graemeball / omero_scripts_processing

python module to abstract image processing in omero scripts

omero processing scripts

Package for processing of images in omero.

OMERO is a platform for visualizing, managing, and annotating scientific image data. One of its components is the OMERO.processor which can launch python scripts for the processing of images and distribute the load over multiple processor nodes via OMERO.grid. This makes it ideal for batch processing of images, specially computationally-heavy steps.

The omero.scripts.processing package aims at creating a common interface for image processing scripts in omero.

The package defines two base classes, processing block and chain. The first is meant for individual processing steps, e.g., denoising with a specific algorithm, channel alignment, SI reconstruction, while the later is sequences of processing blocks.

This design of block is based on the idea that the processing of an image has the following stages:

1. get image: in most cases, this is simply downloading the image from the omero database. More complex cases can be multiple input images, conversion of the image into a different format.
2. parse options: each block may have multiple arguments which should be validated. This is performed after obtaining the image in case the default is image dependent. Still, in many cases it will be possible to obtain such default values from the Omero image and not from the file.
3. process: do the actual image processing. This step is the main change between the multiple subclasses of block, and is often split into further substeps.
4. import image: simply importing the image back into OMERO. At the simplest level, this will be a single image file
5. annotate: it is useful to annotate the processed images. The base class will attach a log file to the processed image, and add links between the original and processed (parent and child) images.

Examples

The aim is that a processing script becomes something as simple as:

import omero.scripts.processing
import omero.scripts.processing.denoise

dn = omero.scripts.processing.denoise.ndsafir()

chain = omero.scripts.processing.chain([dn])
chain.launch()

Of course, most of the work is in defining the new class, but by inheriting most of the steps from other classes, it should be much simpler. Actual examples that are currently in production at Micron Oxford, are available online.

To create a chain of blocks should be equally simple:

import omero.scripts.processing
import omero.scripts.processing.denoise
import omero.scripts.processing.deconvolution
import omero.scripts.processing.frobnicate

dn    = omero.scripts.denoise.ndsafir()
decon = omero.scripts.deconvolution.decon3d()
frob  = omero.scripts.frobnicate.frobnicate()

chain = omero.scripts.processing.chain([dn, decon, frob])
chain.launch()

Issues and Wishlist

• it seems that we can't group related options together without having a parent and a parent requires some sort of an option. See ome/openmicroscopy#2463

• handle version and contact values when creating the script interface.

• we upload and download the image from omero when going through a chain. This is safer but we could have an option where the intermediary files are used which saves us from downloading. Also, maybe some users are not interested in keeping the intermediary processing steps.

• should probably rename the block subclasses

• arguments on the GUI must have unique names which will not always be true. For example multiple blocks may have a correction option.

• we have checkboxes for each block in a chain but that was forced on us so that subgroups work properly. However, it may actually be a nice feature but we are not yet checking their values.

• have the chain handling the keep alive

• investigate having the chain getting the image and putting back into omero, possibly accepting functions from the block. Think chain working like pipes in a command line and blocks as filters.