Feature Extraction

The Feature Extraction module is designed to extract local and global image features (measurements).
Each image feature extraction can be configured by following a sequence of steps. The configuration and launching of feature extraction computational jobs is accessible from the "Feature Extraction" tab of the top menu bar. The feature extraction module provides the following functionalities:

• - feature extraction per image
• - feature extraction per region of interest or ROIs (images + masks)
• - feature extraction per small partition of ROI (images + masks + tiles)
• - download of feature values in a comma separated value (CSV) file format
• - interactive spatial and feature value based selection of data points for statistical modeling
• - interactive modeling of feature histogram distribution
• - histogram plots for electronic publications with traceable data points to the source image collections

Extractors documentation

This documentation page displays information about each software packages that has been integrated to extract image features. By following the links, a user can learn about all image features supported by each software package and download the executable.
From the Feature Extraction view, click on the "Extractors Documentation" tab on the left menu to access the Feature Extractors documentation.

Click on a software name to go to the details page.

Feature Extraction jobs

This page displays all executed feature extraction jobs.
From the Feature Extraction view, click on the "Feature Extraction Jobs" tab on the left menu to access the Feature Extraction jobs view.

Creating a Feature Extraction job to extract image features

To create a new Feature Extraction job, click on the "Create new feature extraction job" button. To configure each input parameter, follow the instructions below.

Configure:

Type a unique name and optionally an email address, then click on the button "NEXT STEP"

Select images:

Input images

The minimum input for running a feature extraction job is a collection for the Input Images (images on which the features will be computed) and a filename pattern to filter the images in the collection by using a regular expression. For using all files in the collection, you can click on "Use all .ome.tif. images".
Click on "Check pattern" to verify that your regular expression is valid. Optional Pyramid input: by specifying a pyramid, you can use the statistical modeling tool to filter and analyze all computed features. The specified pyramid should be one representing the same image collection (or a processed version of the same image collection, for example after intensity scaling for rendering if the images are not 8bbb, but a warning will be displayed) and generated from an assembled image (not a grid of image tiles).
If you are working with a grid of tiles, you should first assemble them in the Image Processing module before using them in the Feature Extraction module.

Input partitions (masks) - optional

Specify a mask collection for features to be computed on Regions of Interest (ROI). The masks should be labeled images.
Click on "Check pattern and files alignment" to verify that your image name pattern is valid and that your input images and mask images are properly matching.

Input groups (tiles) - optional

Specify a mask collection for features to be computed on tessellations (tiles or partitions) inside of the Regions of Interest (ROI) defined in the masks. The tessellation images should be labeled.
Click on "Check pattern and files alignment" to verify that your image name pattern is valid and that your input images, mask images and tessellation images are properly matching.

If one of the image collection is not locked (i.e., it can still be modified), you have to lock it using the "Lock collection" button before submitting the job.

Once the images configuration is done, click on "NEXT STEP".

Select features:

Click on the "Add feature(s)" button to add features. If you would like to know more about the available features and the feature extractors, then visit the Documentation section.

You can filter features by Extractor, Category, and also by name. If you click on the "Select all displayed features" button, all the features that are matching your filters and that you see if the "Search results" are going to be selected. You can also select/unselect features individually. Once you finalize your selection, you can click on the "Add selected feature(s)" button.
The features you selected are displayed below "Selected image features to compute". Each box displays the name and the short description of a selected feature, the name of the extractor computing the chosen feature, and the code of the feature that you will see in the header of a resulting table. You can remove features that you have selected, or add new ones. If some extractors are configurable, then you will find the options in the right panel.
Once the features configuration is done, click on "NEXT STEP".

Review and submit:

You can review the job configuration configuration and then submit the job by clicking on the "Submit" button. If one of the image collections you specified has not been locked, then you still have the possibility to lock it from this page (locking collections is mandatory before launching a job).
Once you click on the "Submit" button, a spinner will appear and you will be re-directed to the results page.

Viewing the Feature Extraction job results

To view the job results and download the features values, you can go to the Feature Extraction Jobs section, where you will find the list of jobs. You can click on each job to go to the details page.

On the results details page of each job, you will find general information, such as the configuration of the job and its status (SUBMITTED, QUEUED, RUNNING, DONE or FAILED). If the status is DONE, a button "Download results" will appear for you to download the feature values as CSV files. If you selected a pyramid, a "View stat modeling" button will also appear.

Statistical modeling view

If available for the job, a "View stat modeling" button will be present from the job detail page.
The statistical modeling view gives a representation of the feature values as a histogram using the pyramid views of individual objects (ROIs) in each bin. The functionality of this module includes:

• 1. interactive filtering of image objects and their features based on spatial location and feature values ("Spatial filter" and "Feature filter"),
• 2. computing statistics of filtered features ("Statistical modeling" -> "Show Stats"),
• 3. estimating statistical probability distribution function (PDF) models from image feature histograms and generating feature values according to the estimated PDF model ("Statistical modeling"->"Recommend PDF model" and "Estimate PDF parameters"),
• 4. saving feature histograms with thumbnail object images that are traceable to persistent large image collections ("Publish"),
• 5. interactive rendering of a dynamically filtered histogram that contains zoom-able images of objects described by a selected object feature (right panel)

One can select a histogram feature and the number of histogram bins between the maximum and minimum feature values from "Feature selection".