Generating a spatial biology image is only half the job. The other half is turning pixels into numbers: cell boundaries, marker intensities, cell types, neighbourhoods and spatial statistics. For many UK labs, image analysis is the bottleneck. A single whole-slide multiplex image can contain billions of pixels, and manually counting cells is neither practical nor reproducible. This guide explains the core tasks and the software tools most widely used in UK research settings in 2026.
1. The core tasks in spatial image analysis
Most spatial-biology image-analysis workflows follow the same sequence, even if the software differs:
- Pre-processing: illumination correction, background subtraction, stitching of tiled images, registration across staining cycles.
- Segmentation: defining cell or nucleus boundaries. This is the most critical step because every downstream measurement depends on it.
- Feature extraction: measuring marker intensity, area, shape and texture for each segmented object.
- Cell typing: classifying cells based on marker combinations, for example CD8+ T cells, CD68+ macrophages or tumour cells.
- Spatial analysis: calculating nearest-neighbour distances, clustering, co-occurrence and tissue-region interactions.
- Visualisation and export: overlays, heatmaps and tables ready for statistical testing or integration with other omics data.
2. Open-source tools UK labs use
QuPath
QuPath is the de facto standard for digital pathology and multiplex IF analysis in the UK. It handles whole-slide images, provides annotation tools, cell detection and classification, and supports scripting in Groovy for custom workflows. Bankhead et al. introduced QuPath as open-source software for digital pathology image analysis, and it has since become a core tool for multiplexed tissue projects. The software is particularly strong for H&E and IF overlays, region-of-interest analysis and exporting single-cell measurement tables.
CellProfiler
CellProfiler is a modular, high-content image-analysis pipeline built for biologists who need reproducible batch processing. It excels at object identification and feature measurement in multi-well plates and tiled images. McQuin et al. described CellProfiler 3.0 as the next generation of image processing for biology, with a pipeline-based interface that makes it easy to document and share workflows.
Fiji / ImageJ
Fiji is the Swiss army knife of biological image analysis. It is not specifically designed for spatial biology, but its plugin ecosystem makes it invaluable for custom preprocessing, registration and quantification. Schindelin et al. described Fiji as an open-source platform for biological-image analysis, and it remains the fallback tool when no off-the-shelf solution exists.
StarDist and Cytokit
StarDist uses deep-learning-based star-convex polygons for nucleus and cell segmentation, often outperforming classical thresholding in dense tissues. Czech et al. developed Cytokit as a single-cell analysis toolkit for high-dimensional fluorescent microscopy imaging, providing an end-to-end pipeline from raw images to single-cell data.
3. Commercial platforms
| Platform | Strengths | Typical use case | Notes |
|---|---|---|---|
| QuPath | Free, whole-slide IF, cell typing, scripting | Digital pathology, multiplex IF, TME studies | Open source; very large user community in UK |
| CellProfiler | Batch pipelines, reproducibility, modularity | High-content screens, tiled tissue quantification | Open source; good for non-pathologists |
| HALO (Indica Labs) | Whole-slide analysis, certified for clinical research | Clinical translational studies, pharma workflows | Commercial; per-seat licensing |
| Visiopharm Phenoplex | Multiplex spatial biology workflow, cell typing, neighbourhoods | Cancer microenvironment, immuno-oncology | Commercial; built for spatial biology |
| Imaris / Aivia | 3D visualisation, object tracking, segmentation | Confocal/organoid 3D spatial analysis | Commercial; strong for volumetric data |
4. Segmentation is the make-or-break step
Spatial analysis is only as good as the segmentation. Over-segmentation splits one cell into several objects; under-segmentation merges touching cells. Both introduce false cell types and false neighbourhoods. For this reason, labs often validate segmentation by overlaying outlines on the original image and spot-checking across tissue regions. Deep-learning methods such as StarDist reduce but do not eliminate this problem, especially in dense tumour regions or tissues with variable nuclear morphology.
Analysing EVOS S1000 OME-TIFF outputs
The EVOS S1000 Spatial Imaging System exports spectrally unmixed whole-slide images as OME-TIFF files. These files open directly in QuPath, where each channel becomes an image layer for cell detection and phenotyping. Because the S1000 captures up to 9 channels in a single round, a QuPath project can contain all protein markers — whether from directly conjugated Alexa Fluor primaries or from Aluora signal amplification — in one project, avoiding the registration errors that occur when stitching images from multiple staining rounds.
For labs mixing S1000 protein images with 10x Genomics Xenium or Akoya PhenoCycler transcriptomic data, the standard OME-TIFF format also makes it easier to align adjacent sections using morphological landmarks such as DAPI and pan-cytokeratin. CellProfiler can read the same TIFFs for batch feature extraction, while Python libraries such as squidpy can compute neighbourhoods and co-occurrence across protein and RNA tables.
5. Spatial statistics move beyond counting
Once cells are classified, the interesting questions are spatial. How close are cytotoxic T cells to tumour cells? Are macrophages clustered or dispersed? Which cell types co-localise more than random chance would predict? Tools such as the spatial statistics functions in QuPath, R packages like spatstat, and Python libraries such as squidpy can compute Ripley’s K, nearest-neighbour distances and neighbourhood enrichment. The goal is to move from descriptive counts to testable hypotheses about tissue organisation.
6. Data formats and interoperability
Large spatial images are usually stored as OME-TIFF or, increasingly, Zarr arrays. The recently introduced SpatialData framework by Marconato et al. provides an open, universal data framework for spatial omics, designed to unify images, labels, shapes and tables from different platforms. For UK labs running multiple instruments, adopting a common format early saves enormous integration effort later.
7. Practical workflow tips for UK labs
- Keep raw images safe: raw microscope files are the only record that can be re-analysed as methods improve.
- Version your analysis pipeline: store QuPath projects, CellProfiler pipelines and R/Python scripts in a version-controlled repository.
- Use positive and negative controls: these help set intensity thresholds and validate cell-type calls.
- Plan compute and storage: a single whole-slide multiplex image can exceed 10 GB; spatial transcriptomics datasets can reach hundreds of gigabytes.
Video: TIA centre seminar on scaling multiplexed protein imaging and image analysis.
Video: Overview of next-generation single-cell multiplexed spatial proteomics and analysis.