DigitalPathology&Oncology
Whole-slide image analysis with QuPath. Develop and evaluate tumor-region, TIL and IHC analysis workflows with batch pipelines and deep-learning integration.
Research workflow planning. Each intended use requires its own evaluation and approval process.
Pathology labs must turn whole-slide images (1–20 GB per slide) into actionable, quantifiable findings at scale — tumor burden, margins, TIL density, PD-L1 scoring, grading
Plan a reproducible whole-slide analysis pilot
Reviewed October 7, 2026. This engineering guide describes project decisions; it does not establish diagnostic suitability or measured clinical benefit.
Define the analysis unit
Specify the tissue, stain, scanner, region of interest and measurement being studied. Define whether the output is a research feature, an annotation for review or part of a proposed clinical workflow. Tumor area within a selected tissue region is different from tumor area divided by the entire slide, including background. Record the denominator and exclusions in the output contract.
Check images before inference
Inspect focus, folds, pen marks, tissue completeness, stain variation and pixel-size metadata on a representative sample. Confirm that the image reader opens the required formats and resolutions. Keep rejected slides visible in the run report rather than quietly removing difficult cases from the cohort. Test both scanner and stain variation before extrapolating a result.
Keep patients together in evaluation
Hold out cases at the patient or specimen level so related sections and tiles do not cross training and evaluation splits. Agree an annotation protocol with domain reviewers and retain disagreements. Evaluate region overlap or cell detection alongside the downstream measurement error; a good tile classifier does not by itself establish a reliable slide-level score.
Reproduce the complete measurement
Record the image identifier, analysis resolution, stain settings, model hash, extension versions, thresholds and script revision. Verify coordinate and unit conversions when moving masks between tools. Include missing measurements and failed jobs in exports. Compare reruns on the same inputs; fixed random seeds alone do not guarantee identical results across engines or hardware.
Benchmark a representative batch
Measure read time, tissue selection, inference, overlay writing and export separately. Test cold-cache and warm-cache behavior and concurrent jobs against available memory and storage bandwidth. Report slide characteristics, hardware and failure rate with throughput. Choose worker limits from those measurements instead of promising one slides-per-hour number for every lab.
Hand off to reviewers and LIMS
Agree stable case identifiers, field names, units, revision handling and approval status with the receiving system. Preserve overlays so reviewers can inspect what was counted. Validate imports and corrected reruns on a small batch before scaling. A pilot handover should include the project configuration, scripts, evaluation report, exception log and operating instructions.
QuPath’s command-line documentation describes the script subcommand; its workflow scripting guide explains running scripts across projects. Pin your installed version and test against its documentation. Project execution can overwrite saved analysis, so verify behavior on a project copy first.
Medical imaging integration and evaluation · Healthcare workflow planning
Candidate Analysis Workflows
Illustrative Research Blueprints
Planning examples, not completed client results or diagnostic performance guarantees.
Solid Tumor Burden at Cohort Scale
PD-L1 IHC Scoring with Reviewer QA
TIL Density + Spatial Heterogeneity
Validation & Operations
QuPath \ script run_pipeline.groovy \ --project my_project.qpproj # Illustrative invocation: verify your installed CLI, paths and script # Use Groovy/QuPath scripts to: # - Open project # - Iterate entries # - Run detectors/classifiers # - Save overlays # - Export tables
Frequently Asked Questions
Can the system process whole-slide images at scale?
Batch processing can be designed around a defined cohort, image reader and analysis pipeline. Measure image reading, preprocessing, inference and export on representative slides and the proposed hardware. Slide dimensions, storage bandwidth, cache state and model complexity determine the achievable throughput.
Which scanner formats are supported?
Format support depends on the selected image reader, software version, scanner export and installed dependencies. Test representative files at the required resolutions, including pixel-size metadata and multi-series images, before confirming support for a project.