QuPath Weka Extension

The QuPath Weka Extension enables classifiers from the popular open source machine learning library Weka to be used from within QuPath v0.1.2, and also facilitates the export of QuPath data in a Weka-friendly format.

The QuPath Weka Extension is not compatible with QuPath v0.2.0.

The QuPath Weka Extension is not an official piece of software by the developers of Weka, nor does it include or modify Weka in any way. Rather, it simply enables some of the useful features of Weka to be accessible from within QuPath's user interface.

Use of the QuPath Weka Extension requires Weka to be downloaded (and optionally installed) separately from http://www.cs.waikato.ac.nz/ml/weka/downloading.html

Installation

• Download Weka from http://www.cs.waikato.ac.nz/ml/weka/downloading.html
• Download the latest release of the QuPath Weka Extension from here
• Start running QuPath and drag the qupath-weka-extension.jar file onto the main window to install it
• Do either one of the following:
  1. Run the Classify → Create detection classifier (Weka) command
  2. If prompted, find the weka.jar file from within your Weka installation, and select it in the file chooser

If the above steps are successful, it should be possible to use the Classify → Create detection classifier (Weka) command for classifying detections as an alternative to the default Classify → Create detection classifier command.

For more information on classifiers and their usefulness, see the Classifying objects section of the documentation.

Why use the Weka extension?

Since QuPath already provides some machine learning capabilities with the Classify → Create detection classifier command (using OpenCV) to enable tasks such as cell classification or region identification, you may wonder why an alternative might be useful.

There are three main reasons for the existence of this Weka extension:

1. To provide access to (at least some of) Weka's classifiers

In some cases, there may be advantages to using one of Weka's classifiers (rather than the QuPath's OpenCV-based defaults) - either because of accuracy or performance, or simply because they are written in Java and therefore may be easier to explore, adapt or serialize.

2. To enable training data export to view in Weka Explorer

The Weka Explorer can be a very handy tool for examining data in more detail - including tasks such as feature selection, or assessing the performance of different classifiers.

Using the Weka extension, QuPath is able to export the data used to train a classifier in Weka's preferred .arff format, for easy import into Weka Explorer. This can be used to help gain insights into how best to approach classification in QuPath.

To do this, start training a classifier using Create detection classifier (Weka) within QuPath, and then click the Export training data for Weka button at the bottom of the classifier panel.

3. To demonstrate how additional machine learning libraries could be used with QuPath

There are many machine learning libraries that could be usefully employed within QuPath, but which are not included by default. The Weka extension shows one way in which a new library could be integrated with QuPath - giving a pattern to developers who might want to integrate another library.