Rare-event detection and targeted slide scanning on a ZEISS Axioscan
Faculty of Science, Charles University, Prague
Viničná Microscopy Core Facility · discussed and delivered with Ondřej Šebesta, Deputy Head of VMCF and Head of the Light Microscopy Unit (KONFMI) and Image Analysis Unit
Rare-event detection without scanning the whole slide
The Viničná Microscopy Core Facility at Charles University operates a ZEISS Axioscan for shared life-science imaging. This project added a rare-event detection workflow for experiments where sparse structures are identified on a preview and only the selected positions are acquired at high resolution.
The goal was to detect rare events or structures using both traditional image analysis and machine-learning software. In the client's case, that meant detecting parasites in blood smears without scanning the whole slide, capturing only the parasites at the highest possible resolution. The same principle may be applied to other sparse targets, such as pollens, specific tissue structures, or particular cells. The solution was designed so it would be easy to implement new software, algorithms, and AI models into the pipeline as they become available.
Why the standard Axioscan sequence was not enough
The Axioscan and ZEN Slidescan automate slide scanning end to end, and the native detection reliably finds where a section sits on the slide. It does not identify specific or rare targets within the specimen, and the scan profile runs as one fixed sequence with no native route to connect a more capable detector. For work that depends on locating sparse targets, such as parasites in a blood smear, the operator either scanned the whole slide and searched afterwards, or marked regions by hand.
Because the steps of the scan profile were designed as one end-to-end process, operators either repeated unwanted steps or worked around the process manually. That slowed scanning, introduced inconsistency, and made it harder to keep images, ROI files, metadata, and logs together for reproducible review.
- The scan profile ran as one fixed sequence, so existing previews could not be reused without repeating earlier steps.
- The native tissue detection separates specimen from background. It finds where a section sits on the slide, but cannot pick out specific or rare targets within it, and there was no native way to connect an external analysis tool or model to do that detection.
- A review step was needed between detection and scanning, and the native path does not allow pausing at that point.
- Images, ROI files, metadata, and logs were produced separately, which made it harder to keep a run together for review and reproducibility.
Preview analysis and automated ROI detection
SmartLabs built a workflow layer inside ZEISS ZEN Blue that opens the two constraints of the native path. It lets an analysis tool decide where to scan, and it lets the operator enter the run at the step that fits the session.
The detection itself is done by the tool best suited to the sample, whether ZEN Blue's internal analysis, QuPath, or another model the user connects. What we built is the layer that carries that detection into the scan and then drives acquisition from its result: the analysis returns the regions, and only those are imaged at high resolution. Once a run starts, the workflow keeps focus, exposure, illumination, output structure, and metadata handling consistent, whichever mode the operator chooses.
Three entry points
Depending on what already exists in the session, the operator chooses:
| Mode | When to use it | What it does |
|---|---|---|
| Full Run | Starting fresh, no prior data | Preview, detect ROIs, scan |
| Detect and Scan | Preview already captured | Detect ROIs, scan |
| Scan Only | ROIs already defined externally | Use existing ROIs, scan directly |
ROI generation and import
When ROI detection is needed, the workflow uses the analysis route that fits the sample. For simple brightfield cases that can be ZEN Blue's internal analysis. For more complex sample structures, the workflow exports the preview to QuPath, runs the detection macro, and re-imports the annotations into ZEN as scan ROIs. It is also designed to send images to an external analysis tool and take a set of coordinates back into ZEN, so externally generated ROIs can be used directly for acquisition instead of being drawn by hand inside ZEN.
Guided acquisition
High-resolution scanning is executed only on the targeted ROIs. The workflow keeps the acquisition parameters consistent and leaves operator control where it matters: choosing the workflow mode, selecting the analysis route, and validating the session before scanning.
Data handling
For each run, the workflow automatically creates a structured session folder. Preview images, scan data, ROI files, metadata, and logs are stored together. Acquired images are saved as .CZI, and detection and scan metadata are exported as CSV and JSON for analysis or record keeping.
Technical stack: ZEN, QuPath and AI integrations
What the targeted-scanning workflow enables
| Flexible entry | The operator starts at whichever step fits the session. If preview images, detected ROIs, or previous acquisition data already exist, the workflow continues from that point instead of forcing a full restart. |
|---|---|
| Reuse of existing data | The same sample or image dataset can be analysed several times through different tools or routes to extract different information. One route may detect tissue regions while another identifies fluorescence-positive areas within the same sample. |
| Protocol fit | The workflow adapts to the lab protocol. Existing validated detection logic is reused in acquisition, not replaced. |
| Automated ROI generation | Preview images are processed through the selected analysis route, and detected regions are converted into ROIs for high-resolution scanning. |
| Reproducibility | Imaging parameters are standardized across sessions and operators, so the same protocol can be applied consistently and outputs stay comparable across repeated experiments. |
| Traceability | Acquisition data, ROI files, and metadata are stored together per session. No separate tracking is required. |
| User experience | A guided workflow reduces the training burden and lowers the chance of operator error, while preserving scientific control over analysis choices, ROI selection, and acquisition settings. |