Documentation
WIPP
Table of contents

Machine Learning

The Machine Learning module is designed to classify regions of interest using image features (measurements).
It is accessible from the "Machine Learning" tab from the top menu bar, and allows:
  • - K-means clustering
WIPP Machine Learning module screenshot
WIPP Machine Learning module screenshot

From the Machine Learning view, click on the "K-means clustering Jobs" tab on the left menu to access the K-means clustering jobs view.
WIPP K-means clustering Jobs screenshot
WIPP K-means clustering Jobs screenshot

K-means clustering jobs

To create a new Machine Learning job, click on the "Create new machine learning job" button. Below is a description on how to configure each input parameter.
WIPP K-means clustering job with MIST screenshot
WIPP K-means clustering job screenshot

Job name

Unique name for the job.

Feature Extraction Result

Name of the Feature Extraction job whose result (image features) will be used for clustering.

Features Configuration

Choose which features will be used for clustering (all of the computed ones or only a subset).

Package

Choose which machine learning package to use. Only Weka is available in WIPP 2.0.0.

Max Number of Clusters

The maximum number of clusters (k) - indicates the maximum number of subsets to partition the selected feature vectors.

Merge Feature Extraction Results option

This option will run the clustering on all ROIs from all images at once, otherwise clustering will be run per image.

Merge K-means Results option

This option will merge all results from all images into one file.
The k-means clustering algorithm is an iterative algorithm which assigns each feature vector to the closest cluster in terms of the least squared Euclidean distance. Wikipedia page: https://en.wikipedia.org/wiki/K-means_clustering. The initialization of the clusters is performed randomly. The current algorithm is implemented in Java using the Weka library ( https://www.cs.waikato.ac.nz/ml/index.html).