Inferensys

Glossary

DICOM Whole Slide Imaging

A DICOM supplement defining the storage and handling of gigapixel digital pathology images, enabling the use of standard PACS infrastructure for microscopy data.
Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.
DIGITAL PATHOLOGY INTEROPERABILITY

What is DICOM Whole Slide Imaging?

DICOM Whole Slide Imaging (WSI) is a standard supplement to the DICOM protocol that defines a data structure and metadata framework for the storage, retrieval, and viewing of gigapixel digital pathology images within existing Picture Archiving and Communication Systems (PACS).

DICOM Whole Slide Imaging (WSI) is a specification that extends the DICOM standard to accommodate the unique data characteristics of digital pathology. It defines a dual-layer storage model: a low-resolution overview image for rapid navigation and a high-resolution, tiled pyramid representation of the entire glass slide. This structure enables efficient streaming and pan-and-zoom viewing of massive, multi-gigabyte images without requiring the entire file to be loaded into memory.

By encoding pathology images as a native DICOM SOP Class, WSI enables the convergence of radiology and pathology workflows on a single VNA or PACS infrastructure. This integration leverages existing DICOMweb services like WADO-RS for efficient, HTTP-based retrieval of specific image tiles, eliminating departmental data silos and allowing for the unified management of a patient's complete diagnostic imaging record.

DIGITAL PATHOLOGY INTEROPERABILITY

Core Characteristics of DICOM WSI

DICOM Whole Slide Imaging (WSI) extends the DICOM standard to manage gigapixel digital pathology images, enabling their storage, retrieval, and viewing within standard PACS infrastructure.

01

Multi-Resolution Pyramid Structure

A DICOM WSI file stores the image as a pyramidal data structure, not a single flat image. This pyramid contains multiple downsampled versions of the original slide, from the highest resolution base layer to a low-resolution thumbnail. This enables efficient progressive streaming, where a viewer requests only the resolution and spatial region needed for the current zoom level, avoiding the need to transfer the entire gigapixel file over a network.

02

Tiled Organization for Sparse Access

Each layer of the pyramid is subdivided into a regular grid of rectangular tiles, typically 256x256 or 512x512 pixels. This tiling is the core mechanism for sparse data access. A viewing client can use the DICOMweb WADO-RS service to fetch only the specific tiles that fall within the current viewport, rather than downloading the entire image. This HTTP-based tile retrieval is fundamental to the responsive pan-and-zoom experience in digital pathology.

03

Dual-Personality SOP Classes

DICOM WSI introduces VL Whole Slide Microscopy Image Storage SOP Classes. These are unique because they are 'dual-personality' objects. They contain both traditional DICOM patient and study metadata in a standard header and the pixel data pyramid. This allows a PACS to index and manage a WSI file like any other DICOM image while specialized viewers can interpret the pyramid for navigation. The primary SOP Class UID is 1.2.840.10008.5.1.4.1.1.77.1.6.

04

Companion Segmentation Objects

WSI is often analyzed by AI to produce tissue maps. The DICOM Segmentation SOP Class is used to store these results as a companion object. A segmentation object references the source WSI instance and encodes binary or fractional maps (e.g., tumor regions) as a tiled, multi-resolution pyramid. This maintains the spatial alignment between the original image and the AI-generated overlay, allowing them to be viewed in perfect registration within a PACS viewer.

05

Concatenation for Unlimited Size

A single gigapixel WSI can exceed the 4 GB limit of a classic DICOM file. To handle this, the standard uses DICOM Concatenation. The image is split across multiple DICOM instances (parts) that share a single SOP Instance UID. The Concatenation UID (0020,9161) and In-concatenation Number (0020,9162) tags link these parts together, allowing a reader to reassemble the complete dataset seamlessly. This ensures even the largest slides remain compliant.

06

Optical Path and Z-Stack Encoding

WSI supports advanced microscopy techniques. The Optical Path Sequence (0048,0105) describes the illumination, filter, and objective lens used to acquire the image, which is critical for fluorescence or multispectral imaging. Furthermore, the standard can encode Z-stacks—multiple focal planes at the same XY position—by using the Dimension Organization module, allowing pathologists to focus through the depth of a tissue sample directly within a DICOM-compliant viewer.

DICOM WHOLE SLIDE IMAGING

Frequently Asked Questions

Essential questions about the DICOM supplement that brings gigapixel digital pathology into the radiology workflow, enabling standard PACS infrastructure to manage microscopy data.

DICOM Whole Slide Imaging (WSI) is a supplement to the DICOM standard that defines how gigapixel digital pathology images are stored, transmitted, and managed using the same infrastructure as radiology. Unlike a standard DICOM CT or MR image, which is a single frame or a small stack of frames, a WSI object represents a multi-resolution pyramid of images. The core difference lies in the data structure: WSI uses a tiled pyramidal organization where the full-resolution baseline layer is divided into thousands of individual square tiles, and multiple downsampled layers are generated to enable smooth zooming and panning. This is defined in the VL Whole Slide Microscopy Image Storage SOP Class. Standard DICOM relies on a single pixel matrix per frame; WSI introduces the concept of a Total Pixel Matrix and a Tiled Organization, allowing a viewer to request only the specific tiles needed for the current viewport, rather than downloading the entire multi-gigabyte file. This enables the use of DICOMweb protocols like WADO-RS for efficient, HTTP-based streaming of pathology data directly into a standard PACS workstation.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.