Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A General Pipeline for Glomerulus Whole-Slide Image Segmentation

Quan Huu Cap

TL;DR

Problem: Glomerulus segmentation in whole-slide kidney images is critical for CKD diagnosis but challenging due to glomerular variability and staining differences. Method: A general stitching-based segmentation pipeline extracts overlapping patches, aggregates their predictions over overlaps, and evaluates multiple models on two large public datasets (KPIs and Mice glomeruli). Findings: The approach achieves state-of-the-art performance for both patch-level and WSI-level segmentation, notably with Mask2Former attaining a WSI Dice of 94.64 on KPIs; stitching consistently improves results across models and datasets. Significance: The work provides a practical, reproducible framework with publicly available code and pretrained models that can accelerate CKD-related diagnostics and further research.

Abstract

Whole-slide images (WSI) glomerulus segmentation is essential for accurately diagnosing kidney diseases. In this work, we propose a general and practical pipeline for glomerulus segmentation that effectively enhances both patch-level and WSI-level segmentation tasks. Our approach leverages stitching on overlapping patches, increasing the detection coverage, especially when glomeruli are located near patch image borders. In addition, we conduct comprehensive evaluations from different segmentation models across two large and diverse datasets with over 30K glomerulus annotations. Experimental results demonstrate that models using our pipeline outperform the previous state-of-the-art method, achieving superior results across both datasets and setting a new benchmark for glomerulus segmentation in WSIs. The code and pre-trained models are available at https://github.com/huuquan1994/wsi_glomerulus_seg.

A General Pipeline for Glomerulus Whole-Slide Image Segmentation

TL;DR

Problem: Glomerulus segmentation in whole-slide kidney images is critical for CKD diagnosis but challenging due to glomerular variability and staining differences. Method: A general stitching-based segmentation pipeline extracts overlapping patches, aggregates their predictions over overlaps, and evaluates multiple models on two large public datasets (KPIs and Mice glomeruli). Findings: The approach achieves state-of-the-art performance for both patch-level and WSI-level segmentation, notably with Mask2Former attaining a WSI Dice of 94.64 on KPIs; stitching consistently improves results across models and datasets. Significance: The work provides a practical, reproducible framework with publicly available code and pretrained models that can accelerate CKD-related diagnostics and further research.

Abstract

Whole-slide images (WSI) glomerulus segmentation is essential for accurately diagnosing kidney diseases. In this work, we propose a general and practical pipeline for glomerulus segmentation that effectively enhances both patch-level and WSI-level segmentation tasks. Our approach leverages stitching on overlapping patches, increasing the detection coverage, especially when glomeruli are located near patch image borders. In addition, we conduct comprehensive evaluations from different segmentation models across two large and diverse datasets with over 30K glomerulus annotations. Experimental results demonstrate that models using our pipeline outperform the previous state-of-the-art method, achieving superior results across both datasets and setting a new benchmark for glomerulus segmentation in WSIs. The code and pre-trained models are available at https://github.com/huuquan1994/wsi_glomerulus_seg.

Paper Structure

This paper contains 11 sections, 3 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. METHOD AND MATERIAL
  3. The Proposed Segmentation Pipeline
  4. Datasets
  5. The KPIs Dataset
  6. The Mice Glomeruli Dataset
  7. EXPERIMENTAL RESULTS
  8. Training Segmentation Models
  9. Model Inference
  10. Results and Discussion
  11. CONCLUSION

Figures (3)

  • Figure 1: The overview of our proposed pipeline for WSI glomerulus segmentation.
  • Figure 2: A comparison between single patch prediction and stitching on overlapping predictions. This strategy effectively increases the detection coverage, especially when glomeruli located near patch image borders.
  • Figure 3: The visual comparison of patch-level segmentation results on the mice glomeruli dataset (first two rows) and the KPIs dataset (last two rows) from different segmentation models.