Interactive Workflow
The following workflow and online showcase is an example of how arkitekt enables an interactive workflow, spanning multiple bioimage apps and deep learning enabled denoising. With this workflow we would like to demonstrate that by leveraging a tight and timely interaction between the apps, users can get immediate feedback on the results of their actions.
Here we set out to use Arkitekt to build a workflow that allows users to interactively denoise and segment cells in regions of interest in a 3D image. The workflow is composed of three apps:
-
Napari a python based 3D volume viewer that will allow the users to interactively annotate regions of interest in an image, while eeasily exploring the 3D volume, as well as the app that will be used to display the results of the workflow.
-
Care Denoising, a deep learning based denoising app that will be used to denoise the regions of interest selected by the user in Napari.
-
Fiji (ImageJ), and its "Analyze Particles" plugin, that will be used to segment the denoised regions of interest.
Why?
The main motivation for showcasing this workflow, is the core believe that interactive workflows are a key component of the current and future bioimage analysis landscape. We believe that the ability to interactively explore and annotate data, and to get immediate feedback on the results of the analysis, is a key necessity for the current analysis where human insight is determing where we look for when analysis our data. And while automation will continue to grow in importance, we believe that the human in the loop will, continue to play a key component of the analtical process.
Prerequisites
Well lets jump right in and see how we can run this workflow. You could of course build this workflow from scratch, but we have already done that for you, and you can easily import it into your arkitekt instance. To do so, please connect this website to your arkitekt instance and then simply click on the "import" button below, which will import it directly into your arkitekt instance.
If you havent installed any of the apps yet, your Arkitekt instance will tell you that you are missing some apps implementing the nodes. So of course make sure you install the apps that are missing. In this case you will need to install the following apps:
Napari
With the mikro-napari plugin, which you can install easily via the napari plugin manager.
or via the command line with pip install mikro-napari
Kare Denoising
An Arkitekt plugin that allows you to denoise images using deep learning. You can install it through the PluginStore or when connected, by clicking on the button below.
Fiji
A ImageJ instance with the "Analyze Particles" plugin installed (comes preinstalled with Fiji)
MikroJ
The Arkitekt enabled FIJI wrapper app that allows your Fiji instance to be connected to Arkitekt. Please install the latest version of MikroJ.
Data Preparation
This workflow was designed to work with the Tribolium Dataset. However the examplary data used in this workflow is only on noisy stack of the Tribolium dataset, which you can download here.
The Image Data was converted through the Convert Omero
Node provided by
Model Preparation
For this workflow you can either use a pretrained model, or train your own model. If you want to train your own model, you can follow the instructions in the Deep Learning Training Tutorial to train your own model. If you want to use a pretrained model, you can upload the pretrained model directly from the link below.
This model was trained on this dataset here.
The Workflow
Once you have installed all the apps, you can simply import the workflow by connecting your arkitekt instance and clicking on the button below.
Once connected this pane will also give you an overview over the apps that still need to be installed, nodes in green have been installed, nodes in yellow still lack an implementation. Of course feel free to explore this graph more interactively in your orkestrator interface
Running the Workflow
For the execution of the Workflow you will need to start the following apps:
- Napari
- Kare Denoising (through deployment of the kare plugin)
- Fiji (through MikroJ)
Once you have started all the apps, you can run the workflow directly from an image uploaded to your arkitekt instance. To do so, simply start the workflow and select the image the converted you want to run the workflow on.
Test Environment
This workflow was tested on the following environments:
Analysis Computer
- Intel Core i9-8700K CPU @ 3.70GHz
- Ubuntu 22.04
- Arkitekt Paper Channel
- Segmentor Plugin 0.3.1
- Stdlib Plugin 0.3.6
Analysis Computer
- Intel Core i9-8700K CPU @ 3.70GHz
- Windows 10
- Arkitekt Paper Channel
- Segmentor Plugin 0.3.1
- Stdlib Plugin 0.3.6