Merge pull request #44 from computational-cell-analytics/intensity-measurement-updates

Update intensity measurement tool

Author: constantinpape

flamingo_tools/measurements.py

@@ -29,7 +29,7 @@ def _measure_volume_and_surface(mask, resolution):
     return volume, surface
 
 
-def _get_bounding_box(table, seg_id, resolution, shape):
+def _get_bounding_box_and_center(table, seg_id, resolution, shape):
     row = table[table.label_id == seg_id]
 
     bb_min = np.array([
@@ -46,38 +46,113 @@ def _get_bounding_box(table, seg_id, resolution, shape):
         slice(max(bmin - 1, 0), min(bmax + 1, sh))
         for bmin, bmax, sh in zip(bb_min, bb_max, shape)
     )
-    return bb
 
+    center = (
+        int(row.anchor_z.item() / resolution),
+        int(row.anchor_y.item() / resolution),
+        int(row.anchor_x.item() / resolution),
+    )
+
+    return bb, center
+
+
+def _spherical_mask(shape, radius, center=None):
+    if center is None:
+        center = tuple(s // 2 for s in shape)
+    if len(shape) != len(center):
+        raise ValueError("`shape` and `center` must have same length")
+
+    # Build a 1-D open grid for every axis
+    grids = np.ogrid[tuple(slice(0, s) for s in shape)]
+    dist2 = sum((g - c) ** 2 for g, c in zip(grids, center))
+    return (dist2 <= radius ** 2).astype(bool)
 
-def _default_object_features(seg_id, table, image, segmentation, resolution):
-    bb = _get_bounding_box(table, seg_id, resolution, image.shape)
+
+def _normalize_background(measures, image, mask, center, radius, norm, median_only):
+    # Compute the bounding box and get the local image data.
+    bb = tuple(
+        slice(max(0, int(ce - radius)), min(int(ce + radius), sh)) for ce, sh in zip(center, image.shape)
+    )
+    local_image = image[bb]
+
+    # Create a mask with radius around the center.
+    radius_mask = _spherical_mask(local_image.shape, radius)
+
+    # Intersect the radius mask with the foreground mask (if given).
+    if mask is not None:
+        assert mask.shape == image.shape, f"{mask.shape}, {image.shape}"
+        local_mask = mask[bb]
+        radius_mask = np.logical_and(radius_mask, local_mask)
+
+        # For debugging.
+        # import napari
+        # v = napari.Viewer()
+        # v.add_image(local_image)
+        # v.add_labels(local_mask)
+        # v.add_labels(radius_mask)
+        # napari.run()
+
+    # Compute the features over the mask.
+    masked_intensity = local_image[radius_mask]
+
+    # Standardize the measures.
+    bg_measures = {"median": np.median(masked_intensity)}
+    if not median_only:
+        bg_measures = {
+            "mean": np.mean(masked_intensity),
+            "stdev": np.std(masked_intensity),
+            "min": np.min(masked_intensity),
+            "max": np.max(masked_intensity),
+        }
+        for percentile in (5, 10, 25, 75, 90, 95):
+            bg_measures[f"percentile-{percentile}"] = np.percentile(masked_intensity, percentile)
+
+    for measure, val in bg_measures.items():
+        measures[measure] = norm(measures[measure], val)
+
+    return measures
+
+
+def _default_object_features(
+    seg_id, table, image, segmentation, resolution,
+    foreground_mask=None, background_radius=None, norm=np.divide, median_only=False,
+):
+    bb, center = _get_bounding_box_and_center(table, seg_id, resolution, image.shape)
 
     local_image = image[bb]
     mask = segmentation[bb] == seg_id
     assert mask.sum() > 0, f"Segmentation ID {seg_id} is empty."
     masked_intensity = local_image[mask]
 
     # Do the base intensity measurements.
-    measures = {
-        "label_id": seg_id,
-        "mean": np.mean(masked_intensity),
-        "stdev": np.std(masked_intensity),
-        "min": np.min(masked_intensity),
-        "max": np.max(masked_intensity),
-        "median": np.median(masked_intensity),
-    }
-    for percentile in (5, 10, 25, 75, 90, 95):
-        measures[f"percentile-{percentile}"] = np.percentile(masked_intensity, percentile)
+    measures = {"label_id": seg_id, "median": np.median(masked_intensity)}
+    if not median_only:
+        measures.update({
+            "mean": np.mean(masked_intensity),
+            "stdev": np.std(masked_intensity),
+            "min": np.min(masked_intensity),
+            "max": np.max(masked_intensity),
+        })
+        for percentile in (5, 10, 25, 75, 90, 95):
+            measures[f"percentile-{percentile}"] = np.percentile(masked_intensity, percentile)
+
+    if background_radius is not None:
+        # The radius passed is given in micrometer.
+        # The resolution is given in micrometer per pixel.
+        # So we have to divide by the resolution to obtain the radius in pixel.
+        radius_in_pixel = background_radius / resolution
+        measures = _normalize_background(measures, image, foreground_mask, center, radius_in_pixel, norm, median_only)
 
     # Do the volume and surface measurement.
-    volume, surface = _measure_volume_and_surface(mask, resolution)
-    measures["volume"] = volume
-    measures["surface"] = surface
+    if not median_only:
+        volume, surface = _measure_volume_and_surface(mask, resolution)
+        measures["volume"] = volume
+        measures["surface"] = surface
     return measures
 
 
-def _regionprops_features(seg_id, table, image, segmentation, resolution):
-    bb = _get_bounding_box(table, seg_id, resolution, image.shape)
+def _regionprops_features(seg_id, table, image, segmentation, resolution, foreground_mask=None):
+    bb, _ = _get_bounding_box_and_center(table, seg_id, resolution, image.shape)
 
     local_image = image[bb]
     local_segmentation = segmentation[bb]
@@ -106,21 +181,31 @@ def _regionprops_features(seg_id, table, image, segmentation, resolution):
 FEATURE_FUNCTIONS = {
     "default": _default_object_features,
     "skimage": _regionprops_features,
+    "default_background_norm": partial(_default_object_features, background_radius=75, norm=np.divide),
+    "default_background_subtract": partial(_default_object_features, background_radius=75, norm=np.subtract),
 }
 """The different feature functions that are supported in `compute_object_measures` and
 that can be selected via the feature_set argument. Currently this supports:
 - 'default': The default features which compute standard intensity statistics and volume + surface.
 - 'skimage': The scikit image regionprops features.
+- 'default_background_norm': The default features with background normalization.
+- 'default_background_subtract': The default features with background subtraction.
+
+For the background normalized measures, we compute the background intensity in a sphere with radius of 75 micrometer
+around each object.
 """
 
 
+# TODO integrate segmentation post-processing, see `_extend_sgns_simple` in `gfp_annotation.py`
 def compute_object_measures_impl(
     image: np.typing.ArrayLike,
     segmentation: np.typing.ArrayLike,
     n_threads: Optional[int] = None,
     resolution: float = 0.38,
     table: Optional[pd.DataFrame] = None,
     feature_set: str = "default",
+    foreground_mask: Optional[np.typing.ArrayLike] = None,
+    median_only: bool = False,
 ) -> pd.DataFrame:
     """Compute simple intensity and morphology measures for each segmented cell in a segmentation.
 
@@ -133,6 +218,8 @@ def compute_object_measures_impl(
         resolution: The resolution / voxel size of the data.
         table: The segmentation table. Will be computed on the fly if it is not given.
         feature_set: The features to compute for each object. Refer to `FEATURE_FUNCTIONS` for details.
+        foreground_mask: An optional mask indicating the area to use for computing background correction values.
+        median_only: Whether to only compute the median intensity.
 
     Returns:
         The table with per object measurements.
@@ -147,12 +234,19 @@ def compute_object_measures_impl(
         table=table,
         image=image,
         segmentation=segmentation,
-        resolution=resolution
+        resolution=resolution,
+        foreground_mask=foreground_mask,
+        median_only=median_only,
     )
 
     seg_ids = table.label_id.values
     assert len(seg_ids) > 0, "The segmentation table is empty."
+    measure_function(seg_ids[0])
     n_threads = mp.cpu_count() if n_threads is None else n_threads
+
+    # For debugging.
+    # measure_function(seg_ids[0])
+
     with futures.ThreadPoolExecutor(n_threads) as pool:
         measures = list(tqdm(
             pool.map(measure_function, seg_ids), total=len(seg_ids), desc="Compute intensity measures"
@@ -206,14 +300,14 @@ def compute_object_measures(
         table = None
     elif s3_flag:
         seg_table, fs = s3_utils.get_s3_path(segmentation_table_path)
-        with fs.open(seg_table, 'r') as f:
+        with fs.open(seg_table, "r") as f:
             table = pd.read_csv(f, sep="\t")
     else:
         table = pd.read_csv(segmentation_table_path, sep="\t")
 
     # filter table with largest component
     if len(component_list) != 0 and "component_labels" in table.columns:
-        table = table[table['component_labels'].isin(component_list)]
+        table = table[table["component_labels"].isin(component_list)]
 
     # Then, open the volumes.
     image = read_image_data(image_path, image_key)

scripts/intensity_annotation/gfp_annotation.py

@@ -27,7 +27,7 @@ def __init__(self, statistics, default_stat, bins: int = 32, parent=None):
         self.canvas = FigureCanvasQTAgg(self.fig)
 
         # We exclude the label id and the volume / surface measurements.
-        self.stat_names = statistics.columns[1:-2]
+        self.stat_names = statistics.columns[1:-2] if len(statistics.columns) > 2 else statistics.columns[1:]
         self.param_choices = self.stat_names
 
         self.param_box = QComboBox()
@@ -110,7 +110,30 @@ def _extend_sgns_simple(gfp, sgns, dilation):
     return sgns_extended
 
 
-def gfp_annotation(prefix, default_stat="median"):
+def _create_mask(sgns_extended, gfp):
+    from skimage.transform import downscale_local_mean, resize
+
+    gfp_averaged = downscale_local_mean(gfp, factors=(16, 16, 16))
+    # The 35th percentile seems to be a decent approximation for the background subtraction.
+    threshold = np.percentile(gfp_averaged, 35)
+    mask = gfp_averaged > threshold
+    mask = resize(mask, sgns_extended.shape, order=0, anti_aliasing=False, preserve_range=True).astype(bool)
+    mask[sgns_extended != 0] = 0
+
+    # v = napari.Viewer()
+    # v.add_image(gfp)
+    # v.add_image(gfp_averaged, scale=(16, 16, 16))
+    # v.add_labels(mask)
+    # # v.add_labels(mask, scale=(16, 16, 16))
+    # v.add_labels(sgns_extended)
+    # napari.run()
+
+    return mask
+
+
+def gfp_annotation(prefix, default_stat="median", background_norm=None):
+    assert background_norm in (None, "division", "subtraction")
+
     gfp = imageio.imread(f"{prefix}_GFP_resized.tif")
     sgns = imageio.imread(f"{prefix}_SGN_resized_v2.tif")
     pv = imageio.imread(f"{prefix}_PV_resized.tif")
@@ -125,7 +148,16 @@ def gfp_annotation(prefix, default_stat="median"):
     sgns_extended = _extend_sgns_simple(gfp, sgns, dilation=4)
 
     # Compute the intensity statistics.
-    statistics = compute_object_measures_impl(gfp, sgns_extended)
+    if background_norm is None:
+        mask = None
+        feature_set = "default"
+    else:
+        mask = _create_mask(sgns_extended, gfp)
+        assert mask.shape == sgns_extended.shape
+        feature_set = "default_background_norm" if background_norm == "division" else "default_background_subtract"
+    statistics = compute_object_measures_impl(
+        gfp, sgns_extended, feature_set=feature_set, foreground_mask=mask, median_only=True
+    )
 
     # Open the napari viewer.
     v = napari.Viewer()
@@ -135,6 +167,8 @@ def gfp_annotation(prefix, default_stat="median"):
     v.add_image(pv, visible=False, name="PV")
     v.add_labels(sgns, visible=False, name="SGNs")
     v.add_labels(sgns_extended, name="SGNs-extended")
+    if mask is not None:
+        v.add_labels(mask, name="mask-for-background", visible=False)
 
     # Add additional layers for intensity coloring and classification
     # data_numerical = np.zeros(gfp.shape, dtype="float32")
@@ -212,9 +246,10 @@ def threshold_widget(viewer: napari.Viewer, threshold: float = (max_val - min_va
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("prefix")
+    parser.add_argument("-b", "--background_norm")
     args = parser.parse_args()
 
-    gfp_annotation(args.prefix)
+    gfp_annotation(args.prefix, background_norm=args.background_norm)
 
 
 if __name__ == "__main__":