Tissue Detection

Two things in StrataQuest share this name. The Tissue Detection engine sits in a layer's Pre-Processing section (Add > Detection > Tissue Detection) and returns a mask, tissue white and everything else black. The Tissue Detection tool, opened with Autodetect ROIs on the ROI toolbar, uses the same settings to draw regions of interestLoading... on the sample. Both run on the preview image, the top level of the cache, so the cache must be built first.

The settings describe four adjustments. An intensity threshold, computed automatically or typed, separates tissue from background. Fragments within Tissue Fragments – max. merging distance are joined into one region, regions smaller than Tissue Fragments – min. area are dropped, and Contour Inflate Size widens each region by a number of pixels so the tissue edges come along. Exclude border connected regions drops regions touching the border.

The Engines Reference's recipe is to start permissive and tighten. Set the threshold low so everything shows, raise it until the tissue is right, raise the merging setting in small steps until fragments join, and raise the minimum area last to clear small specks. Detection Profiles save a tuned set of values for reuse.

The image pyramid. A whole-slide scan holds far more pixels than the tissue outline needs. In digital pathology generally, scans are stored as pyramids: the same image at several resolutions, each level a fraction of the one below. Finding the edge between tissue and glass needs only a coarse level, which makes the job fast. StrataQuest's cache is this kind of pyramid, and Tissue Detection reads its top level.

Thresholds and histograms. On many brightfield previews the histogramLoading... of pixel values shows two populations, bright glass and darker tissue, with a valley between them, and a method such as Otsu'sLoading... picks a cutoff in that valley. The Engines Reference doesn't name the method behind Automatic, though it does name Otsu for the Nuclei and Total Area engines. Its own example is a fluorescence sample, where tissue is typically brighter than the background, so don't assume tissue is always the darker class.

From mask to regions. After thresholding, the tissue is a set of separate patches. In image processing generally, connected component labelingLoading... turns patches into distinct regions, each with its own number. Merging distance and minimum area then act on regions: the first joins pieces of one section split by tears or faint staining, and the second drops dust and debris.

Why inflate. The outermost cells of a section are often faint, or only partly inside a tight outline. Contour Inflate Size adds a margin of pixels so the tissue edges and a small surrounding area are included. A layer's Events Strategy then determines what happens to cells cut by a region's boundary.

In the ROI tool the User Manual describes Automatic differently: it determines red, green and blue channel values within three intervals, from a lower value (default 5) to an upper value (default 255). Applying Illumination Correction first improves Autodetect ROIs.

Source: StrataQuest 8.1 Engines Reference, ch. 44 (pp. 550–553); User Manual §10.3.1–10.3.2 (pp. 91–98), §11.1.1 (pp. 155–157).

Suppose you have a tissue microarray slide with many cores, and a whole-section slide from the same study.

1. Build the cache for each sample, and apply Illumination Correction.
2. Whole section: press Autodetect ROIs. In the dialog, start with a low Intensity Threshold, raise it until the section is outlined, then raise Tissue Fragments – max. merging distance until the section is one region. Save the settings as a profile for the rest of the batch.
3. Microarray: press TMA Detection on the toolbar instead. If spots are broken or the contrast is poor, set Spot Mean Radius by hand, then choose Create Samples for each Spot.
4. Inside a layer: add the Tissue Detection engine in Pre-Processing when a later engine needs a tissue mask as input.

Each TMA spot becomes its own sample, so NucleiLoading... detection, measurements and diagrams are reported per core.