StarDist

The content of this page has not been vetted since shifting away from MediaWiki. If you’d like to help, check out the how to help guide!

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 need any help, please first take a look at the StarDist documentation and frequently asked questions (FAQ). If that doesn’t solve your issue, you can browse the existing stardist-tagged forum posts or search the image.sc forum. If you can’t find what you’re looking for, please create a new topic at the forum (and use the tag stardist). If you open a new topic, please provide a clear and concise description to understand and ideally reproduce the issue you’re having. Please only file an issue here for bug reports or if you have technical questions regarding the plugin.

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

Installation

Start Fiji (or download and install it from here first).
Select Help > Update... from the menu bar.
Click on the button Manage update sites.
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.)
Click on Apply changes to install the plugin.
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[1]:

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.