Analyze Images
Imaging turns microscopy into numbers: counting cells, measuring how much area a marker covers, reading fluorescence intensity, and quantifying colocalization, all in the browser and without writing code.
Upload your images, answer a couple of questions about what you're measuring, and Conspecta assembles a pipeline you can tune on one image and then run across the whole set. The numbers land in a data table you can chart and drop into a figure.
The workspace has two tabs. Images holds your image collections. Analysis holds the runs you've built from them.
Getting Images In
Images live in collections. A collection is the set of images you want to analyze together: a plate, a timepoint, a condition.
Upload images asks for the files, then where they should go, either a new collection you name or one you already have. You can also drag files onto an open collection, or onto the empty workspace to start one.
Conspecta accepts TIFF, PNG, JPEG, BMP, and LIF (Leica Image File) files, up to 200 files at a time, 100 MB per ordinary image, and 5 GB per LIF. Files it can't take are listed individually with the reason, so one bad file never blocks the batch.
LIF Files
A LIF is read in the browser as it uploads, without loading the whole thing into memory, and every series inside it becomes an image with its z-slices, channels, and timepoints intact. The original LIF isn't stored. What you get is the imagery, which is what you analyze.
Keep the tab open while a large upload runs. LIF files are parsed and uploaded from your browser, so closing the tab cancels what's left. The collection header shows live progress and lets you retry an individual file that failed.
Inside a Collection
A collection shows its images as thumbnails or as a table with dimensions, slice and channel counts, file size, and upload date. A z-stack carries a badge like 10z · 3ch so you can see its shape without opening it.
Click the collection's title to rename it, and Add a description… under the title to record what the set is (the experiment, the prep, the condition), so the context travels with the images. Press Enter to save.
A collection holding series from a LIF or CZI gets a Group control in its toolbar. Source file buckets every series under the filename it came from, with anything you uploaded on its own under Standalone images — the answer to "my one LIF became forty rows". A collection of ordinary images has one bucket, so the control stays out of the way.
Select images to Rename, Move them to another collection, or Remove them. Removing an image that other analysis depends on warns you first.
Analyze collection starts an analysis with these images already chosen.
Building an Analysis
New analysis doesn't ask you to pick an algorithm. It asks what you're looking at, in plain language, and builds the pipeline from your answers.
The first question is What are you analyzing?
- Individual objects for cells, nuclei, colonies, particles
- A whole region or image for coverage, overall brightness, or overlap
From there the questions narrow. Counting objects asks whether each container needs its own count, whether you want all of them or only the marker-positive ones, and what your images actually show: colonies on a plate, fluorescent cells or nuclei, or simple light or dark shapes. Measuring a whole region asks whether you want area covered, overall signal, the positive fraction, or how much two markers overlap.
The question about what your images show is where AI detection comes in. Colonies and fluorescent cells are detected with a model. Simple light or dark shapes is the fallback for everything else — stained tissue, particles, spots — and it runs on your own machine with an adjustable cutoff, so it's the one to pick when your objects already stand out clearly from their background.
At the end you get a review screen naming the analysis Conspecta built, a diagram of its steps, a title to type, and a choice of private or shared. Then you pick your images.
The eight analysis types it can assemble are Count objects, Count marker-positive objects, Count objects inside each region, Measure intensity inside each object, Marker intensity, Marker classification, Area coverage, and Colocalization.
Tip: Colocalization and marker classification need real acquisition channels, so they work on a LIF and not on a flattened RGB screenshot. If the option is disabled, that's why, and the message says so.
Tuning and Running
An analysis opens on one image with its pipeline in the sidebar, and the workspace tells you which stage you're at:
- Tune on this image while you adjust settings
- Quality-check the preview to make sure it found what you meant
- Scale to every image once it looks right
- Review result quality
Open any step in the sidebar to see its settings. Changing one re-runs the pipeline live on the image in front of you, so tuning is a conversation with the picture rather than a guess followed by a wait. The settings are named for what they do: whether to measure bright or dark areas, an automatic or fixed threshold, a minimum object size, whether to split touching objects, fill holes, clean up background, or reduce noise.
Under the image, a filter band trims what counts by size, confidence, or brightness, plus color. Filtering happens on results you already have, so narrowing a size range uses nothing and takes effect immediately.
Each step reports its own Checks, which is where a step tells you it found nothing, or found suspiciously many things, before you scale that mistake across 200 images.
When the preview is right, Run on all N images. Conspecta gives you a time estimate up front, shows progress as it goes, and offers to notify you in the browser when it finishes.
Z-Stacks
A z-stack uses as much as its number of slices, so Conspecta checks which slices are actually in focus and offers to run only those. The Depth rail tells you what it found in plain words, such as "Slices 5–17 in focus · 13 of 40 will run", and you can accept it, widen it, use every slice, or reset to the suggestion.
Across a whole set, Choose which slices lets you decide once: detect focus per stack, or apply one range everywhere. The focus check itself runs on your own machine and uses nothing.
Results
Open results switches the center of the workspace to what the run produced, in four views:
| View | What it shows |
|---|---|
| Objects | Every detected object as a cut-out, across the set |
| Images | One card per image with its numbers |
| Table | The measurements as rows and columns |
| Chart | A quick chart of the results |
Images that couldn't be analyzed are named in a band rather than silently dropped, and with more than one image contributing you can filter down to one.
Every run writes its numbers into a data table automatically. You don't have to export anything to keep them. The table appears in Data Tables named after the analysis, and it's what the Chart view and the figure picker read from, so results are one click from a chart and two from a figure.
Tip: Reopening a finished analysis restores its results without re-running anything, which uses no compute usage. The object cut-outs are the exception, since masks aren't stored, but the table, cards, and chart are all there.
Comparing Runs
One run tells you what one condition did. The question an experiment actually asks is treated against control, and Compare answers it by putting the results of two or more runs side by side in one data table.
Select several runs on the Imaging landing and choose Compare, use the Compare button in the landing header, or click Compare at the top of an open analysis. In the dialog box, pick the runs to include. The filter icon at the right end of the search box narrows a long list by status, who created the run, or how recently it was edited, and never hides a run you already picked. A run that can't join your picks stays visible but dimmed, with the reason under its name, so a count is never mixed with a measurement in different units. Then choose which shared measures to compare and check the preview, which shows each group's image count and average before anything is created. Create data table lands you in Data Tables on a table with one row per image and a Group column naming each row's run, with the chart already open.
Group by switches the comparison from one group per run to Condition, Timepoint, or Replicate. Those groupings need the experiment design recorded on the images, so Label images, in that same step, opens a grid of every image in the comparison. Type the labels in, or fill the whole grid at once from the filenames: under Fill from filenames, retype one of your names with {condition}, {replicate}, or {timepoint} where each label sits, and {*} for a part you don't need, such as a plate id or a date. A pattern of {timepoint}_{condition}_r{replicate} reads t01_control_r1.tif. As you type, the line under the field says how many of your filenames the pattern fits and spells out what it pulls from one of them, so you can see it works before Apply pattern fills the grid.
Every cell stays editable before you save, and Sort order is the number a timepoint sorts by, so t9 comes before t10 instead of after it. Filenames that don't fit the pattern keep the labels they had, and the dialog box says how many were skipped. Saving turns on the grouping you needed.
A timepoint comparison keeps the course in its stored order, so a growth or closure curve reads left to right. Images without the chosen label gather in an Unlabeled group rather than disappearing, and if the compared images were acquired at different pixel sizes a warning appears above the preview before anything is created.
The comparison stays live the same way an embedded table does. If a source run's numbers change, the table flags the difference and a refresh pulls the new values. For error bars or a significance test, add the chart to a figure and use its Comparisons step, or run Compare groups from the table's Analyze menu.
Keeping Track
An analysis carries a status through its life, from Draft to Approved or Archived, and can be commented on and signed off like a notebook page. Collections and analysis are each either private to you or shared with the project, set when you create them and changeable afterwards.
Folders, stars, sorting, and filtering work the same as elsewhere in the app.