- Guide
- Imaging
- Image analysis pipeline steps
Image Analysis Pipeline Steps
An image analysis runs a pipeline, which is a list of steps that each read what an earlier step produced. Add step lists every step, and this page is that full list.
Adding a Step
Add step at the foot of the pipeline puts the step you pick at the end, open so you can set it. The grip beside any step offers Add step above and Add step below, which put it at that place instead.
Either way the menu lists every step under six headings in this order. They are Preprocess, Segment, Measure, Relate, Gate, Aggregate.
A step that reads something the steps above it don't produce yet is grayed out, with a short note beside its name, such as "Needs a measure". Point at it, or tap it on a phone, and the panel at the foot of the menu says what to add first. It becomes available once that step is in.
The search box at the top of the menu narrows the list, and Enter adds the first step that matches and can go in.
A step's input is picked at the top of its settings, and what the choice lists depends on what the step reads.
- Where a step reads an image, the choice lists the image itself and any plane an earlier cleanup produced.
- A LIF, or another acquisition that stores a plane per channel, adds each channel by the name the microscope recorded.
- A JPEG, PNG or flattened TIFF is a single picture, so its list holds the image and the cleanups alone.
- Where a step reads objects, the choice lists each step that found some.
An input with one candidate is bound for you and shows no picker.
The tables below name each step's settings and leave the explaining to the app, which does it where you set them.
- The info button beside a setting's name covers that setting.
- How it works at the top of an open step covers the step.
Tune an image analysis covers the settings you reach for most, among them Threshold strategy, Expect threshold between, Match by and Overlap thresholds.
The two steps that detect with an AI model have an AI tag in the menu and count toward your team's monthly usage, as Detecting with an AI model describes.
What Each Step Does
Preprocess
Cleans up an image before anything reads it. A cleanup added above a detection becomes the image that detection reads, and a measure keeps reading the image as it was acquired until you point it at the cleaned plane.
| Step | What it reads | Its settings |
|---|---|---|
| Trim the image edges | The image as stored | Top, Bottom, Left, Right, each in px |
| Remove speckle noise | An image | Smoothing radius |
| Smooth away noise | An image | Blur radius |
| Even out the background | An image | Background level, Background offset, Expect background offset between |
Trim the image edges leaves a band along each edge out of the analysis, such as the data bar along the bottom of an SEM image. Every later step reads only what's kept, including a measure that reads the image as acquired, so bright text on a dark strip stays out of an automatic threshold, and Area covered and Total pixels cover the kept region. The stored file stays as it was.
It works differently from the other steps.
- It is always first, and a pipeline holds one.
- It has no image to choose and no Try a range.
- A detector backed by an AI model still detects on the whole image, then drops any object whose center falls in the trimmed band.
- Results with one row per image gain Analyzed width (px) and Analyzed height (px).
Segment
Finds objects, or reshapes objects an earlier step found.
| Step | What it reads | Its settings |
|---|---|---|
| Find objects by brightness cutoff | An image | Measure the, Threshold, Minimum region size, Threshold strategy, Expect threshold between |
| Find colonies on a plate | An image as acquired | Detection sensitivity, Overlap between objects |
| Find fluorescent cells or nuclei | An image as acquired | Detection sensitivity, Overlap between objects |
| Split touching objects | Objects found earlier | Split depth, Closest two centers |
| Clean up object shapes | Objects found earlier | Shape cleanup, Cleanup radius, Cleanup passes |
Measure
Turns pixels into numbers. Every measure reads an image, so each one has the same channel choice. The first four can measure the frame as a whole, which is why they are offered before anything has found objects.
| Step | What it measures over | Its settings |
|---|---|---|
| Score overlap between two markers | The whole image or objects found earlier | Marker A channel, Marker B channel, Overlap thresholds, Marker A cutoff, Marker B cutoff |
| Measure how much area is covered | The whole image or objects found earlier | Measure the, Threshold, Threshold strategy, Expect threshold between |
| Measure marker brightness | The whole image or objects found earlier | Nothing to adjust |
| Sort objects into positive and negative | The whole image or objects found earlier | Positive cutoff |
| Sort objects by color | Objects found earlier | Nothing to adjust |
| Measure each object's shape | Objects found earlier | Nothing to adjust |
Measure each object's shape records each object's area, perimeter, equivalent circular diameter (ECD), circularity, solidity, convexity, extent, aspect ratio and eccentricity. Eccentricity runs from 0 for a round object to near 1 for a needle, and convexity falls below 1 for a notched or ragged outline. On an image with a scale it adds Area (µm²), Perimeter (µm) and ECD (µm) beside the pixel values.
Sort objects by color reads the color of each object's pixels, so it needs a color (RGB) image. The menu offers it wherever objects exist, which includes a single-channel fluorescence image, where there is no color to sort on.
Relate
Works out which objects sit inside which.
| Step | What it reads | Its settings |
|---|---|---|
| Match objects to the region that holds them | Two sets of objects, the containing regions and the objects inside them | Match by, Smallest share inside |
An object that sits outside every region isn't counted toward any of them. When most of the objects fall outside, Checks warns, which usually means the regions are drawn too tight or on the wrong channel.
Gate
Drops objects from everything below it.
| Step | What it reads | Its settings |
|---|---|---|
| Keep only objects in the chosen range | Objects found earlier, and a measure of them | A measurement, a comparison, and the number or numbers it compares against |
Its settings read as one sentence under Keep objects whose.
- The first box lists every measurement of an object under Shape, Brightness and Color, such as Circularity, Aspect ratio or mean brightness.
- The second sets the comparison. at least and at most take one number, between takes two, and one of takes a list of values.
Picking a measurement the pipeline doesn't record yet adds the measure step that records it, directly above. That is why adding this step to a pipeline with no measure also adds Measure each object's shape.
Filter under the image does the same job with a handle you drag, described under Filtering what counts.
Aggregate
Turns what the pipeline found into the rows of the results table.
| Step | What it reads | Its settings |
|---|---|---|
| Count what was found | Objects found earlier | Nothing to adjust |
| Count objects inside each region | A match step | Nothing to adjust |
| Summarize the measurements | A measure that ran over objects | Nothing to adjust |
| Check controls | Every image's finished number | The conditions to check and the range each should fall in |
Check controls always goes last, and a pipeline holds one. Its ranges, and what a run's control images came to, are covered under Controls.
Settings That Add a Step
Four settings on the counting analysis types each add or remove a whole step. Count objects has all four, and Count touching objects has Background cleanup and Reduce noise.
| Setting | What it adds |
|---|---|
| Background cleanup | Even out the background, above the detection |
| Split touching objects | Split touching objects, below the detection |
| Fill holes | Clean up object shapes, set to fill interior gaps, between the detection and the split |
| Reduce noise | Remove speckle noise, at the top of the pipeline |
Turning one off takes its step back out. Open the step it added to reach that step's own settings, such as Cleanup radius on the shape cleanup, and to point it at a different image or set of objects.