Published April 18, 2024 | Version v1
Other Open

Kidney Pathology Image Segmentation (KPIs) Challenge 2024: Structured description of the challenge design

  • 1. Department of Computer Science, School of Engineering, Vanderbilt University, Nashville, USA
  • 2. Department of Pathology, Microbiology and Immunology, Vanderbilt University Medical Center, Nashville, USA
  • 3. Department of Biostatistics, Vanderbilt University Medical Center, Nashville, USA
  • 4. Department of Electrical and Computer Engineering, School of Engineering, Vanderbilt University, Nashville, USA
  • 5. NVIDIA Corp, Bethesda, USA

Description

Chronic kidney disease (CKD) poses a significant health risk, causing more deaths annually than breast and prostate cancer combined. Impacting over 10% of the worldwide population, it affects upwards of 800 million individuals. Kidney biopsy, encompassing both open and percutaneous methods, is the gold standard for diagnosing and guiding the treatment of CKD.
 
In pathological image analysis, particularly in kidney disease, tissue segmentation is of paramount importance. The rise of deep learning has been transformative in kidney pathology image segmentation, yet it has also exposed a lack of comprehensive benchmarks for developing and evaluating these techniques. A major hurdle has been the scarcity of large-scale disease data in existing public datasets for kidney pathology segmentation, as they predominantly comprise samples from normal patients. This is mainly because the tissue samples from humans are typically obtained through needle biopsies, yielding only small tissue samples. Consequently, there's a pressing need to release extensive kidney pathology digital data spanning various CKD disease models.
 
In our challenge, we've expanded the dataset from CKD disease models by utilizing preclinical animal models, particularly whole kidney sections from diseased rodents. The primary rationale for using rodent data is the morphological similarity between rodent and human kidney pathologies, making them a prevalent choice in pre-clinical medical research and drug discovery. Secondly, whole kidney sections can be sourced from comprehensive disease models, providing an abundance of tissue in each whole slide image (WSI). This is a significant advantage, as a single WSI from these models can encompass more tissue content than what would be achievable from thousands of needle biopsies in human disease models, an approach that is often impractical.
 
The Kidney Pathology Image Segmentation (KPIs) challenge encompasses a broad spectrum of kidney disease models, including normal and multiple specific CKD conditions, derived from preclinical rodent models. As a pioneering effort in the MICCAI community, the challenge features an extensive collection of 10,000 normal and diseased glomeruli from over 60 Periodic acid Schiff (PAS) stained whole slide images. Each image includes nephrons, with each nephron containing a glomerulus, and a small cluster of blood vessels. The objective for participants is to develop algorithms that can precisely segment glomeruli at a pixel level. To the best of our knowledge, this represents the first MICCAI challenge focused exclusively on segmenting functional units in kidney pathology across various CKD disease models.

Files

Kidney Pathology Image Segmentation Challenge 2024.pdf

Files (97.8 kB)