Parameter tuning and optimization - Evaluating nuclei segmentation with Galaxy

Author(s) orcid logoRiccardo Massei avatar Riccardo Massei
Reviewers Leonid Kostrykin avatarBeatriz Serrano-Solano avatar
Overview
Creative Commons License: CC-BY Questions:

  • What are the challenges of using the same settings for every biological image, and how does parameter tuning address these challenges?

  • How can we choose the best filters for preprocessing images for nuclei segmentation?

Objectives:

  • Understand the importance of parameter tuning in bioimage analysis for achieving accurate results

  • Learn how to perform parameter tuning for segmentation in Galaxy

Requirements:
Time estimation: 1 hour
Level: Introductory
Supporting Materials:
Published: Apr 28, 2025
Last modification: Apr 28, 2025
License: Tutorial Content is licensed under Creative Commons Attribution 4.0 International License. The GTN Framework is licensed under MIT
Revision: 1

Parameter tuning is super important in image analysis. When you’re doing image analysis — like segmentation, quantification, or feature extraction — the settings you choose make a big difference in how accurate your results are. But biological images can vary strongly in terms of quality, contrast, and structure. So you can’t just use the same settings for every image.

That’s where parameter tuning comes in. By adjusting these settings carefully, researchers can make sure their tools are capturing the right biological info without adding noise or messing up the images. This helps make sure their work is repeatable and reliable, which is key for getting accurate scientific results. Basically, it’s all about finding the perfect balance to get the most useful and accurate data possible.

In this tutorial, we will show how to perform parameter tuning for segmentation. In particular, we will fine-tune a parameter of Gaussian filters for nuclei segmentation by quantitatively comparing ground-truth segmentation with the ones processed in Galaxy. As a dataset, we will use GFP-GOWT1 mouse stem cell images, which were acquired from Bártová et al. 2011. The training dataset contains a collection of images plus their ground truth (Maška et al. 2014). The images are publicly available at the ISBI Cell Tracking Challenge repository.

graph_comparison.png.

So, let’s proceed!

Get the image data

Hands On: Data Upload

  1. Create a new history for this tutorial.

  2. Download the following image and import it into your Galaxy history.

    If you are importing the image via URL:

    • Copy the link location

    • Click Upload Data at the top of the tool panel

    • Select Paste/Fetch Data

    • Paste the link(s) into the text field

    • Press Start

    • Close the window

    If you are importing the image from the shared data library:

    As an alternative to uploading the data from a URL or your computer, the files may also have been made available from a shared data library:

    1. Go into Libraries (left panel)
    2. Navigate to the correct folder as indicated by your instructor.
      • On most Galaxies tutorial data will be provided in a folder named GTN - Material –> Topic Name -> Tutorial Name.
    3. Select the desired files
    4. Click on Add to History near the top and select as Datasets from the dropdown menu

    5. In the pop-up window, choose

      • “Select history”: the history you want to import the data to (or create a new one)
    6. Click on Import

  3. Rename the datasets appropriately if needed (e.g. "original_image", "ground_truth")

  4. Confirm the datatypes are correct (tiff for both images)

    • Click on the pencil icon for the dataset to edit its attributes
    • In the central panel, click Datatypes tab on the top
    • In the Assign Datatype, select datatypes from “New type” dropdown
      • Tip: you can start typing the datatype into the field to filter the dropdown menu
    • Click the Save button

    As an alternative to uploading the data from a URL or your computer, the files may also have been made available from a shared data library:

    1. Go into Libraries (left panel)
    2. Navigate to the correct folder as indicated by your instructor.
      • On most Galaxies tutorial data will be provided in a folder named GTN - Material –> Topic Name -> Tutorial Name.
    3. Select the desired files
    4. Click on Add to History near the top and select as Datasets from the dropdown menu

    5. In the pop-up window, choose

      • “Select history”: the history you want to import the data to (or create a new one)
    6. Click on Import

Define a series of values to test

The parameter of Gaussian filters that you are going to tune in this tutorial is sigma, that is the size of the filters. You can now input a series of sigma values you want to test. Values need to be imported as a tabular file as shown in the figure:

fetch_data.jpg.

Next, copy and paste or manually add the desired values. Additionally, change the file format to tabular. Then, rename the file to sigma_values and click Start. The dataset will appear now in your history.

You now have everything you need to build the parameter tuning workflow!

Evaluate nuclei segmentation - Full workflow

Hands On: Create a workflow to evaluate nuclei segmentation

  1. Create a new workflow in the workflow editor.

    1. Click Workflow on the top bar
    2. Click the new workflow button
    3. Give it a clear and memorable name
    4. Clicking Save will take you directly into the workflow editor for that workflow
    5. Need more help? Please see the How to make a workflow subsection here

  2. Add three Input dataset:
    • 1: Input Dataset, 2: Input Dataset and 3: Input Dataset will appear in your workflow. Change the “Label” of these inputs to Dataset to perform segmentation, Input ground truth and Sigma values to test
  3. Add Split by group ( Galaxy version 0.6)
  4. Add Parse parameter value from the list of tools with the following recommended parameters:
    • Select type of parameter to parse: Float
  5. Add Filter 2-D image ( Galaxy version 1.12.0+galaxy1) from the list of tools with the following recommended parameters:
    • Filter Type: Gaussian
    • Sigma: Click the “Add Connection to Module” symbol
  6. Add Perform histogram equalization ( Galaxy version 0.18.1+galaxy0) from the list of tools with the following recommended parameters:
    • Histogram equalization algorithm: CLAHE
  7. Add Threshold image ( Galaxy version 0.18.1+galaxy3) from the list of tools with the following recommended parameters:
    • Thresholding method : Globally Adaptive/Otsu
    • Offset: 0.0
  8. Add Compute image segmentation and object detection performance measures ( Galaxy version 1.4.0-2) from the list of tools with the following recommended parameters:
    • Unzip: No
    • Segmentation is uniquely labeled: No
    • Ground truth is uniquely labeled: No
    • Performance measure(s):
      • Region-based / Dice
      • Region-based / Jaccard Coefficent
      • Region-based / Rand Index
  9. Add Collapse Collection ( Galaxy version 5.1.0) from the list of tools with the following recommended parameters:
    • Prepend File name: Yes
    • Where to add dataset name: Same line and only once per dataset
  10. Connect the following inputs:
    • Connect the output of 3: Sigma values to test to the “File to split” input of 4: Split by group.
    • Connect the output of 4: Split by group to the “Object ID” input of 5: Input file containing parameter to parse out of.
    • Connect the output of 5: Input file containing parameter to parse out of to the “Sigma” input of 6: Filter 2-D image.
    • Connect the output of 2: Dataset to perform segmentation to the “Input Image” input of 6: Filter 2-D image.
    • Connect the output of 6: Filter 2-D image to the “Input Image” input of 7: Perform histogram equalization.
    • Connect the output of 7: Perform histogram equalization to the “Input Image” input of 8: Threshold image.
    • Connect the output of 8: Threshold image to the “Segmented images” input of 9: Compute image segmentation and object detection performance measures.
    • Connect the output of 2: Input Ground Truth to the “Ground truth images” input of 9: Compute image segmentation and object detection performance measures.
    • Connect the output of 9: Compute image segmentation and object detection performance measures to the “Collection of files to collapse into single dataset” input of 10: Collapse Collection.

This is how the workflow should look like!

workflow_overview.png.

In this workflow, the Parse parameter value tool is passing sigma values one by one to the workflow downstream. In this way, a collection of datasets is created for each sigma value and each one is processed individually. Such strategy can be adapted for any parameter of the workflow and, in general, for any tool.

Coming back to the results, here are the results from segmentation with four different sigmas…

sigma_comparisons.png.

…and below we can see the output of the analysis:

output_workflow.png.

We are comparing three different performance measures region-based segmentation Dice, Jaccard and Rand. All three indexes measure the agreement of the predicted segmentation image foreground and the ground truth image foreground

Overall, the results suggest that as the sigma value increases too much, the coefficient values decrease. Lower sigma values result in segmentations that more closely match the ground truth, with the highest Dice coefficient of 0.74 achieved at a sigma value of 5.

Conclusions

In this tutorial, we demonstrate how to test different sigma values to evaluate segmentation with a ground truth image.

Overall, the Parse parameter value tool can be easily reused for testing other parameters and applied to different workflows.