ImageJ

StarDist (ImageJ)
Author	Uwe Schmidt, Martin Weigert
Update site	StarDist
Maintainer	Uwe Schmidt, Martin Weigert
Source	on GitHub
Website	https://github.com/mpicbg-csbd/stardist

Overview

This is the ImageJ/Fiji plugin for StarDist, a cell/nuclei detection method for microscopy images with star-convex shape priors. The plugin can be used to apply already trained models to new images. See the main repository for links to our publications and the full-featured Python package that can also be used to train new models. If you encounter problems with the plugin, please file an issue here.

The plugin currently only supports 2D image and time lapse data. If you have 3D data, please use our python library.

Installation

1. Start Fiji (or download and install it from here first).
2. Select Help > Update... from the menu bar.
3. Click on the button Manage update sites.
4. Scroll down the list and tick the checkboxes for update sites CSBDeep and StarDist, then click the Close button.
   (If StarDist is missing, click Update URLs to refresh the list of update sites.)
   
5. Click on Apply changes to install the plugin.
6. Restart Fiji.

Usage

Plugin

Open the image that should be segmented. Note, that currently only 2D and 2D+time images are supported. Suitable test images can for instance be found at the Broad Bioimage Benchmark Collection:

Start the plugin from Plugins > StarDist > StarDist 2D. The following parameters can be set:

	Select a neural network model from the dropdown list, which can be one of the following: A built-in model. We currently provide: Versatile (fluorescent nuclei) and DSB 2018 (from StarDist 2D paper) that were both trained on a subset of the DSB 2018 nuclei segmentation challenge dataset. Versatile (H&E nuclei) that was trained on images from the MoNuSeg 2018 training data and the TCGA archive. A custom user-trained model (via the training code) that has been exported as a zip file and can be loaded from a file or URL (see Advanced options below). If necessary, one can change/disable the percentile-based input image normalization.
	Adjust the NMS (non-maximum suppression) postprocessing parameters: Probability/Score Threshold - higher values lead to fewer segmented objects, but will likely avoid false positives. Overlap Threshold - higher values allow segmented objects to overlap substantially. If in doubt, load the default NMS parameters of the selected built-in model (see below). The segmented objects can be returned as a Label Image or in the ROI Manager (or both).
	Advanced options: Specify a user-trained model file or URL Increase the number of tiles (in case of GPU memory limitations/errors, i.e. for larger images) Load default NMS parameters for the selected built-in model. Restore all default parameters.

Example of running the plugin, showing the returned label image and ROIs overlaid on the input image (check Show All in the ROI Manager):

Scripting/Batch-Processing

Please have a look at the Fiji/Jython script batch-processing example that runs stardist on all files of a folder.

Citation

Please cite the paper if you are using the plugin in your research:

• Uwe Schmidt, Martin Weigert, Coleman Broaddus, and Gene Myers. Cell Detection with Star-convex Polygons. International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), Granada, Spain, September 2018.