Tissue Detection

Tissue Detection operates on the preview image — the lowest-resolution layer of the image pyramid — for fast whole-slide processing. The algorithm proceeds in four stages:

1. Threshold — Separates tissue (colored or dark) from glass background (white or bright) using Otsu's automatic threshold or a manual intensity value.
2. Merge — Joins nearby tissue fragments that are within the configurable merging distance. This prevents small tears or folds in the tissue from creating separate analysis regions.
3. Filter — Removes small debris and fragments below the minimum area threshold. Dust, staining artifacts, and coverslip edges are eliminated.
4. Inflate — Expands the detected boundary outward by a configurable margin. This safety margin ensures cells at the tissue edge are fully included in the analysis region.

The cache must be built before running Tissue Detection. Detection Profiles allow saving and reusing optimized parameter sets across batches of similarly prepared tissue.

The image pyramid: Whole-slide images can be enormous — 40,000 × 30,000 pixels or more. Processing this at full resolution for tissue detection would be wasteful since we only need to find the coarse boundary between tissue and glass. Image pyramids solve this by storing the same image at multiple resolutions. The lowest level might be 1/64th the original dimensions, reducing a billion-pixel image to a few hundred thousand pixels — fast enough for instant analysis.

Why Otsu works here: Tissue detection is a near-ideal use case for Otsu's method. The histogram is strongly bimodal — glass background forms one sharp peak (bright), tissue forms another (darker). The valley between them is deep and clean. Otsu's algorithm tries every possible threshold, picks the one that makes the two groups internally most uniform, and this reliably falls in the valley between the peaks. This is one of the few situations where global thresholding works almost perfectly.

Connected components: After thresholding, the binary mask may contain multiple disconnected tissue regions (tissue sections, TMA cores, fragments). Connected component labeling identifies each separate region — the algorithm iteratively grows from a seed pixel within each region, constrained to stay within the thresholded area, until all pixels in that region are labeled. Each connected component becomes a potential analysis region.

The inflation step matters: Cells at the tissue edge often have only partial nuclear staining or sit right at the tissue-glass boundary. Without inflation, these edge cells may fall outside the detected tissue region and be excluded from analysis. The inflation parameter (typically 5-20 pixels at preview resolution) adds a buffer zone. This is related to Howard's brick rule in stereology — partial objects at boundaries need systematic inclusion or exclusion rules to avoid counting bias.

On a tissue microarray slide with 60 cores of breast cancer biopsies:

1. Build the cache to generate the image pyramid
2. Enable TMA Detection mode
3. Set Minimum Area to exclude tissue debris between cores
4. Set Inflate Size to 10 pixels to capture edge cells

Result: 60 individual tissue cores are identified automatically, each becoming a separate region of interest. Subsequent engines (Nuclei Detection, measurements, phenotyping) process each core independently, enabling per-core statistics for the entire cohort in a single automated run.