ZEISS Knowledge Base

Introduction

This guide explains you how to detect nuclei and cells using the pipeline with the Machine Learning Segmenter operation. Machine learning (ML) is a branch of Artificial intelligence (AI) in which, based on the training dataset that are first provided, the computer develops its own logic for answering future questions. The key concept of machine learning is to produce accurate predictions on new unseen data after being trained on a finite learning dataset.

Demo dataset

The data set is a multi-dimensional, discrete representation of your real sample volume. It can be structured as a z-series of planes (optical slices) of multiple channels (dyes) in a temporal sequence of time points located at multiple spatial positions. Usually, the dataset shows a single experimental situation. A complete experiment can be composed by several datasets. The datasets are available as graphic files saved in plenty of file formats (standard formats as well as proprietary formats).

The link for the specific demo dataset for this guide is displayed below. All datasets are listed here: https://demodata.arivis.com

Downloading the demo dataset

To download the demo dataset, click http://demodata.arivis.com/data/arivisVision4D-DemoData-SamplePipelines-ML.zip
The download is starting automatically.
Create a new folder on your local disk. Move the *.zip file from the download folder inside it.
Unzip the folder.

Opening the demo dataset

Click File > Open...
Select the *.sis file from the Windows Explorer.
The demo dataset is displayed in the viewer. The dataset is visualized according to the current rendering setting parameters.

Activating the Sample Pipeline

In the Shortcut Toolbar, click Analysis Panel .
In the Sample Pipelines list, double-click the Detect Nuclei And Cells Using Machine Learning Pipeline.
If you have activated a pipeline, it will be replaced by the new one.

You can open the appropriate How to guide. By hovering over a pipeline, this button is displayed. When clicking on it, the option Open How to appears.

Executing the pipeline

You can execute the pipeline step by step or in a single run. To do this, use the executing buttons in the Pipeline toolbar.

Executing step by step

You can execute the pipeline step by step (back and forth). This method allows to run and undo a single operation. You can either use the executing buttons in the Pipeline toolbar or in the Operation toolbar to go through the operation list.

To run the single operation, click .
To undo the single operation, click .
Note: Undo the last operation executed if you need to change the operation settings.

Operation status

Executing buttons in the Operation toolbar after executing

When the operation is running, this icon is shown.

When the operation is completed, this icon is shown.

Viewing the Results

If not already visible, open the Objects dialog. In Shortcut Toolbar, click Objects table .
Measurements are now visible in the objects table.
To add or remove table columns, click Feature Columns...
For more information, refer to the Online Help (F1).

Results in the Viewer

Opening the ML Trainer panel

Click Analysis > Machine Learning Trainer (Image)
The Machine Trainer panel is displayed.

Select the Channel.
Note: You can select a single or multiple channels.
To select the features that are used for the machine learning, select from the Feature set dropdown list.
To set up the Custom feature, click Feature .
The Feature dialog is displayed.

Note: You can select all or a combination of the available features and resolutions. You can apply the features dimensions to 2D or 3D.
To disable a feature totally, click on a feature row header.
To disable a resolution totally, click on a resolution column header.
Click OK.
Select the Classes by which to classify the pixels in the image.
Note: By default, the Background class and the Class 1 object class is added.
To add classes, click + Add Class
To rename the class, double-click on the class header.
Click Brush tool.
The cursor switched to a brush.
To select regions representative for the classes, drag the brush on the image.
Note: Annotate several small segments rather than large few. Try to cover all the different structure conditions (e.g., intensity, texture, etc.) present in your training images. Don’t overlay annotation to the background and other structures.
Each annotation is shown under the related class.
To delete an annotation, right-click on it and click Remove selected object.
To check the quality of the training , click Preview tabs.
The preview shows which pixels will be segmented.
Click Train.
To export the training for further usage, click Panel menu > Export.
To open the pipeline with the Machine Learning Segmenter operation, click Open in Pipeline.