Stephanie Peeters

Maastro

In short, Stephanie Peeter's research entails the following: 

Artificial intelligence for histopathological classification and recurrence prediction of thymic epithelial tumors (INTHYM project)

Thymic epithelial tumors are rare neoplasms in the anterior mediastinum. The cornerstone of the treatment is surgical resection. Administration of postoperative radiotherapy is usually indicated in patients with more extensive local disease, incomplete resection and/or more aggressive subtypes, defined by the WHO histopathological classification.

In this classification thymoma types A, AB, B1, B2, B3, and thymic carcinoma are distinguished. Studies have shown large discordances between pathologists in subtyping these tumors. Moreover, the WHO classification alone does not accurately predict the risk of recurrence, as within subtypes patients have divergent prognoses.

We will develop AI models using digital pathology and relevant clinical variables to improve the accuracy of histopathological classification of thymic epithelial tumors, and to better predict the risk of recurrence.

In this multicentric and international project three existing databases will be used from Rotterdam, Maastricht and Lyon. For all models one database will be used to build AI models, and the other two for external validation.

The ultimate goal of this project is to develop AI models that support the pathologist in correctly subtyping thymic epithelial tumors, in order to prevent patients from under- or overtreatment with adjuvant radiotherapy.

Brief summary of progress / results juni 2024

The project aims to develop AI models to improve the accuracy of histopathological diagnoses and recurrence predictions for TET patients. We will use pathology and clinical data from three centers in Rotterdam, Maastricht, and Lyon.

In the first year, a PhD student organized the EMC thymoma panel's digital pathology dataset, which includes diagnoses with 100% and 70-99% consensus. Standardization of whole slide image staining with hematoxylin and eosin was done, followed by detailed annotation of tumor tissue, hemorrhage, fibrosis, and artifacts. Patches were extracted from regions of interest at different magnifications to exclude background and non-tumor tissue, ensuring robust data for model training.

Initial AI models for thymic epithelial tumor classification were developed using 100% consensus data. Model 1, distinguishing cell types A, B3, TC versus B1, B2, showed high accuracy and reliability. Model 2.1, sequentially classifying A, B3, and TC, was also initiated. A clinical database has been established, and the first clinical data entries are underway.