Loes Hollestein

Erasmus MC

In short, Loes Hollestein's research entails the following:

AI-Driven Exploration of the Melanoma Tumor Microenvironment in the Dutch Early Stage Melanoma (D-ESMEL) study

The development of distant metastasis after thin melanoma is very rare (e.g. 1% of all stage IA melanoma). In order to study progression of early stage melanoma, we designed the Dutch Early Stage Melanoma Study (D-ESMEL). The D-ESMEL study consists of a discovery set and a population-based validation cohort. From all cutaneous melanoma samples (>750), clinical, multi-omics (RNAseq/DNAseq) and whole slide imaging (digital H&E and multiplex immunofluorescence) is available. In this project we will characterize the interactions in the tumor microenvironment using two AI approaches. First we will apply a segmentation approach, where tumor and immune cell interactions are quantified after Deep Learning based cell segmentation. Second, we will apply a segmentation-free approach where a Deep Learning framework will be applied to learn which cell phenotypes, neighborhoods and area interactions are predictive for prognosis. Thereafter, prognostic features of the tumor microenvironment will be integrated with prognostic features from
RNAseq, DNAseq, clinical and histopathological data.

Brief summary of progress / results March 2026

The project's goal is to improve the identification of patients at risk of distant metastasis by studying the local immune response. We have generated spatial transcriptomic data to collect clues on specific immune markers in melanomas related metastasis, which can be utilized to determine which prognostic markers show potential for investigation in 5-plex immunofluorescence data. Currently, we are developing an end-toend workflow to predict distant metastasis in an internal melanoma multiplex immunofluorescence dataset which can be utilized for the multi-plex data to be generated. Data integration analysis, excluding spatial data, is being finalized and the manuscript is being prepared. This work lays out the foundation for future data integration which incorporates spatial immune data.