolution images (default 37000 / 2 ** 16)
ad_conversion_bits : int
The number of bits used during analog to visual conversion (default 16)
Returns
-------
ImageStack :
synthetic spots
"""
# check some params
if not 0 < camera_detection_efficiency <= 1:
raise ValueError(
f'invalid camera_detection_efficiency value: {camera_detection_efficiency}. '
f'Must be in the interval (0, 1].')
def select_uint_dtype(array):
"""choose appropriate dtype based on values of an array"""
max_val = np.max(array)
for dtype in (np.uint8, np.uint16, np.uint32):
if max_val <= np.iinfo(dtype).max:
return array.astype(dtype)
raise ValueError(
'value exceeds dynamic range of largest skimage-supported type'
)
# make sure requested dimensions are large enough to support intensity values
indices = zip((Indices.Z.value, Indices.Y.value, Indices.X.value),
(num_z, height, width))
for index, requested_size in indices:
required_size = intensities.coords[index].values.max()
if required_size > requested_size:
raise ValueError(
f'locations of intensities contained in table exceed the size of requested '
f'dimension {index}. Required size {required_size} > {requested_size}.'
)
# create an empty array of the correct size
image = np.zeros(
(intensities.sizes[Indices.ROUND.value],
intensities.sizes[Indices.CH.value], num_z, height, width),
dtype=np.uint32)
# starfish uses float images, but the logic here requires uint. We cast, and will cast back
# at the end of the function
intensities.values = img_as_uint(intensities)
for ch, round_ in product(*(range(s) for s in intensities.shape[1:])):
spots = intensities[:, ch, round_]
# numpy deprecated casting a specific way of casting floats that is triggered in xarray
with warnings.catch_warnings():
warnings.simplefilter('ignore', FutureWarning)
values = spots.where(spots, drop=True)
image[round_, ch, values.z, values.y, values.x] = values
intensities.values = img_as_float32(intensities)
# add imaging noise
image += np.random.poisson(n_photons_background,
size=image.shape).astype(np.uint32)
# blur image over coordinates, but not over round_/channels (dim 0, 1)
sigma = (0, 0) + point_spread_function
image = gaussian_filter(image, sigma=sigma, mode='nearest')
image = image * camera_detection_efficiency
image += np.random.normal(scale=background_electrons, size=image.shape)
# mimic analog to digital conversion
image = (image / graylevel).astype(int).clip(0, 2**ad_conversion_bits)
# clip in case we've picked up some negative values
image = np.clip(image, 0, a_max=None)
# set the smallest int datatype that supports the data's intensity range
image = select_uint_dtype(image)
# convert to float for ImageStack
with warnings.catch_warnings():
# possible precision loss when casting from uint to float is acceptable
warnings.simplefilter('ignore', UserWarning)
image = img_as_float32(image)
return cls.from_numpy_array(image)