Define the intended use
A scanner-validation study, image-analysis biomarker study, telepathology study and AI diagnostic study require different endpoints and validation designs.
Digital pathology research essentials
Image acquisition
Record scanner model, objective, numerical aperture where relevant, pixel resolution, focus strategy, z-stacking, file format, compression, staining method and image QC.
Ground truth
Define how reference labels were established: single pathologist, consensus, adjudication, ancillary tests or clinical outcome. Measure interobserver variability when it affects the endpoint.
AI and computational pathology
Separate training, tuning/validation and independent test data at the patient level. Prevent slide or patient leakage across datasets. Report external validation and subgroup performance.
Performance reporting
Use clinically meaningful metrics with confidence intervals. Depending on the task, report sensitivity, specificity, AUROC, precision/recall, calibration, agreement or segmentation metrics.
Reproducibility
Preserve image-processing parameters, model version, code, preprocessing and hardware/software environment. Follow emerging digital pathology and AI reporting standards.
Research-use note: This resource is educational and methodological. Study-specific protocols require scientific review, appropriate ethics approval, validated laboratory methods and a prespecified statistical plan.
Planning a digital pathology study?
Our team can collaborate on whole-slide imaging, computational pathology and AI validation studies.