Gigapixel Pyramid

The pyramid organizes a WSI into a stack of discrete resolution levels, each stored as a grid of small, fixed-size image tiles. The base level contains the full-resolution data, while each subsequent level is a 2x or 4x downsampled version of the previous one. This structure allows a viewer to request only the specific tiles intersecting the viewport at the appropriate zoom level, avoiding the need to decompress the entire gigapixel image.

Unlike flat image files that require sequential decoding, a pyramidal format supports random access to any tile at any resolution. A software client calculates the required tiles based on the user's pan and zoom coordinates, fetching only the data needed to render the current screen. This lazy loading mechanism ensures that navigation feels instantaneous, even when the underlying image is hundreds of thousands of pixels in diameter.

Gigapixel pyramids rely on advanced wavelet-based codecs like JPEG2000 to achieve high compression ratios while preserving diagnostic quality. The newer High-Throughput JPEG 2000 (HTJ2K) standard significantly reduces computational overhead, enabling faster decoding on standard hardware. These codecs support both lossless archival storage and visually lossless compression for real-time streaming, often reducing file sizes by an order of magnitude compared to raw data.

A gigapixel pyramid embeds precise spatial metadata, including physical pixel size (microns per pixel), objective magnification, and a global coordinate system. This allows software to map pixel coordinates to physical tissue locations and to co-register multiple slides or overlay analysis results. The coordinate system ensures that a point identified at one zoom level corresponds exactly to the same tissue location at any other resolution.

A high-resolution digital scan of an entire glass pathology slide, creating the gigapixel image file that the pyramid structure is designed to store and serve. WSIs are the raw input for computational analysis, typically captured at 40x magnification, resulting in files exceeding 100,000 x 100,000 pixels. The pyramid representation is essential because a standard WSI cannot be loaded into memory as a single flat image.

The process of dividing a massive WSI into smaller, manageable image tiles (e.g., 256x256 or 512x512 pixels) that can be processed by a convolutional neural network. The gigapixel pyramid facilitates this by allowing extraction at a specific resolution layer. Key considerations include:

• Overlap strategy to avoid boundary artifacts
• Foreground filtering to ignore glass background
• Magnification selection (20x vs 40x) based on the diagnostic task

The application of encoding algorithms to reduce the massive storage footprint of a gigapixel pyramid. Modern WSI formats use JPEG2000 or HTJ2K codecs, which offer:

• Lossless compression for primary diagnosis archives
• Lossy compression (up to 10:1) for algorithmic analysis without diagnostic degradation
• Progressive decoding that aligns with pyramid level-of-detail requests Compression efficiency directly impacts storage costs and network transfer speeds in clinical PACS environments.

The process of rendering a color-coded probability overlay on a WSI to visualize the spatial distribution of model predictions. The gigapixel pyramid enables efficient heatmap rendering by allowing the overlay to be generated at a lower resolution and then mapped back to the base layer coordinates. Common visualizations include:

• Attention heatmaps from MIL models highlighting diagnostically relevant regions
• Tumor probability maps for margin assessment
• Cell density maps for TIL quantification