Anne-Roos Schrader

LUMC

In short, Anne-Roos Schrader's research entails the following: 

Artificial Intelligence for Diagnosis of Cutaneous Lymphomas – the AID-CLYM study  

'Cutaneous lymphoma (CL) is a rare and heterogeneous group of hematological skin cancers that is largely distinct from its systemic counterparts in clinical presentation, histopathology, and biological behavior. Therefore, CLs require different therapeutical management. For correct classification and therapeutic management, patients with (suspicion of) CL need timely referral to an expert center. In most CL subtypes, however, diagnosis is challenging and cases are easily mistaken for inflammatory dermatoses, such as eczema or drug reactions.

Recent studies demonstrated that artificial intelligence (AI) is capable of recognizing patterns in digital pathology images and that these can be used to generate computer-based predictions. This novel AI-based approach, however, has not yet been studied in diagnosis of CL patients. With the funding by the Hanarth Fonds, the departments of Pathology and Dermatology of the Leiden University Medical Center (LUMC), the national referral center for patients with CL, will study whether artificial intelligence can improve the process of referral and diagnostics for patients with these rare variants of skin cancer. For this project, the LUMC will collaborate with the Leiden Institute of Computer Science (LIACS).' 

Brief summary of progress / results December 2024

During the first year of the AID-CLYM study, we established a solid foundation for achieving our research objectives. This was accomplished in part by expanding the datasets used for training and validation. A key element of this effort was the creation of an international validation cohort, for which we established the CLIDIPA registry. To date, 11 centers from various European countries have joined this registry, and significant efforts are underway to collect as much data as possible. More information about the registry and its progress can be found on www.clidipa.org.

In 2024, the first study from this project was published in the Journal of Investigative Dermatology (PMID: 39306030). In this proof-of-concept study, we developed a weakly-supervised deep learning model to classify mycosis fungoides (MF) and MF-like inflammatory skin diseases using only H&E-stained images that performed at a level comparable to three expert pathologists. The next step is to further expand the dataset to train an improved model and validate it internationally using data from the CLIDIPA registry.

Additionally, we laid the groundwork for the supervised component of our AI study. This part focuses on training a cell nucleus classification model to identify lymphocytes and distinguish between reactive and neoplastic lymphocytes in the skin, with immunohistochemistry and immunofluorescence as the gold standard. The pathology images of MF patients are also being linked to clinical outcomes, which is expected to result in the first predictive models for MF progression and survival by 2025. These models will likewise be validated using data from the CLIDIPA registry.

Finally, we have embarked on an innovative AI approach using graph neural networks (see our review paper on this topic via arXiv:2406.12808v3, recently accepted for publication in Medical Image Analysis). In a pilot study, we demonstrated that simple graphs perform comparably to our trained weakly-supervised deep learning models in differentiating MF from MF-like inflammatory diseases. In 2025, we aim to further develop and expand this approach, integrating it with the supervised and weakly-supervised methods.

Brief summary of progress / results February 2024

The AID-CLYM study officially started on December 1, 2023. The team has appointed two enthusiastic researchers who will shape the project over the next 4 years, namely Siemen Brussee and Pieter Valkema. Even though the project has only just officially started, major steps have already been taken in building the dataset that now consists of whole-slide images of more than 1000 skin lymphoma patients. Based on this set, the first pilot models were run to distinguish the most common type of skin lymphoma, mycosis fungoides (MF), from MF-like inflammatory skin diseases. We expect to publish the results shortly. In addition, the research team has laid the foundation for European collaboration by establishing the Cutaneous Lymphoma International Digital Pathology (CLIDIPA) Registry, which will serve as an international validation cohort for the models developed within the AID-CLYM study.