Key concepts

Pathology in Gradient Biotech is organized around tissue images, histological structure, image-derived measurements, and reproducible analysis runs. The goal is not to diagnose disease automatically. The goal is to turn slide images into auditable research metrics: tissue masks, segmented cell objects, region-level summaries, spatial density maps, and morphology context for translational studies.

Tissue images

A tissue image is the visual input for computational pathology. It may be a full scanner export, a cropped region image, or a microscopy image from a specific assay.

Common image concepts:

• Whole-slide image (WSI): a high-resolution digital scan of an entire glass slide.
• Region image: a smaller field or crop used for faster analysis, development, or focused review.
• Tissue microarray (TMA): a slide containing many small tissue cores arranged in a grid.
• Multiplex immunofluorescence (multiplex IF): microscopy assay that measures several fluorescent markers on the same tissue section.
• Pixel size: physical scale represented by one image pixel, often recorded as microns per pixel.
• Magnification: scanner or microscope zoom level, such as 20x or 40x, which affects apparent cell size and analysis parameters.

Image-derived metrics depend on stain quality, scanner settings, resolution, focus, tissue folds, and whether the uploaded file preserves scale metadata.

Histology and stains

Histology is the microscopic study of tissue structure. Computational pathology pipelines analyze the visual patterns created by stains, cells, and tissue organization.

Core stain concepts:

• Hematoxylin and eosin (H&E): common tissue stain where hematoxylin highlights nuclei and eosin highlights cytoplasm and extracellular matrix.
• Immunohistochemistry (IHC): stain that uses antibodies to detect a specific protein target in tissue.
• Immunofluorescence (IF): stain that uses fluorescent labels to detect targets under fluorescence imaging.
• Multiplex immunofluorescence (multiplex IF): immunofluorescence assay with multiple marker channels.
• Color deconvolution: computational separation of stain channels so nuclei, cytoplasm, or antibody signal can be analyzed more directly.

Stain type determines what can be inferred. Hematoxylin and eosin supports morphology and broad tissue architecture. Immunohistochemistry and immunofluorescence can support marker-specific research questions when the assay and image channels are available.

Tissue compartments

A tissue compartment is a region with biological or histological meaning. Compartments can be detected algorithmically, imported from annotations, or reviewed visually.

Common compartment concepts:

• Tumor region: area dominated by malignant tissue.
• Stroma: connective tissue, fibroblasts, vessels, extracellular matrix, and other non-tumor structural components.
• Necrosis: tissue death region that may affect segmentation, density, and interpretation.
• Invasive margin: boundary region where tumor and surrounding tissue interact.
• Lymphoid aggregate: localized immune-rich structure that may be relevant to immune infiltration studies.
• Region of interest (ROI): user-defined area selected for focused analysis.

Compartment labels are strongest when they come from expert annotation or a validated detection workflow. Automated region labels should be treated as research annotations, not clinical sign-out.

Tiling and slide viewing

Whole-slide images are too large to load into the browser as a single full-resolution image. Pathology uses a tile pyramid so the viewer can load only the visible pieces at the current zoom level.

Important tiling concepts:

• Tile: small image patch, often 256 pixels square, loaded on demand.
• Tile pyramid: multi-resolution set of tiles covering the same slide at several zoom levels.
• Tile manifest: metadata describing tile size, available levels, thumbnail path, and coordinate layout.
• Thumbnail: low-resolution preview used for fast orientation.
• Pan and zoom viewer: interactive viewer that navigates the tile pyramid instead of loading the full slide at once.

Generate tiles before large-slide viewing and before segmentation workflows that depend on tiled image access.

Tissue detection

Tissue detection separates tissue foreground from blank glass, background, coverslip edges, and other non-informative regions.

Key tissue detection outputs:

• Tissue mask: binary or grayscale image marking tissue foreground.
• Tissue fraction: proportion of image area classified as tissue.
• Region bounding box: rectangular summary around a detected tissue region.
• Focus score: blur or focus quality indicator used for quality control.
• Quality control (QC) warning: flag for image problems such as low tissue area, blur, or tile-level artifacts.

Tissue detection makes downstream analysis more stable because segmentation and density calculations can ignore blank background. It still depends on stain contrast and image quality.

Cell segmentation

Cell segmentation identifies individual nuclei or cells in the image and converts visual objects into measurements.

Core segmentation concepts:

• Nucleus detection: identifying nuclei, often the most reliable object in hematoxylin and eosin images because nuclei are darkly stained.
• Cell boundary: estimated outline of an individual cell or nucleus.
• Centroid: x/y coordinate representing the center of a detected object.
• Area: size of the detected object in pixels or physical units when scale is available.
• Confidence score: model or algorithm score for the detected object.
• Overlay: visual layer showing segmentation boundaries on top of the slide.

The current baseline segmentation is research-oriented and depends heavily on stain contrast, resolution, focus, and tissue type. Counts and boundaries should be reviewed visually before biological interpretation.

Spatial quantification

Spatial quantification measures how detected cells and tissue regions are arranged across the slide.

Common spatial concepts:

• Cell density: number of detected cells per image area or tissue area.
• Density heatmap: visual map showing local areas of high or low cell density.
• Nearest-neighbor distance: distance from one detected cell to its closest neighboring detected cell.
• Spatial clustering: tendency for cells to group together instead of being evenly distributed.
• Infiltration phenotype: research summary of how cells appear distributed through tissue compartments, such as inflamed, excluded, or sparse patterns.
• Region-level metric: statistic calculated within a labeled compartment rather than across the whole slide.

Spatial metrics are strongest when segmentation, tissue masks, and annotations are reliable. They describe image-derived patterns; they do not prove cell identity or biological mechanism by themselves.

Annotations

Annotations are region labels drawn or exported from another tool and imported into the pathology workspace.

Annotation concepts:

• GeoJSON: geospatial JavaScript Object Notation format used here for polygon and multipolygon region annotations.
• QuPath: open-source digital pathology tool that can export annotated regions.
• Feature collection: GeoJSON container holding multiple annotated regions.
• Polygon: coordinate-defined region boundary.
• Classification or label: region name such as tumor, stroma, necrosis, or margin.

Annotations let region-level quantification follow biologically meaningful boundaries instead of only whole-slide measurements.

Tumor microenvironment context

For cancer tissue studies, pathology often contributes image-based context for the tumor microenvironment (TME): the tumor cells, stromal cells, immune cells, vasculature, matrix, and tissue architecture around the tumor.

Pathology metrics can support questions such as:

• Is the tissue cell-dense or sparse?
• Are cells concentrated near the invasive margin?
• Do annotated tumor and stroma regions show different densities?
• Does the image show an inflamed, excluded, or low-infiltration pattern?
• Do image-derived regions align with molecular findings from spatial omics?

Pathology does not identify every immune cell type from hematoxylin and eosin alone. Marker-specific assays, molecular data, or expert annotation are needed for stronger cell-type interpretation.

Study

A study is the top-level container for one pathology project. It holds slides, sample metadata, pipeline runs, annotations, and artifacts. Create one study per trial, tissue collection, cohort, or analysis project.

Optional metadata includes disease area and tissue type.

Slide

A slide is one uploaded image attached to a study. Each slide records:

• Modality, such as whole-slide image, region image, tissue microarray, immunofluorescence, or multiplex immunofluorescence
• Stain type, such as hematoxylin and eosin, immunohistochemistry, immunofluorescence, or multiplex immunofluorescence
• Scanner format and pixel dimensions
• Microns per pixel when detected
• Ingestion status and storage path

Open a slide to generate tiles, run pipelines, inspect overlays, and export results.

Sample and subject

A sample links a slide to cohort metadata such as diagnosis, tumor grade, treatment, response status, group label, and timepoint. A subject is the person, animal, or experimental source that contributed one or more samples.

Good sample identifiers matter because they connect pathology outputs to oncology, immunology, computational biology, or clinical metadata without mixing slides from different sources or timepoints.

Pipeline run

A run is one execution of an analysis pipeline. Each run stores:

• Unique run_id
• Pipeline type and version
• Parameter JavaScript Object Notation (JSON)
• Status: queued -> running -> complete or failed
• Output artifacts, such as masks, overlays, cell tables, metric tables, and summaries
• Timestamps and summary statistics

Runs chain through prerequisites. Spatial quantification requires a completed segmentation run. Segmentation works best after tile generation.

Artifacts

Artifacts are files produced by analysis runs. Pipeline outputs are written under data/pathology/artifacts/{run_id}/ and tile pyramids under data/pathology/tiles/{slide_id}/.

Common artifacts:

• tissue_mask.png: tissue foreground mask
• tissue_regions.json: detected region labels and bounding boxes
• cell_overlay.png: segmentation visualization for the viewer
• cells.json: detected cell centroids, areas, labels, and confidence scores
• density_heatmap.png: spatial density visualization
• cell_spatial_metrics.csv: comma-separated values table with per-cell spatial measurements
• region_metrics.json: region-level quantification summary

Download artifacts from slide detail pages when export links are available.

Relationship to other research areas

Pathology owns whole-slide imaging, tissue masks, cell segmentation, image overlays, morphology-guided regions, and image-derived spatial metrics.

Computational Biology owns spatial transcriptomics ingestion, single-cell analysis, deconvolution, differential expression, and generic omics spatial statistics. Pathology contributes the tissue morphology layer that helps interpret those molecular outputs.

Oncology adds clinical endpoints, mutation context, immune profiling, survival analysis, and tumor-specific interpretation. Oncology can use pathology outputs as histological context for tumor microenvironment and translational cohort analysis.

Provenance and reproducibility

Every analysis is reproducible at the run level. The study run history records pipeline type, parameters, status, timestamps, versions, and artifact links.

For reports or methods sections, include the slide identifier, stain type, pixel scale when available, pipeline type, pipeline version, parameters, and run identifier.

Interpretation scope

Pathology outputs are research metrics. They are not clinical diagnoses, regulatory-grade pathology results, or pathologist sign-out.

Review image quality, stain context, tissue masks, segmentation overlays, and annotation quality before making biological claims. When possible, validate automated measurements against expert review, orthogonal assays, or known controls.