Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SPLICE -- Streamlining Digital Pathology Image Processing

Areej Alsaafin, Peyman Nejat, Abubakr Shafique, Jibran Khan, Saghir Alfasly, Ghazal Alabtah, H. R. Tizhoosh

TL;DR

SPLICE addresses the challenge of indexing and retrieving gigapixel whole-slide images by unsupervisedly selecting a small, non-redundant set of patches that form a collage representing each WSI. The method uses sequential color-feature analysis and percentile thresholds to minimize redundancy, then indexes collage patches with deep embeddings and binary encodings for rapid retrieval. Across Mayo Clinic datasets and TCGA, SPLICE delivers competitive retrieval accuracy while using far fewer patches and requiring less storage than mosaic-based approaches, enabling scalable, efficient computational pathology workflows. The approach reduces storage needs by about 50% and achieves robust performance with a single tunable percentile parameter, highlighting practical benefits for large-scale digital pathology archives.

Abstract

Digital pathology and the integration of artificial intelligence (AI) models have revolutionized histopathology, opening new opportunities. With the increasing availability of Whole Slide Images (WSIs), there's a growing demand for efficient retrieval, processing, and analysis of relevant images from vast biomedical archives. However, processing WSIs presents challenges due to their large size and content complexity. Full computer digestion of WSIs is impractical, and processing all patches individually is prohibitively expensive. In this paper, we propose an unsupervised patching algorithm, Sequential Patching Lattice for Image Classification and Enquiry (SPLICE). This novel approach condenses a histopathology WSI into a compact set of representative patches, forming a "collage" of WSI while minimizing redundancy. SPLICE prioritizes patch quality and uniqueness by sequentially analyzing a WSI and selecting non-redundant representative features. We evaluated SPLICE for search and match applications, demonstrating improved accuracy, reduced computation time, and storage requirements compared to existing state-of-the-art methods. As an unsupervised method, SPLICE effectively reduces storage requirements for representing tissue images by 50%. This reduction enables numerous algorithms in computational pathology to operate much more efficiently, paving the way for accelerated adoption of digital pathology.

SPLICE -- Streamlining Digital Pathology Image Processing

TL;DR

SPLICE addresses the challenge of indexing and retrieving gigapixel whole-slide images by unsupervisedly selecting a small, non-redundant set of patches that form a collage representing each WSI. The method uses sequential color-feature analysis and percentile thresholds to minimize redundancy, then indexes collage patches with deep embeddings and binary encodings for rapid retrieval. Across Mayo Clinic datasets and TCGA, SPLICE delivers competitive retrieval accuracy while using far fewer patches and requiring less storage than mosaic-based approaches, enabling scalable, efficient computational pathology workflows. The approach reduces storage needs by about 50% and achieves robust performance with a single tunable percentile parameter, highlighting practical benefits for large-scale digital pathology archives.

Abstract

Digital pathology and the integration of artificial intelligence (AI) models have revolutionized histopathology, opening new opportunities. With the increasing availability of Whole Slide Images (WSIs), there's a growing demand for efficient retrieval, processing, and analysis of relevant images from vast biomedical archives. However, processing WSIs presents challenges due to their large size and content complexity. Full computer digestion of WSIs is impractical, and processing all patches individually is prohibitively expensive. In this paper, we propose an unsupervised patching algorithm, Sequential Patching Lattice for Image Classification and Enquiry (SPLICE). This novel approach condenses a histopathology WSI into a compact set of representative patches, forming a "collage" of WSI while minimizing redundancy. SPLICE prioritizes patch quality and uniqueness by sequentially analyzing a WSI and selecting non-redundant representative features. We evaluated SPLICE for search and match applications, demonstrating improved accuracy, reduced computation time, and storage requirements compared to existing state-of-the-art methods. As an unsupervised method, SPLICE effectively reduces storage requirements for representing tissue images by 50%. This reduction enables numerous algorithms in computational pathology to operate much more efficiently, paving the way for accelerated adoption of digital pathology.
Paper Structure (6 sections, 10 figures, 4 tables, 1 algorithm)

This paper contains 6 sections, 10 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Results
  3. WSI Search & Match Using Mayo Clinic Datasets
  4. WSI Search & Match Using Public Datasets
  5. Discussion
  6. Methods

Figures (10)

  • Figure 1: SPLICE approach for patch selection. SPLICE selects representative patches, a "collage", by analyzing their color characteristics through sequential comparisons, including only patches with unique features in the WSI collage.
  • Figure 2: Simplified Example for SPLICE collage applied on a WSI region (From top to bottom) -- Starting from an image, we define a lattice. In every pass, the reference patch (pointer) is marked with a pink square while the similar patches to the reference patch are overlaid. First Pass: Patch 1 selected and compared to all other patches, similar patches are marked by 1 and excluded, Second Pass: Patch 2 is compared to all remaining patches; similar patches are marked by 2 and excluded, Third Pass: Patch 3 is compared to all remaining patches; similar patches are marked by 3 and excluded, Fourth Pass: Patch 4 is compared to all remaining patches; similar patches are marked by 4 and excluded, Fifth Pass: Patch 5 is compared to all remaining patches; similar patches are marked by 5 and excluded, Last Pass: the collage of selected patches is generated.
  • Figure 3: Representative patch selection using SPLICE in histopathology whole slide image from Mayo-CRC dataset. These diverse patches, extracted by SPLICE, showcase various patterns within the histopathology image of a colon specimen. The patches are selected by applying sequential analysis, leveraging the color characteristics of the patches for their distinctive representation.
  • Figure 4: Yottixel approach for patch selection. Yottixel employs a two-step process, beginning with color clustering, followed by spatial clustering, to select a certain percentage of patches from each color cluster. The final set of selected patches is called the mosaic that represents the WSI.
  • Figure 5: Distribution of patches across various collage sizes, ranging from the 10th to the 50th percentile, employing whole-slide images sourced from the Mayo-CRC dataset. The histogram provides a visual representation of the patch distribution, while the rug plot at the bottom offers a detailed view of individual data points, enhancing the understanding of the data's spread and density.
  • ...and 5 more figures