rhizoTrak

Development

Table of Contents

1. Analyze and process csv files using R
2. Automatic segmentation in rhizoTrak
   • How to write your own segmentation operator in rhizoTrak
   • Template project for developing operators
   • Install your new operator in rhizoTrak

Analyze and process csv files using R

We provide the R package rhizoTrakAddons to assist in analyzing and processing csv files generated by rhizoTrak. Specifically the package aims at the analysis of csv files generated by rhizoTraks in ’Detailed Mode’. Detailed documentation is included in the vignette of the package. It may be download from this file or in near future installed from CRAN.

Automatic segmentation in rhizoTrak

How to write your own segmentation operator in rhizoTrak

In rhizoTrak you can aid the manual annotation process through an automatic segmentation. This automatic segmentation is performed by a MiToBo operator. MiToBo is a Java toolbox and programming environment for image processing and analysis algorithms deeply linked to ImageJ and Fiji. To learn more about MiToBo’s basic concepts of operators and how to implement your own operators, take a look at MiToBo’s User Guide. In addition, MiToBo’s Javadoc API gives you an overview of its functionality and more details about the programming interface.

For adding segmentation functionality to rhizoTrak an existing segmentation operator can be used or an own operator can be implemented. Every segmentation operator takes an image or several images as an input and returns the segmentation.

To get a good, i.e. useful, segmentation the operator has to fit the data and their specific characteristic. In some applications the existing operators do not perform well. Then it may be a good choice to write an own implementation of a segmentation operator. The following guide explains the parameters of the interface.

The class RootImageSegmentationOperator (in package de.unihalle.informatik.MiToBo.apps.minirhizotron.segmentation) is the interface for the operator. Therefore every operator has to extend this RootImageSegmentationOperator class.

Input parameter

Input images, existing treelines and meta data are hash maps with an integer key corresponding to the layerID and entries are of respective data type. inputImages are of type ImagePlus and inputTreelines are different MTBRootTree stored in a vector. The type of layerMetadata is RhizoProjectLayerMetadataContainer.

Operator configuration

The operator configuration defines the interaction with the rhizoTrak GUI.
The class RootImageSegmentationOperator has three different OpWorkingMode: SEGMENTATION_CREATE, SEGMENTATION_UPDATE and SEGMENTATION_CREATE_AND_UPDATE. The choosen operator working modus is passed via getOperatorWorkingMode() to rhizoTrak where the treelines are drawn. For different working modi different options can be configured:

1. SEGMENTATION_CREATE In create modus the result treelines of the operator are added. When adding in rhizoTrak one status can be choosen for all treelines.
2. SEGMENTATION_UPDATE In this modus the result treelines of the operator modify the existing treelines in rhizoTrak. Treelines can be deleted or transformed. When updating in rhizoTrak the status and/or diameter of the existing treelines can be modified (true) or preserved (false). The flag is set in the method isToTransferStatusOnUpdate(), respectively isToTransferDiameterOnUpdate().
3. SEGMENTATION_CREATE_AND_UPDATE The third modus is a combination of the two previous ones. First, treelines are updated while status and/or diameter can be preserved. Then, remaining treelines are added if there are any.

Choosing the layers to analyze, respectively images and treelines, a LayerSubset is given to select between ACTIVE, PREVIOUS, NEXT, ALL, FIRST_TO_PREVIOUS and NEXT_TO_LAST.
The operator specific EnumSet of different LayerSubset is programmed by overriding the methods getLayerSubsetForInputImages() and getLayerSubsetForInputTreelines(). The user can choose via this EnumSet which input images and treelines to analyze. In addition it can be choosen to get only selected treelines (getOnlySelectedTreelines()).
The operator configuration will be described in more detail in the section Template project for developing operators.

Output parameter

The operator returns the segmentation results as treelines of type MTBRootTree. If used within rhizoTrak, the results will be imported. The output parameter resultTreelines is of type HashMap<Integer, Vector<MTBRootTree>>.

• The Integer identifies the layerID.
  
  
• Vector<MTBRootTree> gives all treelines per layer.

A treeline consists of nodes of type MTBTreeNode with the data class MTBRootTreeNodeData, storing e.g. x and position, as well as radius and status.

Template project for developing operators

For a quick start in operator development we provide a pre-configured Maven project.
It can be downloaded from our Github repository.

To setup the project clone the repository to a folder of your choice.

In the sub-folder src/main/java/rhizoTrak/segmentation of the project you can find the dummy segmentation operator class RhizoTrakDummySegmentationDemo.java which you can use as template for your own implementation.
Note that the operator just creates some arbitrary treelines and only uses the width and height of the actual input image - it’s really just a dummy.

Some important notes regarding the demo operator:

• lines 76-78:
  your new operator class needs to extend RootImageSegmentationOperator and you need to add the annotation @ALDDerivedClass to your class definition, otherwise rhizoTrak will not be able to find your new operator
• lines 83-85:
  your operator must have a public constructor without any arguments, otherwise rhizoTrak cannot invoke an instance of your operator class
• lines 94ff:
  the operate() method is the method called from rhizoTrak to invoke your operator; at the time of invocation you can assume that the member variable this.inputImages inherited from the parent class contains the current image to be processed
• line 103:
  here we instantiate the result data structure which is subsequently filled with the result treelines
• line 120:
  node data, e.g. x and y position, is set in MTBRootTreeNodeData
• line 122:
  a new treeline is created via MTBRootTree giving the root node of type MTBTreeNode
• line 126:
  further nodes are added with method addChild(MTBTreeNode t)
• lines 146-148:
  every operator class is required to implement the method getUniqueClassIdentifier() to allow rhizoTrak to index objects of this class; choose a unique and comprehensive string as identifier
• lines 150ff:
  @Override the operator methods - will be explained more detailed in the following lines
• lines 151-154:
  the user can choose which images to analyze by selecting the layers, here all LayerSubset identifiers are given
• lines 157-159:
  the choice for selectable layers for input treelines is given, in this case no LayerSubset identifier is returned as this operator only creates a treeline and does not operate on existing treelines
• lines 162-164:
  if true only treelines selected by the user are provided, otherwise all treelines are passed to the operator
• lines 167-169:
  the operate can create (new) treelines, update existing treelines or do both, which means updating and also adding treelines. Here the operator creates one new treeline, existing treelines will not be modified
• lines 172-174 & 177-179:
  if true, diameter, respectively status, of result treelines is modified when updating in rhizoTrak. Otherwise the existing diameters/status labels of treelines are preserved.

Install your new operator in rhizoTrak

To install your new operator in rhizoTrak you need to build a jar archive containing the class file of your new operator class.

The easiest way to build such a jar is to run mvn package on the command line in your project.

Afterwards you can find the jar file in the sub-folder target of your project. Just copy the jar file to the jars folder of your local Fiji installation and restart Fiji. When you run rhizoTrak then your new operator should appear in the list of available detection operators in rhizoTrak.