Virtual Slides

Virtual slide creation and usage encompasses the complete digital pathology workflow:

1. Scanning — A whole-slide scanner acquires the tissue at high resolution (20x or 40x), producing the base level of the image pyramid.
2. Pyramid construction — Lower-resolution levels are generated by successive 2× downsampling. A 40x scan at 100,000 × 50,000 pixels generates levels at 50K×25K, 25K×12.5K, etc., down to a thumbnail.
3. File format — Virtual slides are stored in tile-based formats (SVS, NDPI, CZI, MRXS) where each pyramid level is divided into tiles that can be loaded independently.
4. Viewing — Slide viewer software streams tiles at the current zoom level, loading adjacent tiles as the user pans. This provides seamless navigation without loading the entire multi-GB file into memory.

The sampling theorem and pixel size (Hanrahan): The scanning resolution determines the information content of the virtual slide. At 40x with 0.25 µm pixels, the Nyquist limit allows detection of features down to 0.5 µm — sufficient for nuclear morphology and subcellular localization. At 20x with 0.5 µm pixels, features below 1 µm are unresolvable. The choice between 20x and 40x should match the analytical need: nuclear detection and biomarker quantification work at 20x; subcellular localization, mitotic figure identification, and nuclear grading may require 40x.

Image pyramids (Hanrahan/Pawley): The multi-resolution pyramid exploits a fundamental property of visual information: most navigational decisions require only low resolution ("where is the tumor?"), while detailed analysis requires high resolution ("what does this nucleus look like?"). The pyramid pre-computes the low-resolution views rather than computing them on the fly, trading storage space for viewing speed. Each pyramid level is exactly 2× the linear resolution of the level below, enabling standard 2× zoom steps.

Compression and information preservation: Virtual slide formats use lossy (JPEG) or lossless (JPEG2000, LZW) compression. Lossy compression reduces file size 5-10× but introduces artifacts — most visibly as block artifacts around sharp edges at high compression. For quantitative analysis (intensity measurement, spectral unmixing), lossless compression preserves the exact pixel values. For morphological assessment and detection, moderate JPEG compression is acceptable. Pawley warns: "avoid lossy compression" for data intended for quantitative analysis.

Building a digital pathology archive for a clinical trial:

• 300 patients × 2 slides each = 600 virtual slides
• Scanned at 20x, multispectral (9 channels): ~5 GB per slide = 3 TB total
• Stored on a networked server with DICOM-compliant metadata
• Pathologists review remotely for quality control and ROI annotation
• StrataQuest analyzes all 600 slides with the same parameter set for consistency
• Virtual slides archived for regulatory review, re-analysis with updated algorithms, and publication

The virtual slide archive ensures that every analysis can be reproduced, every result can be verified against the original image, and every slide can be re-analyzed with future methods — glass slides can fade or break, but digital archives persist.