Skip to content
This repository has been archived by the owner on Aug 10, 2020. It is now read-only.

ieivanov/aicsimageio

 
 

AICSImageIO

Build Status Documentation Code Coverage

Delayed Parallel Image Reading for Microscopy Images in Python


Features

  • Supports reading metadata and imaging data for:
    • CZI
    • OME-TIFF
    • TIFF
    • LIF
    • Any additional format supported by imageio
  • Supports writing metadata and imaging data for:
    • OME-TIFF

Installation

Stable Release: pip install aicsimageio
Development Head: pip install git+https://github.com/AllenCellModeling/aicsimageio.git

Documentation

For full package documentation please visit allencellmodeling.github.io/aicsimageio.

Quick Start

Full Image Reading

from aicsimageio import AICSImage, imread

# Get an AICSImage object
img = AICSImage("my_file.tiff")
img.data  # returns 6D STCZYX numpy array
img.dims  # returns string "STCZYX"
img.shape  # returns tuple of dimension sizes in STCZYX order

# Get 6D STCZYX numpy array
data = imread("my_file.tiff")

Delayed Image Slice Reading

from aicsimageio import AICSImage, imread_dask

# Get an AICSImage object
img = AICSImage("my_file.tiff")
img.dask_data  # returns 6D STCZYX dask array
img.dims  # returns string "STCZYX"
img.shape  # returns tuple of dimension sizes in STCZYX order
img.size("STC")  # returns tuple of dimensions sizes for just STC
img.get_image_data("CZYX", S=0, T=0)  # returns 4D CZYX numpy array
img.get_image_dask_data("CZYX", S=0, T=0)  # returns 4D CZYX dask array

# Read specified portion of dask array
lazy_s0t0 = img.get_image_dask_data("CZYX", S=0, T=0)  # returns 4D CZYX dask array
s0t0 = lazy_s0t0.compute()  # returns 4D CZYX numpy array

# Or use normal numpy array slicing
lazy_data = imread_dask("my_file.tiff")
lazy_s0t0 = lazy_data[0, 0, :]
s0t0 = lazy_s0t0.compute()

Speed up IO and Processing with Dask Clients and Clusters

If you have already spun up a distributed.Client object in your Python process or your processing is running on a distributed worker, great, you will naturally gain IO and processing gains. If you haven't done that or don't know what either of those are, there are some utility functions to help construct and manage these for you.

from aicsimageio import AICSImage, dask_utils

# Create a local dask cluster and client for the duration of the context manager
with AICSImage("filename.ome.tiff") as img:
    # do your work like normal
    print(img.dask_data.shape)

# Specify arguments for the local cluster initialization
with AICSImage("filename.ome.tiff", dask_kwargs={"nworkers": 4}) as img:
    # do your work like normal
    print(img.dask_data.shape)

# Connect to a dask client for the duration of the context manager
with AICSImage(
    "filename.ome.tiff",
    dask_kwargs={"address": "tcp://localhost:12345"}
) as img:
    # do your work like normal
    print(img.dask_data.shape)

# Or spawn a local cluster and / or connect to a client outside of a context manager
# This uses the same "address" and dask kwargs as above
# If you pass an address in, it will create and shutdown the client
# and no cluster will be created.
# Similar to AICSImage, these objects will be connected and useable
# for the lifespan of the context manager.
with dask_utils.cluster_and_client() as (cluster, client):

    img1 = AICSImage("1.tiff")
    img2 = AICSImage("2.tiff")
    img3 = AICSImage("3.tiff")

    # Do your image processing work

Note: The AICSImage context manager and the dask_utils module require that the processing machine or container have networking capabilities enabled to function properly.

Metadata Reading

from aicsimageio import AICSImage

# Get an AICSImage object
img = AICSImage("my_file.tiff")
img.metadata  # returns the metadata object for this image type
img.get_channel_names()  # returns a list of string channel names found in the metadata

Napari Interactive Viewer

napari is a fast, interactive, multi-dimensional image viewer for python and it is pretty useful for imaging data that this package tends to interact with.

from aicsimageio import AICSImage

# Get an AICSImage object
img = AICSImage("my_file.tiff")
img.view_napari()  # launches napari GUI and viewer

We have also released napari-aicsimageio, a plugin that allows use of all the functionality described here, but in the napari default viewer itself.

Performance Considerations

  • If your image fits into memory and you are not using a distributed cluster: use AICSImage.data or Reader.data which are generally optimal.
  • If your image is too large to fit into memory: use AICSImage.get_image_data to get a numpy array or AICSImage.get_image_dask_data to get a dask array for a specific chunk of data from the image.
  • If you are using a distributed cluster: all functions and properties in the library are generally optimal.
  • If you are using a distributed cluster with less than ~6 workers: use aicsimageio.use_dask(False). From our testing, 6 workers is the bare minimum for read time reduction compared to no cluster usage.
  • When using a dask array, it is important to know when to compute or persist data and when to keep chaining computation. Here is a good rundown on the trade offs.

Notes

  • Image data and dask_data are always returned as six dimensional in dimension order STCZYX or Scene, Time, Channel, Z, Y, and X.
  • Each file format may use a different metadata parser it is dependent on the reader's implementation.
  • The AICSImage object will only pull the Scene, Time, Channel, Z, Y, X dimensions from the reader. If your file has dimensions outside of those, use the base reader classes CziReader, OmeTiffReader, TiffReader, or DefaultReader.
  • We make some choices for the user based off the image data during img.view_napari. If you don't want this behavior, simply pass the img.dask_data into napari.view_image instead.

Development

See CONTRIBUTING.md for information related to developing the code.

Free software: BSD-3-Clause

About

Delayed Parallel Image Reading for Microscopy Images in Python

Resources

License

Code of conduct

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 99.3%
  • Makefile 0.7%