Cross-Organ and Cross-Scanner Adenocarcinoma Segmentation Challenge

The field of digital pathology has witnessed remarkable advancements. Gland segmentation, particularly in adenoma regions, has emerged as a central focus due to the widespread occurrence of adenocarcinomas in bodily tissues. Key challenges such as CAMELYON16/17, GlaS, DigestPath, have played a pivotal role in propelling segmentation algorithms forward in digital pathology.
 
Despite these advancements, the efficacy of current algorithms encounters a significant challenge due to the inherent diversity present in digital pathology images and tissues. The variances arise from diverse organs, tissue preparation methods, and image acquisition processes, resulting in what is termed as domain-shift. This phenomenon markedly impacts the performance of machine learning algorithms when transitioning from one organ or laboratory to another. Consequently, research on domain adaptation and domain generalization for pathological images has gained momentum, exemplified by competitions like MIDOG 21/22. 
 
In this challenge, we have digitally captured over 800 patch pathology images spanning five distinct adenocarcinomas, utilizing five different types of scanners. The primary objective is to formulate strategies that empower machine learning solutions to be robust against domain-shift, ensuring consistent performance across diverse organs and scanners employed in image digitization.
 
This challenge holds profound significance in advancing the development of machine learning-based algorithms for routine diagnostic applications in laboratories. Particularly noteworthy, it marks the inaugural challenge in histopathology, offering a platform for the comparative evaluation of domain adaptation methods on a competitive, extensive dataset featuring different organs and scanners from various manufacturers. Anticipated outcomes encompass valuable insights into domain generalization approaches applicable to pathology images at large.