Field of View

Field of View management handles the tile-based architecture of whole-slide imaging:

1. Acquisition — The scanner moves systematically across the slide, capturing one FOV per camera exposure. Tiles overlap slightly (typically 5-10%) for reliable stitching.
2. Stitching — Overlapping regions are aligned using cross-correlation, and the tiles are merged into a seamless mosaic. The stitching is stored as a coordinate map rather than creating a single massive image file.
3. Tile-wise processing — Analysis engines load and process one tile at a time, keeping memory requirements manageable even for multi-gigabyte images.
4. Border handling — A configurable border (typically 50-200 pixels) from adjacent tiles is loaded alongside each FOV. Objects detected within this border zone are deduplicated to avoid double-counting.

Sampling and Nyquist (Hanrahan): Every pixel is a point sample of the underlying continuous image. The Nyquist-Shannon sampling theorem states that a signal can be perfectly reconstructed from its samples if the sampling rate is at least twice the highest frequency present. In microscopy: the pixel size must be ≤ half the smallest resolvable feature (the Rayleigh criterion distance). For a 1.4 NA objective at 500 nm, lateral resolution is ~180 nm, requiring pixels ≤ 90 nm. Most whole-slide scanners at 40x provide ~250 nm pixels — adequate for nuclear detection but not for diffraction-limited features.

Field illumination uniformity (Dobrucki): Within each FOV, illumination may not be perfectly uniform. Dobrucki warns that "mercury arc lamps have non-uniform field illumination" — intensity can drop 20-40% from center to edge. This creates artifacts: cells at the FOV center appear brighter than identical cells at the edge. Background Removal corrects this within each FOV, but the correction must be applied before any intensity-dependent analysis.

The tile boundary artifact: If tiles are not perfectly stitched (sub-pixel misalignment), features at tile boundaries may show intensity discontinuities — visible as faint lines in the image. More importantly, cells split across tile boundaries may be detected as two partial objects (one in each tile). Border handling solves this by extending each tile's analysis region into the adjacent tile, then deduplicating objects that appear in both tiles' border zones.

Processing a 20x whole-slide image tiled into 1,200 FOVs:

• Each FOV: 2048 × 2048 pixels at 0.5 µm/pixel = ~1 mm × 1 mm of tissue
• Border handling: 100 pixel overlap with adjacent tiles
• Tissue Detection identifies 450 FOVs containing tissue (remaining 750 are empty glass)
• Analysis processes only the 450 tissue-containing FOVs, saving 63% of computation time
• 45,000 nuclei detected across all tiles, with ~500 at tile boundaries handled by deduplication