Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Scalable Whole Slide Image Representation Using K-Mean Clustering and Fisher Vector Aggregation

Ravi Kant Gupta, Shounak Das, Ardhendu Sekhar, Amit Sethi

TL;DR

This work addresses scalable WSI classification by replacing end-to-end full-image models with a patch-based pipeline. It extracts patch embeddings from fixed-size patches, clusters them with K-means ($k=10$), and encodes cluster-wise distributions via Fisher vectors using a Gaussian mixture model with $m=5$ centers, forming a concatenated WSI descriptor that feeds a classifier (AMIL). Key hyperparameters include $\pi_m=0.2$ and $\sigma_m=0.1$, with data augmentations; the approach demonstrates strong performance across four datasets (Warwick HER2, TCGA-BRCA, TCGA-LUAD, CAMELYON17) in HER2 scoring, mutation prediction, and metastasis detection, indicating robustness and scalability. The method provides a practical, scalable representation of WSIs that preserves local heterogeneity while enabling global discrimination, with potential clinical impact and avenues for future real-time and interpretability research.

Abstract

Whole slide images (WSIs) are high-resolution, gigapixel sized images that pose significant computational challenges for traditional machine learning models due to their size and heterogeneity.In this paper, we present a scalable and efficient methodology for WSI classification by leveraging patch-based feature extraction, clustering, and Fisher vector encoding. Initially, WSIs are divided into fixed size patches, and deep feature embeddings are extracted from each patch using a pre-trained convolutional neural network (CNN). These patch-level embeddings are subsequently clustered using K-means clustering, where each cluster aggregates semantically similar regions of the WSI. To effectively summarize each cluster, Fisher vector representations are computed by modeling the distribution of patch embeddings in each cluster as a parametric Gaussian mixture model (GMM). The Fisher vectors from each cluster are concatenated into a high-dimensional feature vector, creating a compact and informative representation of the entire WSI. This feature vector is then used by a classifier to predict the WSI's diagnostic label. Our method captures local and global tissue structures and yields robust performance for large-scale WSI classification, demonstrating superior accuracy and scalability compared to other approaches.

Scalable Whole Slide Image Representation Using K-Mean Clustering and Fisher Vector Aggregation

TL;DR

This work addresses scalable WSI classification by replacing end-to-end full-image models with a patch-based pipeline. It extracts patch embeddings from fixed-size patches, clusters them with K-means (), and encodes cluster-wise distributions via Fisher vectors using a Gaussian mixture model with centers, forming a concatenated WSI descriptor that feeds a classifier (AMIL). Key hyperparameters include and , with data augmentations; the approach demonstrates strong performance across four datasets (Warwick HER2, TCGA-BRCA, TCGA-LUAD, CAMELYON17) in HER2 scoring, mutation prediction, and metastasis detection, indicating robustness and scalability. The method provides a practical, scalable representation of WSIs that preserves local heterogeneity while enabling global discrimination, with potential clinical impact and avenues for future real-time and interpretability research.

Abstract

Whole slide images (WSIs) are high-resolution, gigapixel sized images that pose significant computational challenges for traditional machine learning models due to their size and heterogeneity.In this paper, we present a scalable and efficient methodology for WSI classification by leveraging patch-based feature extraction, clustering, and Fisher vector encoding. Initially, WSIs are divided into fixed size patches, and deep feature embeddings are extracted from each patch using a pre-trained convolutional neural network (CNN). These patch-level embeddings are subsequently clustered using K-means clustering, where each cluster aggregates semantically similar regions of the WSI. To effectively summarize each cluster, Fisher vector representations are computed by modeling the distribution of patch embeddings in each cluster as a parametric Gaussian mixture model (GMM). The Fisher vectors from each cluster are concatenated into a high-dimensional feature vector, creating a compact and informative representation of the entire WSI. This feature vector is then used by a classifier to predict the WSI's diagnostic label. Our method captures local and global tissue structures and yields robust performance for large-scale WSI classification, demonstrating superior accuracy and scalability compared to other approaches.
Paper Structure (4 sections, 2 equations, 1 figure, 4 tables)

This paper contains 4 sections, 2 equations, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Datasets, Experiments, and Results
  4. Conclusion and Discussion

Figures (1)

  • Figure 1: Patches from WSIs are encoded into embeddings, clustered via K-means, and transformed into Fisher vectors using a GMM model. These Fisher vectors are concatenated into a single vector representing the WSI, which is then used for classification.