Pipelines

Pipelines define the sequence of processing steps applied to each image. EVAnalyzer supports multiple pipelines per project; each pipeline is executed independently and can process a different image channel.

Creating a Pipeline

In the Pipelines tab, click:

• New pipeline — start with an empty pipeline.
• The arrow beside New pipeline — choose a preset template to load a pre-configured set of steps.

Available presets include:

• EV channel — optimised for extracellular vesicle quantification in single-vesicle imaging with low background.
• Cell brightfield — optimised for cell segmentation in brightfield images.
• Nucleus — optimised for nucleus segmentation from fluorescent labelling (Hoechst, DAPI).
• EV in cell — optimised for EV quantification inside cells.

Pipeline Editor

Click a pipeline name to open the pipeline editor.

Pipeline settings

Setting	Description
Pipeline name	Human-readable label; use distinct names to simplify troubleshooting
Enabled	Disabled pipelines are skipped during analysis

Pipeline input

Setting	Description
Image channel	The channel index (0-based) to use as the starting image
Z-projection	Which projection mode to use when the project Z-stack setting is Intensity Projection
Z index	Which Z plane to use when the Z-stack setting is Exact one
T index	Which time frame to use
Empty input	Start with a blank image (useful for pure object-manipulation pipelines)

Pipeline steps

Steps are listed top-to-bottom and executed in that order. Click + Add step (the — + — button) to open the command picker, which shows only commands compatible with the current pipeline state.

Commands fall into categories indicated by colour:

• Grey — image processing (input: image, output: image)
• White — segmentation (input: image, output: binary mask)
• Green — object operations (input: objects, output: objects or measurements)

A typical pipeline flow:

1. Image processing commands reduce noise and enhance the signal (Rolling Ball, Gaussian Blur, …).
2. Segmentation commands separate foreground from background (Threshold → Connected Components → Watershed).
3. Extract ROIs converts the binary mask into a set of segmentation-class objects.
4. Classify ROIs applies size/circularity filters and assigns an object class.
5. Object processing commands perform further analysis (Colocalization, Distance Transform, …).

Live preview

The viewport on the right shows the result of all pipeline steps applied to the currently selected image. Changing any parameter immediately updates the preview. A live object count is shown in the legend.

Use the zoom controls to inspect segmentation quality, and the side-by-side button to compare the original and processed image simultaneously.

Pipeline History

Every settings change is recorded. Click History to open the change log (last 64 changes). Double-click any entry to restore that state. Click Tag to mark the current state so it is easy to find later.

Running the Analysis

Once all pipelines are configured, click Play (▶) in the toolbar to start the analysis.

• A progress dialog shows per-image and per-pipeline progress.
• Click Stop to interrupt; in-progress tasks will finish before halting.
• Click Open results folder to locate the output files immediately.

Results are written to:

<image_directory>/evanalyzer/<job_name>/results.evadb

Best Practices

• Add one pipeline per image channel you want to analyse.
• Add separate pipelines for object-processing steps (colocalization, in-cell counting) that operate on already-extracted objects.
• Use the Tag history feature to bookmark a known-good state before experimenting with parameters.