Image segmentation and extract segmentation features

[Nasrine26]

... Hello, I had a question concerning image segmentation and feature extraction. I am using a Visium image with H&E stains. When I follow your tutorial with your image, I get no errors but when I use mine I get the following error:

ValueError: Expected height to be in interval [0, 2000], found 9761.

My image has the following size:

ImageContainer object with 3 layers:

image: y (2000), x (1935), z (1), channels (3)

image_smooth: y (2000), x (1935), z (1), channels (3)

segmented_watershed: y (2000), x (1935), z (1), channels:0 (1)

Many thanks!

[giovp]

Hi @Nasrine26 , thank you for the interest in squidpy! from the log you posted it seems like the segmentation was already performed (looking at the segmented_watershed layer). Would you mind sharing at what point of the tutorial it fails as well as te tracebak? thank you!

[Nasrine26]

Yes segmentation is performed but then when I try to extract the features with:

sq.im.calculate_image_features(
    adata_vis,
    my_image,
    layer="image",
    features="segmentation",
    key_added="segmentation_features",
    features_kwargs={
        "segmentation": {
            "label_layer": "segmented_watershed",
            "props": ["label", "area", "mean_intensity"],
            "channels": [1, 2],
        }
    },
    mask_circle=True,
)

following this [tutorial](https://squidpy.readthedocs.io/en/stable/auto_examples/image/compute_segmentation_features.html, I get the following error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-13-090ef3048512> in <module>
----> 1 sq.im.calculate_image_features(
      2     adata_vis,
      3     my_image,
      4     layer="image",
      5     features="segmentation",

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/squidpy/im/_feature.py in calculate_image_features(adata, img, layer, library_id, features, features_kwargs, key_added, copy, n_jobs, backend, show_progress_bar, **kwargs)
     89     start = logg.info(f"Calculating features `{list(features)}` using `{n_jobs}` core(s)")
     90 
---> 91     res = parallelize(
     92         _calculate_image_features_helper,
     93         collection=adata.obs_names,

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/squidpy/_utils.py in wrapper(*args, **kwargs)
    166             pbar, queue, thread = None, None, None  # type: ignore[assignment]
    167 
--> 168         res = jl.Parallel(n_jobs=n_jobs, backend=backend)(
    169             jl.delayed(runner if use_runner else callback)(
    170                 *((i, cs) if use_ixs else (cs,)),

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/parallel.py in __call__(self, iterable)
   1039             # remaining jobs.
   1040             self._iterating = False
-> 1041             if self.dispatch_one_batch(iterator):
   1042                 self._iterating = self._original_iterator is not None
   1043 

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/parallel.py in dispatch_one_batch(self, iterator)
    857                 return False
    858             else:
--> 859                 self._dispatch(tasks)
    860                 return True
    861 

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/parallel.py in _dispatch(self, batch)
    775         with self._lock:
    776             job_idx = len(self._jobs)
--> 777             job = self._backend.apply_async(batch, callback=cb)
    778             # A job can complete so quickly than its callback is
    779             # called before we get here, causing self._jobs to

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/_parallel_backends.py in apply_async(self, func, callback)
    206     def apply_async(self, func, callback=None):
    207         """Schedule a func to be run"""
--> 208         result = ImmediateResult(func)
    209         if callback:
    210             callback(result)

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/_parallel_backends.py in __init__(self, batch)
    570         # Don't delay the application, to avoid keeping the input
    571         # arguments in memory
--> 572         self.results = batch()
    573 
    574     def get(self):

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/parallel.py in __call__(self)
    260         # change the default number of processes to -1
    261         with parallel_backend(self._backend, n_jobs=self._n_jobs):
--> 262             return [func(*args, **kwargs)
    263                     for func, args, kwargs in self.items]
    264 

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/joblib/parallel.py in <listcomp>(.0)
    260         # change the default number of processes to -1
    261         with parallel_backend(self._backend, n_jobs=self._n_jobs):
--> 262             return [func(*args, **kwargs)
    263                     for func, args, kwargs in self.items]
    264 

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/squidpy/im/_feature.py in _calculate_image_features_helper(obs_ids, adata, img, layer, library_id, features, features_kwargs, queue, **kwargs)
    117 ) -> pd.DataFrame:
    118     features_list = []
--> 119     for crop in img.generate_spot_crops(
    120         adata, obs_names=obs_ids, library_id=library_id, return_obs=False, as_array=False, **kwargs
    121     ):

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/squidpy/im/_container.py in generate_spot_crops(self, adata, spatial_key, library_id, spot_scale, obs_names, as_array, squeeze, return_obs, **kwargs)
    813                 y = int(y - self.data.attrs[Key.img.coords].y0)
    814                 x = int(x - self.data.attrs[Key.img.coords].x0)
--> 815             crop = self.crop_center(y=y, x=x, radius=radius, library_id=obs_library_ids[i], **kwargs)
    816             crop.data.attrs[Key.img.obs] = obs
    817             crop = crop._maybe_as_array(as_array, squeeze=squeeze, lazy=False)

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/squidpy/im/_container.py in crop_center(self, y, x, radius, **kwargs)
    647         """
    648         y, x = self._convert_to_pixel_space((y, x))
--> 649         _assert_in_range(y, 0, self.shape[0], name="height")
    650         _assert_in_range(x, 0, self.shape[1], name="width")
    651 

/data/BCI-CRC/nasrine/anaconda/envs/scanpy/lib/python3.8/site-packages/squidpy/gr/_utils.py in _assert_in_range(value, minn, maxx, name)
    186 def _assert_in_range(value: float, minn: float, maxx: float, *, name: str) -> None:
    187     if not (minn <= value <= maxx):
--> 188         raise ValueError(f"Expected `{name}` to be in interval `[{minn}, {maxx}]`, found `{value}`.")
    189 
    190 

ValueError: Expected `height` to be in interval `[0, 2000]`, found `9761`.

[dst53]

Hi @Nasrine26

I was having the same issue and the solution here (set scale factors) worked for me: https://github.com/theislab/squidpy/issues/399

[DaisyCuttie]

Hi, I had the same issue as above "ValueError: Expected height to be in interval [0, 2000], found 21430". I tried adding a scale factor as mentioned in #399 but it does not solve my problem.

My code: `scale = 1 img = sq.im.ImageContainer("image_path", scale=scale) feature_name = f"features_summary_scale{scale}"

sq.im.calculate_image_features( adata, img, features="summary", key_added=feature_name, n_jobs=4, scale=scale )`

[giovp]

hi @Charlottehero ,

the scaling factor should be in adata.uns["spatial"]["V1_Adult_Mouse_Brain"]["scalefactors"], You should choose the one you are interested in with respect to your image resolution. May I kjnow what the file name in image_path in the image container you are pointing to? would be useful for debugging

[Nasrine26]

Hi @dst53,

the scale (set scale factors) worked for me as well. Many thanks!

[Nasrine26]

Hi @giovp,

I had a question concerning extracting segmentation features.

I'm using the full H&E image fromadata.uns['spatial'][library_id]['images']['hires'] to infer segmentation features.

More precisely, I perform the following steps:

1. smooth it using squidpy.im.process()
2. squidpy.im.segment() with method = 'watershed' to do the segmentation, use the channel 0 as it is supposed to contain most of the nuclei info for H&E stain
3. calculate segmentation features using:

sq.im.calculate_image_features(
    adata_vis,
    my_image,
    layer='image',
    features="segmentation",
    key_added="segmentation_features",
    features_kwargs={
        "segmentation": {
            "label_layer": "segmented_watershed",
            "props": ["label", "area", "mean_intensity"],
            "channels": [1, 2],
        }
    },
    mask_circle=True,
)

I then look at the number of nuclei per spot adata_vis.obsm["segmentation_features"]['segmentation_label'].describe()

1. How can I be sure the segmentation is well performed and trust these values?
2. Also is it ok if I use the image that's stored in adata.uns['spatial'][library_id]['images']['hires']?
3. Finally, Are the steps are perform ok or would you go about it another way?

Many thanks, I am very new to segmentation!

[DaisyCuttie]

Thanks @giovp. I figured out the scale factor value from your reply, but now I am encountering a resolution issue with the image. I used the sq.im.ImageContainer() to load in the hires_image I got from spatial results, but the resulting segmentation quality is not as good as what is shown in the tutorial here https://squidpy.readthedocs.io/en/stable/auto_tutorials/tutorial_visium_fluo.html. Am I going the right way?

[MxMstrmn]

Hi @Nasrine26,

regarding your questions:

1. I would suggest to look at the segmented image (stored as an additional layer in the ImageContainer) and do a visual inspectino of the results, see this tutorial
2. You can perform the segmentation on any image, so using adata.uns['spatial'][library_id]['images']['hires'] is fine.
3. I believe that the described steps are reasonable - most importantly the results should help your analysis. If they do, that is great!

[MxMstrmn]

Hi @Charlottehero,

plase have a look at the previous post to evaluate segmentation quality. Bad results can be related to the data, do you think that the image quality might be an issue?

[giovp]

closing due to inactivity, feel free to reopen