Glossary

Whole Slide Image (WSI)

A high-resolution digital scan of an entire glass pathology slide, producing a gigapixel image file for computational analysis.

Get in touch Learn more

Cinematic shot of a sleek glass-walled boardroom on the 40th floor of a glass highrise, late afternoon light casting long shadows across a minimalist table with holographic AI workflow projections.

DIGITAL PATHOLOGY FUNDAMENTALS

What is Whole Slide Image (WSI)?

A Whole Slide Image (WSI) is a high-resolution digital replica of an entire glass pathology slide, creating a gigapixel file that enables computational analysis and remote diagnosis.

A Whole Slide Image (WSI) is a high-resolution digital scan of an entire glass pathology slide, produced by specialized scanners that capture the tissue sample at microscopic magnification, typically 20x or 40x. The resulting file is a multi-gigapixel, multi-resolution pyramidal image that preserves the complete histological detail of the original specimen, enabling pan-and-zoom viewing equivalent to a physical microscope.

WSIs serve as the foundational input for computational pathology pipelines, where deep learning models analyze the digitized tissue for tasks like cancer detection, Gleason grading, and biomarker quantification. The gigapixel scale necessitates specialized processing techniques such as patch extraction and Multiple Instance Learning (MIL) to manage the data volume and enable slide-level classification.

GIGAPIXEL PATHOLOGY

Key Characteristics of a WSI

A Whole Slide Image is not merely a large photograph; it is a multi-resolution, pyramidal data structure that encodes histological information at scales ranging from tissue architecture to nuclear morphology.

Gigapixel Scale & Data Volume

A single WSI routinely exceeds 100,000 x 100,000 pixels at 40x magnification, generating file sizes from 1 to 4 gigabytes per slide. This massive scale precludes direct input into standard convolutional neural networks and necessitates specialized patch-based processing and tiling strategies to manage memory constraints.

1-4 GB

Typical File Size

100k+

Pixels per Dimension

Multi-Resolution Pyramid Structure

WSI files are stored as pyramidal TIFFs containing multiple downsampled versions of the original image at different magnification levels. This structure mimics the pathologist's workflow:

Base level: Highest resolution (e.g., 40x or 20x) for nuclear detail.
Intermediate levels: For architectural context and glandular patterns.
Thumbnail level: Low-resolution overview for tissue orientation and navigation.

Tissue Segmentation & Foreground Detection

Before computational analysis, the tissue region must be separated from the glass background. This pre-processing step uses thresholding on pixel intensity in the HSV or LAB color space to generate a binary mask. Accurate segmentation eliminates wasted computation on empty white space, which can constitute over 70% of the slide area.

Patch-Based Processing Paradigm

Due to memory limitations, WSIs are decomposed into thousands of smaller patches (e.g., 256x256 or 512x512 pixels). Each patch is independently processed by a feature extractor, typically a Vision Transformer (ViT) or ResNet, to generate a feature embedding. These embeddings are then aggregated for slide-level classification using Multiple Instance Learning (MIL).

Metadata & Color Management

WSIs exhibit significant stain variability across laboratories due to differences in hematoxylin and eosin (H&E) protocols, scanner models, and reagent batches. Stain normalization techniques, such as Macenko or Vahadane methods, are critical pre-processing steps to standardize color distributions and ensure model generalizability across sites.

Interoperability & DICOM Standard

While many WSIs are stored in proprietary formats (e.g., Aperio SVS, Hamamatsu NDPI), the field is converging on DICOM Supplement 145 for pathology. This standard enables integration with hospital Picture Archiving and Communication Systems (PACS) and ensures consistent metadata handling for pixel spacing, objective power, and z-stack information.

WHOLE SLIDE IMAGING EXPLAINED

Frequently Asked Questions

Clear, technical answers to the most common questions about gigapixel digital pathology and computational analysis.

A Whole Slide Image (WSI) is a high-resolution digital replica of an entire glass pathology slide, captured by a specialized scanner to produce a single, navigable gigapixel image file. The process begins with a glass slide containing a thin tissue section, which is loaded into a slide scanner. The scanner uses a microscope objective lens and a precision motorized stage to systematically tile the slide, capturing thousands of individual high-magnification fields (typically at 20x or 40x, equivalent to 0.5 or 0.25 microns per pixel). These tiles are then digitally stitched together using software algorithms, often with focus stacking to ensure uniform sharpness across the entire tissue area. The resulting file, commonly stored in a pyramidal, multi-resolution format like OpenSlide or DICOM, allows a pathologist or algorithm to pan and zoom seamlessly, mimicking the experience of a physical microscope but with the added power of computational analysis.

COMPARATIVE ANALYSIS

WSI vs. Traditional Microscopy

A feature-by-feature comparison of digital whole slide imaging against conventional light microscopy for pathology workflows.

Feature	Whole Slide Image (WSI)	Traditional Microscopy	Hybrid Scanner
Image Resolution	Gigapixel (up to 100,000 × 100,000 px)	Optical (limited by objective lens)	Gigapixel with live optical overlay
Field of View	Entire tissue section at once	Single field per eyepiece position	Entire slide with region-of-interest zoom
Remote Consultation
Computational Analysis
Z-Stack / Multiplane	Up to 30 focal planes	Continuous manual focus	Up to 15 focal planes
Storage Medium	Digital file (0.5–5 GB per slide)	Physical glass slide	Digital file + physical slide
Slide Throughput	Up to 1,000 slides per 24 hours	60–80 slides per pathologist per day	Up to 400 slides per 24 hours
Standardization Across Sites
Initial Capital Cost	$150,000–$500,000 per scanner	$2,000–$10,000 per microscope	$80,000–$250,000 per scanner
Recurring Cost	Cloud storage ($0.02–$0.05/GB/month)	Slide archiving and maintenance	Storage + physical archiving

Enabling Efficiency, Speed & Accuracy

Intelligent Analysis, Decision & Execution

We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.

Talk to Us

Search across company data

Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.

Useful when people spend too long searching or get different answers from different systems.

Enterprise searchRAGPermissions

Automate internal workflows

Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.

Useful when repetitive work moves across multiple tools and teams.

AI agentsWorkflow automationGovernance

Add AI to products and internal tools

Build assistants, guided actions, or decision support into the software your team or customers already use.

Useful when AI needs to be part of the product, not a separate tool.

AI integrationDecision supportModel routing

ESSENTIAL CONCEPTS

Related Terms

Master the core computational pathology concepts that underpin whole slide image analysis, from weakly supervised learning paradigms to standardized grading systems.

Multiple Instance Learning (MIL)

A weakly supervised learning paradigm where a model is trained on labeled bags of instances rather than individually labeled examples. In WSI analysis, the entire gigapixel slide is the bag, and extracted patches are the instances. Only a slide-level label (e.g., 'cancerous') is required, eliminating the need for costly pixel-level annotations. Key MIL approaches include:

Instance-based: Aggregate patch predictions via pooling (mean, max)
Embedding-based: Learn a slide-level representation from patch features
Attention-based: Use learnable weights to focus on diagnostically relevant regions

The CLAM framework is a widely adopted attention-based MIL architecture for pathology.

EXPLORE

Patch Extraction & Tiling

The process of tessellating a gigapixel WSI into smaller, manageable tiles (e.g., 256×256 or 512×512 pixels) for processing by convolutional neural networks. Critical considerations include:

Magnification level: Typically 20× or 40×, balancing cellular detail with tissue context
Overlap strategy: Contiguous vs. overlapping tiling to avoid boundary artifacts
Tissue masking: Foreground detection to discard glass background regions
Metadata preservation: Maintaining spatial coordinates for reconstruction

Tools like OpenSlide and MONAI provide optimized libraries for efficient patch extraction from proprietary WSI formats.

Stain Normalization

A computational technique to standardize color appearance across pathology images, mitigating variability introduced by:

Different staining protocols and reagent batches
Scanner calibration differences
Tissue preparation artifacts

Common approaches include:

Reinhard normalization: Matching color statistics to a reference image
Macenko method: Decomposing stains via color deconvolution
CycleGAN-based: Deep learning style transfer for stain translation
Vahadane structure-preserving: Preserving biological structure while normalizing

Without normalization, models may learn spurious correlations with stain intensity rather than morphological features.

Gleason Grading

A standardized histological grading system for prostate cancer based on architectural patterns of tumor growth. Deep learning models now automate this task from WSIs:

Patterns 1-5: Range from well-formed glands (1) to sheets of cells (5)
Grade Groups: Modern 5-tier system replacing the original Gleason score
Primary + Secondary: Two most prevalent patterns are summed (e.g., 3+4=7)

AI-based Gleason grading systems achieve pathologist-level performance, providing:

Consistent, reproducible scoring
Quantification of pattern percentages
Identification of tertiary patterns

The ISUP consensus guidelines define the current clinical grading framework.

Slide-Level Classification

The task of assigning a single diagnostic label to an entire gigapixel WSI. Unlike patch-level classification, this requires aggregating information across thousands of tiles. Core strategies include:

MIL pooling: Aggregate patch predictions via attention or max pooling
Graph-based: Model spatial relationships between patches using GNNs
Transformer-based: Apply self-attention across patch embeddings

Common classification targets:

Cancer vs. benign detection
Tumor subtyping (e.g., adenocarcinoma vs. squamous cell)
Molecular phenotype prediction (e.g., MSI status)
Grading (e.g., Gleason, Nottingham)

Performance is measured via AUC-ROC and slide-level confusion matrices.

Tumor-Infiltrating Lymphocytes (TILs)

Immune cells that have migrated into the tumor microenvironment, quantified as a prognostic and predictive biomarker via computational pathology. AI-driven TIL analysis from H&E WSIs provides:

Stromal TILs: Percentage of tumor stroma occupied by lymphocytes
Intratumoral TILs: Lymphocytes directly infiltrating tumor cell nests
Spatial distribution: Hot vs. cold tumor regions

Clinical significance:

Triple-negative breast cancer: Higher TILs predict better chemotherapy response
Melanoma: TILs correlate with immunotherapy benefit
Colorectal cancer: TIL density informs prognosis

Deep learning models segment lymphocytes using nuclei detection and classify their spatial context.

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.

Limited slotsGet a Free AI Consultation

How We Work

Custom AI workflows for your Business

One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.