[ome-devel] OMERO.WND-CHARM Developer Preview Release Announcement

Coletta, Christopher (NIH/NIA/IRP) [E] christopher.coletta at nih.gov
Tue Aug 19 14:55:39 BST 2014


OMERO.WND-CHARM Developer Preview Release Announcement

OMERO.WND-CHARM is a machine learning analysis plugin for OMERO 5 being jointly developed by the Swedlow Lab at the University of Dundee and the Goldberg Lab at the National Institute on Aging in Baltimore.

It combines the multi-purpose image feature extraction and classification capabilities of the WND-CHARM project with the image/metadata manipulation and storage capabilities of OMERO.

The purpose of the release is to provide a preview for OMERO developers and is not feature complete, nor is it is ready for production. OMERO.WND-CHARM 1.0 will be released later in tandem with the 1.0 release of WND-CHARM's Python API.

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#motivation>Motivation

WND-CHARM is a multi-purpose image classifier that operates on image features that are not specific to a biological domain or imaging modality, and yet are effective at discriminating a wide range of morphology. Use OMERO.WND-CHARM to:

  *   Create classifiers for OMERO image data that are context-specific/sensitive to individual experimental parameters.
  *   Classify images and tag them in OMERO based on classification results.
  *   Quantitatively measure morphological similarity between images or groups of images. Applications for a properly-calibrated classifier can vary from characterizing cancer metastasis in H&E-stained tissue microarrays (Orlov et al. 2012<http://www.ncbi.nlm.nih.gov/pubmed/22467531>), to locating an anatomical region containing predictive signal for the future development of osteoarthritis in knee X-rays (Shamir et al. 2009<http://www.ncbi.nlm.nih.gov/pubmed/19426848>), to distinguishing subtle sub-cellular morphological differences induced by different mutant alleles (Tadeu et al. 2008<http://www.ncbi.nlm.nih.gov/pubmed/18772885>).

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#implementation-description>Implementation Description

OMERO.WND-CHARM is a pip-installable Python package built using the OMERO.tables and OMERO.scripts APIs, along with OMERO's tagging and file attachment features. The python package calls the WND-CHARM image feature extraction library written in C++ and wrapped in Python via SWIG. It stores and retrieves image features via OMERO.tables, trains a classifier based on user-defined OMERO datasets, and outputs classifier results as tags and/or text files attached to containing projects and datasets.

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#installation>Installation

  1.  Make sure you have the WND-CHARM Python API dependencies, details on the WND-CHARM github page<https://github.com/wnd-charm/wnd-charm>.
  2.  Install wndcharm via pip install git+git://github.com/wnd-charm/wnd-charm.git
  3.  Install the OMERO.WND-CHARM Python package<https://github.com/ome/omero-wndcharm> via pip install git+git://github.com/ome/omero-wndcharm.git
  4.  Copy the OMERO.WND-CHARM scripts<https://github.com/ome/omero-wndcharm/tree/master/scripts> into the OMERO.server script directory

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#workflow>Workflow

  1.  Arrange training images into classes by grouping them into OMERO Datasets within a top-level OMERO Project.
  2.  Calculate image features using "Wndcharm Feature Extraction Multichannel" script.
  3.  If necessary, check the progress of the feature calculation with the "Wndcharm Feature Check Progress" script.
  4.  Train a classifier using the "Wndcharm Build Classifier Script".
  5.  Evaluate classifier performance using the "Wndcharm Cross Validation" script, which performs train/test splits and writes the results to a text file attached to the top-level OMERO Project.
  6.  Use the classifier to tag images in a different project with the "Wndcharm Predict" script. Morphological similarity measurements for each image are saved to a text file attached to the construct pairwise distances matrices for clustering and other morphology analyses.

These instructions are also available online<http://ome.github.io/omero-wndcharm/>, along with screenshots.

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#limitations>Limitations

  1.  The inability of the current release to operate on sub-image regions (e.g. tiles and ROIs) will be addressed in the upcoming 1.0 release.
  2.  The current release's fixed set of feature calculation algorithms will be expanded and changed to allow much greater control over what features are calculated in the upcoming 1.0 release.
  3.  Handling of color images (channels) is currently fixed and limited. The upcoming 1.0 release will be much more flexible in handling multi-D image data.

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#future-directions>Future Directions

  *   Integration with the forthcoming OMERO.features API<https://github.com/ome/omero-features>.
  *   Support for more classifiers/regressors/metrics/hyperparameter optimization tools via integration with scikit-learn<http://scikit-learn.org>.
  *   Object detection/trainable segmentation using scanning-window classifier.

<https://gist.github.com/manics/4fcf7b7062eb838da651/d87cbdfa1aa1bb086cdfcd76cc3a1fff2a91cfd6#credits>Credits

  *   Swedlow group: Simon Li, Josh Moore.
  *   Goldberg group: Chris Coletta, Ilya Goldberg.


Christopher Coletta
Computer Scientist
Image Informatics and Computational Biology Unit
Laboratory of Genetics
National Institute on Aging
Biomedical Research Center
251 Bayview Boulevard, Room 10B125
Baltimore, MD 21224
Desk: 410-558-8170
Cell: 617-943-9745



More information about the ome-devel mailing list