Ronald de Krijger

Princess Máxima

In short, Ronald de Krijger's research entails the following:

Artificial intelligence for risk group classification and staging of Wilms tumors

Wilms tumor are the most common malignant pediatric renal tumors. They are currently treated with preoperative chemotherapy, surgical resection of the primary tumor and postoperative chemo- and/or radiotherapy, according to the UMBRELLA protocol, which is used in most European countries and beyond. Histopathology of the resection specimen is crucial for risk group classification into one of three risk groups and for local staging, both of which are then translated into postoperative management.

However, pathological assessment is inherently subjective, with up to 20% discrepancies between pathologists, according to the literature. Automated analysis with supervised machine learning will lead to robust and reproducible recognition of tissue and tumor components, as we have shown in our preliminary work.

In the current project, we will expand on our previous work to segmentation of all relevant components needed for risk group classification. In addition, we will use the same methodology for detection of staging-related criteria.

Finally, we will combine the risk group classification and staging criteria to come to an integrated diagnosis as now done by pathologists. This will not only be less time-consuming than the current manual inspection, but will also lead to more reproducible classification.

Publications

• Deep Learning and Multidisciplinary Imaging in Pediatric Surgical Oncology: A Scoping Review
  • Cancer Med., January 15, 2025
  • Click here to view the publication
• Automated Deep Learning-Based Classification of Wilms Tumor Histopathology
  • Cancers (Basel), May 9, 2023
  • Click here to view the publication
• Artificial Intelligence in Pediatric Pathology: The Extinction of a Medical Profession or the Key to a Bright Future?
  • Pediatr Dev Pathol, March 3, 2022
  • Click here to view the publication
• Interobserver variability between experienced and inexperienced observers in the histopathological
  analysis of Wilms tumors: a pilot study for future algorithmic approach
  • Diagn Pathol, August 21, 2021
  • Click here to view the publication

Short summary of progress/results March 2026

Building on our dataset established in 2025 (n=130 cases of Wilms tumors), the database has been expanded with 58 French cases of Wilms tumors, resulting in a robust and internationally composed foundation for further model development and validation. Development has begun on a weakly supervised deep learning model for the detection of anaplasia, a rare feature present in approximately 10% of Wilms tumors (around 3 cases per year in the Netherlands) and, when anaplasia is diffuse, it is directly associated with a high-risk profile. Preliminary results show that the model can distinguish this feature with high accuracy (median AUC=0.896). In addition, development has started on a model for the automatic detection of positive lymph node, for which annotations have been done and model development has recently started. These efforts support the development of algorithms within a deep learning–based system for risk group classification and staging of Wilms tumors.

Short summary of progress/results March 2025

The project has made significant progress in its initial phase. We have successfully retrieved and digitized archival pathology slides from the Princess Máxima Center, forming a dataset of 130 Wilms tumor cases. Annotation work is underway, focusing on key features such as anaplasia, nephrogenic rests, and lymph nodes. Since December 2024, a postdoctoral researcher with expertise in pathology-AI has been contributing to model development, now working full-time as of March 2025. This allows for the development of algorithms to establish a deep learning-based system for risk group classification and automated staging of Wilms tumors.