View on GitHub

OMERO-wndcharm (beta)

Scripts for using Wndcharm in OMERO

Download this project as a .zip file Download this project as a tar.gz file

OMERO-wndcharm (beta)

OMERO-wndcharm is a small Python library and set of scripts for using WND-CHARM in OMERO.

WND-CHARM is a set of feature calculation and machine learning algorithms that is designed to work across a large variety of images.

At present OMERO-wndcharm should be considered beta quality. It is likely that the storage formats for features and schemas used for recording classification results and other metadata are likely to change.

Installation

See the README file.

Using OMERO-wndcharm

At present OMERO-wndcharm is limited to supervised classification, where all images have the same number of channels.

  1. Create a Project to hold the training set of images, the classifier will be named after this project. Within this project create some datasets, one for each class. The names of the datasets will be used as class labels. You can also create a test dataset, outside of the training project.

    Train and test datasets

  2. Calculate some features for the training project. Select the project, and load the Wndcharm Feature Extraction Multichannel script. Accept the default parameters and click Run Script. Do the same for the testing set.

    Script menu, feature calculationFeature calculation dialog

  3. Wait for the feature calculation to finish.

    Feature calculation scripts in progressFeature calculation scripts completed

  4. Each dataset should have a file attachment, containing the calculated features.

    Features file attachment

  5. Build the classifier. Select the training project, and load the Wndcharm Build Classifier script. Accept the default parameters and click Run Script.

    Script menu, build classifierBuild classifier dialog

  6. When the classifier has been built refresh the page and you should see three file attachments on the project. These hold the features, the feature weights and the class labels.

    Classifier file attachments

  7. You can test the classifier by predicting the class of the test images. Select the testing dataset, and load the Wndcharm Predict script. You must fill in the Training Project ID with the ID of the project that the classifier is attached to. Click Run Script.

    Script menu, predictionPrediction dialog

  8. When the prediction script has finished the test dataset will have a comment containing the predicted classes and marginal probabilities for all images in the dataset. You may need to refresh the page before it becomes visible.

    Prediction dataset comment

  9. In addition each test image will be tagged with its predicted class.

    Predicted class tag

  10. All tags associated with the built classifier belong to a single tagset, visible from the Tags view.

    Classifier tagset

  11. Advanced users with prior experience in machine learning may want to try varying the training parameters, or to evaluate the classifier using cross-validation.

    Script menu, cross-validationCross-validation dialogCross-validation attachment

  12. Note that in some cases it is not possible to delete the attachments created by OmeroWndcharm due to limitations in OMERO.server. The Wndcharm Remove Annotations script should work.