Warning
AICSImageIO is now in maintenance mode only. Please take a look at its compatible successor bioio
(see here for migration guide)
Image Reading, Metadata Conversion, and Image Writing for Microscopy Images in Pure Python
-
Supports reading metadata and imaging data for:
OME-TIFF
TIFF
ND2
-- (pip install aicsimageio[nd2]
)DV
-- (pip install aicsimageio[dv]
)CZI
-- (pip install aicspylibczi>=3.1.1 fsspec>=2022.8.0
)LIF
-- (pip install readlif>=0.6.4
)PNG
,GIF
, etc. -- (pip install aicsimageio[base-imageio]
)- Files supported by Bio-Formats -- (
pip install aicsimageio bioformats_jar
) (Note: requiresjava
andmaven
, see below for details.)
-
Supports writing metadata and imaging data for:
OME-TIFF
PNG
,GIF
, etc. -- (pip install aicsimageio[base-imageio]
)
-
Supports reading and writing to fsspec supported file systems wherever possible:
- Local paths (i.e.
my-file.png
) - HTTP URLs (i.e.
https://my-domain.com/my-file.png
) - s3fs (i.e.
s3://my-bucket/my-file.png
) - gcsfs (i.e.
gcs://my-bucket/my-file.png
)
See Cloud IO Support for more details.
- Local paths (i.e.
Stable Release: pip install aicsimageio
Development Head: pip install git+https://github.com/AllenCellModeling/aicsimageio.git
AICSImageIO is supported on Windows, Mac, and Ubuntu. For other platforms, you will likely need to build from source.
TIFF and OME-TIFF reading and writing is always available after
installing aicsimageio
, but extra supported formats can be
optionally installed using [...]
syntax.
- For a single additional supported format (e.g. ND2):
pip install aicsimageio[nd2]
- For a single additional supported format (e.g. ND2), development head:
pip install "aicsimageio[nd2] @ git+https://github.com/AllenCellModeling/aicsimageio.git"
- For a single additional supported format (e.g. ND2), specific tag (e.g.
v4.0.0.dev6
):pip install "aicsimageio[nd2] @ git+https://github.com/AllenCellModeling/aicsimageio.git@v4.0.0.dev6"
- For faster OME-TIFF reading with tile tags:
pip install aicsimageio[bfio]
- For multiple additional supported formats:
pip install aicsimageio[base-imageio,nd2]
- For all additional supported (and openly licensed) formats:
pip install aicsimageio[all]
- Due to the GPL license, LIF support is not included with the
[all]
extra, and must be installed manually withpip install aicsimageio readlif>=0.6.4
- Due to the GPL license, CZI support is not included with the
[all]
extra, and must be installed manually withpip install aicsimageio aicspylibczi>=3.1.1 fsspec>=2022.8.0
- Due to the GPL license, Bio-Formats support is not included with the
[all]
extra, and must be installed manually withpip install aicsimageio bioformats_jar
. Important!! Bio-Formats support also requires ajava
andmvn
executable in the environment. The simplest method is to installbioformats_jar
from conda:conda install -c conda-forge bioformats_jar
(which will additionally bringopenjdk
andmaven
packages).
For full package documentation please visit allencellmodeling.github.io/aicsimageio.
If your image fits in memory:
from aicsimageio import AICSImage
# Get an AICSImage object
img = AICSImage("my_file.tiff") # selects the first scene found
img.data # returns 5D TCZYX numpy array
img.xarray_data # returns 5D TCZYX xarray data array backed by numpy
img.dims # returns a Dimensions object
img.dims.order # returns string "TCZYX"
img.dims.X # returns size of X dimension
img.shape # returns tuple of dimension sizes in TCZYX order
img.get_image_data("CZYX", T=0) # returns 4D CZYX numpy array
# Get the id of the current operating scene
img.current_scene
# Get a list valid scene ids
img.scenes
# Change scene using name
img.set_scene("Image:1")
# Or by scene index
img.set_scene(1)
# Use the same operations on a different scene
# ...
The .data
and .xarray_data
properties will load the whole scene into memory.
The .get_image_data
function will load the whole scene into memory and then retrieve
the specified chunk.
If your image doesn't fit in memory:
from aicsimageio import AICSImage
# Get an AICSImage object
img = AICSImage("my_file.tiff") # selects the first scene found
img.dask_data # returns 5D TCZYX dask array
img.xarray_dask_data # returns 5D TCZYX xarray data array backed by dask array
img.dims # returns a Dimensions object
img.dims.order # returns string "TCZYX"
img.dims.X # returns size of X dimension
img.shape # returns tuple of dimension sizes in TCZYX order
# Pull only a specific chunk in-memory
lazy_t0 = img.get_image_dask_data("CZYX", T=0) # returns out-of-memory 4D dask array
t0 = lazy_t0.compute() # returns in-memory 4D numpy array
# Get the id of the current operating scene
img.current_scene
# Get a list valid scene ids
img.scenes
# Change scene using name
img.set_scene("Image:1")
# Or by scene index
img.set_scene(1)
# Use the same operations on a different scene
# ...
The .dask_data
and .xarray_dask_data
properties and the .get_image_dask_data
function will not load any piece of the imaging data into memory until you specifically
call .compute
on the returned Dask array. In doing so, you will only then load the
selected chunk in-memory.
Read stitched data or single tiles as a dimension.
Readers that support mosaic tile stitching:
LifReader
CziReader
If the file format reader supports stitching mosaic tiles together, the
AICSImage
object will default to stitching the tiles back together.
img = AICSImage("very-large-mosaic.lif")
img.dims.order # T, C, Z, big Y, big X, (S optional)
img.dask_data # Dask chunks fall on tile boundaries, pull YX chunks out of the image
This behavior can be manually turned off:
img = AICSImage("very-large-mosaic.lif", reconstruct_mosaic=False)
img.dims.order # M (tile index), T, C, Z, small Y, small X, (S optional)
img.dask_data # Chunks use normal ZYX
If the reader does not support stitching tiles together the M tile index will be
available on the AICSImage
object:
img = AICSImage("some-unsupported-mosaic-stitching-format.ext")
img.dims.order # M (tile index), T, C, Z, small Y, small X, (S optional)
img.dask_data # Chunks use normal ZYX
If the file format reader detects mosaic tiles in the image, the Reader
object
will store the tiles as a dimension.
If tile stitching is implemented, the Reader
can also return the stitched image.
reader = LifReader("ver-large-mosaic.lif")
reader.dims.order # M, T, C, Z, tile size Y, tile size X, (S optional)
reader.dask_data # normal operations, can use M dimension to select individual tiles
reader.mosaic_dask_data # returns stitched mosaic - T, C, Z, big Y, big, X, (S optional)
There are functions available on both the AICSImage
and Reader
objects
to help with single tile positioning:
img = AICSImage("very-large-mosaic.lif")
img.mosaic_tile_dims # Returns a Dimensions object with just Y and X dim sizes
img.mosaic_tile_dims.Y # 512 (for example)
# Get the tile start indices (top left corner of tile)
y_start_index, x_start_index = img.get_mosaic_tile_position(12)
from aicsimageio import AICSImage
# Get an AICSImage object
img = AICSImage("my_file.tiff") # selects the first scene found
img.metadata # returns the metadata object for this file format (XML, JSON, etc.)
img.channel_names # returns a list of string channel names found in the metadata
img.physical_pixel_sizes.Z # returns the Z dimension pixel size as found in the metadata
img.physical_pixel_sizes.Y # returns the Y dimension pixel size as found in the metadata
img.physical_pixel_sizes.X # returns the X dimension pixel size as found in the metadata
If aicsimageio
finds coordinate information for the spatial-temporal dimensions of
the image in metadata, you can use
xarray for indexing by coordinates.
from aicsimageio import AICSImage
# Get an AICSImage object
img = AICSImage("my_file.ome.tiff")
# Get the first ten seconds (not frames)
first_ten_seconds = img.xarray_data.loc[:10] # returns an xarray.DataArray
# Get the first ten major units (usually micrometers, not indices) in Z
first_ten_mm_in_z = img.xarray_data.loc[:, :, :10]
# Get the first ten major units (usually micrometers, not indices) in Y
first_ten_mm_in_y = img.xarray_data.loc[:, :, :, :10]
# Get the first ten major units (usually micrometers, not indices) in X
first_ten_mm_in_x = img.xarray_data.loc[:, :, :, :, :10]
See xarray
"Indexing and Selecting Data" Documentation
for more information.
File-System Specification (fsspec) allows for common object storage services (S3, GCS, etc.) to act like normal filesystems by following the same base specification across them all. AICSImageIO utilizes this standard specification to make it possible to read directly from remote resources when the specification is installed.
from aicsimageio import AICSImage
# Get an AICSImage object
img = AICSImage("http://my-website.com/my_file.tiff")
img = AICSImage("s3://my-bucket/my_file.tiff")
img = AICSImage("gcs://my-bucket/my_file.tiff")
# Or read with specific filesystem creation arguments
img = AICSImage("s3://my-bucket/my_file.tiff", fs_kwargs=dict(anon=True))
img = AICSImage("gcs://my-bucket/my_file.tiff", fs_kwargs=dict(anon=True))
# All other normal operations work just fine
Remote reading requires that the file-system specification implementation for the target backend is installed.
- For
s3
:pip install s3fs
- For
gs
:pip install gcsfs
See the list of known implementations.
The simpliest method to save your image as an OME-TIFF file with key pieces of
metadata is to use the save
function.
from aicsimageio import AICSImage
AICSImage("my_file.czi").save("my_file.ome.tiff")
Note: By default aicsimageio
will generate only a portion of metadata to pass
along from the reader to the OME model. This function currently does not do a full
metadata translation.
For finer grain customization of the metadata, scenes, or if you want to save an array as an OME-TIFF, the writer class can also be used to customize as needed.
import numpy as np
from aicsimageio.writers import OmeTiffWriter
image = np.random.rand(10, 3, 1024, 2048)
OmeTiffWriter.save(image, "file.ome.tif", dim_order="ZCYX")
See OmeTiffWriter documentation for more details.
In most cases, AICSImage.save
is usually a good default but there are other image
writers available. For more information, please refer to
our writers documentation.
AICSImageIO is benchmarked using asv.
You can find the benchmark results for every commit to main
starting at the 4.0
release on our
benchmarks page.
See our developer resources for information related to developing the code.
If you find aicsimageio
useful, please cite this repository as:
Eva Maxfield Brown, Dan Toloudis, Jamie Sherman, Madison Swain-Bowden, Talley Lambert, AICSImageIO Contributors (2021). AICSImageIO: Image Reading, Metadata Conversion, and Image Writing for Microscopy Images in Pure Python [Computer software]. GitHub. https://github.com/AllenCellModeling/aicsimageio
bibtex:
@misc{aicsimageio,
author = {Brown, Eva Maxfield and Toloudis, Dan and Sherman, Jamie and Swain-Bowden, Madison and Lambert, Talley and {AICSImageIO Contributors}},
title = {AICSImageIO: Image Reading, Metadata Conversion, and Image Writing for Microscopy Images in Pure Python},
year = {2021},
publisher = {GitHub},
url = {https://github.com/AllenCellModeling/aicsimageio}
}
Free software: BSD-3-Clause
(The LIF component is licensed under GPLv3 and is not included in this package) (The Bio-Formats component is licensed under GPLv2 and is not included in this package) (The CZI component is licensed under GPLv3 and is not included in this package)