Written and maintained by CASRAI Editorial Board
Last updated
QuPath is free, open-source software for analysing digital pathology and other bioimages, especially whole slide images. This guide is for researchers who need to know what QuPath does, which slide formats it opens, how a basic cell detection run works, whether it is deep learning, and how to cite it. Facts below come from the QuPath website and documentation (checked 2026-11-07) and describe release v0.7.0. QuPath is not intended for clinical, diagnostic or therapeutic purposes; it is research software.
What QuPath is
QuPath describes itself as open software for bioimage analysis and is released under a GPL licence. Its documentation says it is widely used in research digital pathology, particularly for whole slide images. Core capabilities listed in the documentation: annotation and visualisation in a JavaFX user interface, built-in cell and tissue detection, interactive machine learning for object and pixel classification, compatibility with ImageJ, OpenCV, JTS and OMERO, image reading through Bio-Formats and OpenSlide, Groovy scripting, and extensions.
Who runs it and what it covers
QuPath is an open-source project with its code and releases on GitHub, a documentation site on Read the Docs, and community discussion on the Image.sc Forum. The project website does not name a single operating institution, so this guide does not either. It is a tool, not a data resource: it analyses images you bring, and does not host a dataset.
Installers on the GitHub Releases page cover Windows (MSI installer or a portable download), macOS (Intel; Apple silicon is marked beta) and Linux (tar.xz). The current release on the website is v0.7.0.
How do I download QuPath?
Go to the QuPath website, follow the download link to the GitHub Releases page, and choose the installer for your operating system listed above. The documentation has an Installation page with platform-specific notes, including that QuPath is not currently a signed application, so Windows and macOS may show security warnings. Record the version you use, since methods text should state it.
Which image formats does QuPath open?
QuPath reads images through two libraries, and the choice matters for what opens. Bio-Formats supports a wider range of formats and multidimensional data. OpenSlide is faster but handles only 2D, 8-bit RGB images. QuPath is designed for pyramidal images. Format notes from the documentation’s formats page:
| Format | What the documentation says |
|---|---|
| BIF | Readable through both Bio-Formats and OpenSlide. |
| CZI | Read through Bio-Formats, with caveats around JPEG-XR compression and Apple silicon. |
| DICOM | Bio-Formats since v0.4.0; OpenSlide since v0.5.0. |
| iSyntax (Philips) | Not compatible. |
| MRXS | Support is limited. |
| NDPI | Generally works well. |
| TIFF | Varies by file; OME-TIFF is preferred. |
| OME-Zarr | Read and write support since v0.6.0. |
For images on an external server, the documentation points to File > Open URI, and OMERO integration received a major update in v0.6.0. Community extensions for services such as Google Cloud and SlideScore are described as untested and unsupported. If a vendor file will not open, see opening and converting proprietary microscope files and OME-TIFF vs OME-Zarr vs TIFF. For how slides become digital images in the first place, see digital pathology scanners.
How to do cell detection in QuPath
The official cell detection tutorial uses a stained tissue image. Its sequence:
- Draw an annotation around the region to analyse.
- Choose Analyze > Cell detection > Positive cell detection.
- Set the score compartment, for example “Nucleus: DAB OD mean” in the tutorial’s example.
- Tick “Single threshold”.
- Click Run. The tutorial’s example returned 1088 detections, 12.4% of them positive; your numbers will differ.
- Review results through Measure > Show detection measurements or Show annotation measurements.
- If staining colours are poorly separated, use “Estimate stain vectors” and rerun the detection.
The tutorial is a worked example for brightfield staining; the tutorials section of the documentation covers other workflows.
Is QuPath deep learning?
Partly, through extensions. QuPath’s core description lists interactive machine learning for object and pixel classification, meaning you label examples and train a classifier inside the program, plus built-in cell and tissue detection. The cell detection tutorial above does not use deep learning. However, the QuPath documentation has a separate Deep learning section covering the Deep Java Library (DJL) extension, the Bioimage Model Zoo extension with InstanSeg, a StarDist extension, and WSInfer. The StarDist page describes StarDist as a deep-learning-based method of 2D and 3D nucleus detection that runs in QuPath as an alternative method of cell detection once its extension is added. So the answer is: the built-in tools are classical and interactive machine learning, and deep learning models are available as add-ons whose own documentation describes what they do.
How to reference QuPath
The citation page asks users to cite: Bankhead P. et al., “QuPath: Open source software for digital pathology image analysis”, Scientific Reports (2017), doi 10.1038/s41598-017-17204-5. An earlier bioRxiv version has doi 10.1101/099796. The page notes that new QuPath papers will appear and tells readers to consult it to find the correct paper to cite, so check the citation page before submitting a manuscript. It gives no version-specific citation advice; stating the version in your methods is an editorial suggestion, not a QuPath requirement. For the field context, see what is pathology.
Strengths and limits
- Strengths (from the documentation): free and open source; built for whole slide images; reads many formats through Bio-Formats and OpenSlide; scriptable with Groovy; extensible, including deep learning extensions; works with ImageJ and OMERO.
- Limits: research use only, not clinical or diagnostic; Philips iSyntax is incompatible; MRXS support is limited; TIFF behaviour varies; Apple silicon is beta on macOS, with CZI caveats; community extensions for some cloud services are untested.
Alternatives
| Tool | What its own site says |
|---|---|
| CellProfiler | Designed for biologists: load an example pipeline of image-processing modules, adjust settings, process images automatically (even millions), export data to a spreadsheet or database; CellProfiler Analyst adds machine-learning classification. |
| napari | A fast, interactive viewer for multi-dimensional images in Python, with overlays such as points, polygons and segmentations, and annotation of derived datasets. |
These tools have different designs (a pipeline builder, a Python viewer, a whole slide desktop application), so the table describes each rather than ranking them. QuPath’s documentation notes ImageJ compatibility, so ImageJ-based workflows can be combined with it.
FAQ
Is QuPath deep learning?
The core uses interactive machine learning classifiers; deep learning methods such as StarDist, InstanSeg and WSInfer are available through extensions described in the documentation.
How to download QuPath?
From the QuPath website via GitHub Releases: an MSI installer or portable download for Windows, an installer for macOS, tar.xz for Linux.
How to do cell detection on QuPath?
Annotate a region, run Analyze > Cell detection > Positive cell detection, set the score compartment, and review the measurements.
How to reference QuPath?
Cite Bankhead et al., Scientific Reports (2017), doi 10.1038/s41598-017-17204-5, after checking the official citation page.
Sources
- QuPath website (checked 2026-11-07)
- QuPath documentation: About (2026-11-07)
- QuPath documentation: How to cite QuPath (2026-11-07)
- QuPath documentation: Installation (2026-11-07)
- QuPath documentation: Deep learning and StarDist (2026-11-07)
- QuPath documentation: Supported image formats (2026-11-07)
- QuPath tutorial: Cell detection (2026-11-07)
- CellProfiler and napari home pages (checked 2026-11-07)
