S2597
Radiobiology - Preclinical biomarkers
ESTRO 2026
8 Center for Advanced Systems Understanding, (CASUS), Görlitz, Germany. 9 Helmholtz-Zentrum Dresden-Rossendorf, (HZDR), Dresden, Germany. 10 German Cancer Consortium (DKTK), Partner Site Dresden, and German Cancer Research Center (DKFZ), Dresden, Germany. 11 German Cancer Consortium (DKTK), DKFZ, core center Heidelberg, Heidelberg, Germany Purpose/Objective: Radiotherapy is a standard treatment for locally advanced head and neck squamous cell carcinomas (HNSCC); however, patient responses vary. Preclinical models are used to support development of personalized treatment strategies. In this study, we analyzed H&E slides from 10 HNSCC xenograft models by applying unsupervised clustering of deep learning-derived features to link histopathological patterns with radiosensitivity and gene expression. Material/Methods: The samples used here were part of a previous tumor control study [1,2]. Athymic nude mice were xenotransplanted with ten radiobiologically distinct HPV16 negative HNSCC tumor models. Tumor control dose 50% (TCD50) values, obtained from fractionated radiotherapy experiments, were used for classification of the models based on their radioresistance. For the analysis, we used 71 hematoxylin and eosin (H&E)–stained whole-slide images (WSIs) from 68 untreated control tumors approximately 7 mm in diameter.Using the publicly available deep learning model ‘UNI’ pretrained on large histopathological datasets [3], we extracted tile- level features from each slide. All features were standardized using z-score normalization, then dimensionality reduction was performed using principal component analysis (PCA).The Gaussian Mixture Models algorithm was applied to cluster all the features into five clusters. Cluster percentages were calculated, and the median percentages per tumor model were used to evaluate the correlation with TCD50 values.Microarray data from 59 additional untreated tumors were used to validate the clustering results. The data were normalized using the robust multichip average (RMA) method, then gene set enrichment analysis (GSEA) was applied by comparing radiosensitive and radioresistant HNSCC models with respect to the C5 (MsigDB database) ontology gene sets to identify significantly enriched pathways. Results: Unsupervised clustering identified five major clusters representing necrotic and stromal areas, non-keratinized tumor areas (two clusters), tumor tissue with keratin presence, and keratin enriched areas. The median percentage of the latter keratin-enriched cluster per tumor model showed a negative correlation with the TCD50 values (Spearman coefficient -0.75; p-value = 0.001). GSEA analysis supported the identified digital features, with three of the top ten enriched gene sets being keratin-related and associated with radiosensitive tumor models. Conclusion: Unsupervised clustering of H&E slides and GSEA revealed enriched keratinization in radiosensitive HPV16 negative HNSCC xenograft tumors, highlighting the importance of preclinical models in biomarker discovery and reuse of imaging data in accordance with 3R principles. The prognostic value of keratinization and keratin gene expression in HNSCC patients treated with primary radio(chemo)therapy is under investigation and preliminary results will be presented. References: [1] Gurtner et al Radiother Oncol. 2011. https://doi.org/10.1016/j.radonc.2011.05.035 [2] Koi et al. Radiother Oncol. 2017. https://doi.org/10.1016/j.radonc.2017.07.009 [3] Chen et al. Nat Med. 2024. https://doi.org/10.1038/s41591-024-02857-3 Keywords: unsupervised clustering, H&E, HNSCC xenograft
Conclusion: Dynamic OCT paired with radiomics provides a rapid, non invasive, and information rich alternative to current high throughput spheroid assays. By leveraging 3D tomographic images and time variance signatures of intracellular activity, the approach improves prediction of radiation damage over brightfield size proxies, enabling scalable and longitudinal assessments that can accelerate radiobiology studies and therapy optimization. References: [1] Abd El-Sadek I, Miyazawa A, Shen LTW, Makita S, Mukherjee P, Lichtenegger A, Matsusaka S, Yasuno Y. Three-dimensional dynamics optical coherence tomography for tumor spheroid evaluation. Biomedical Optics Express. 2021;12(11):6844-6863. doi:10.1364/BOE.440444.[2] Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RGPM, Granton P, Zegers CML, Gillies R, Boellard R, Dekker A, Aerts HJWL. Radiomics: Extracting more information from medical images using advanced feature analysis. European Journal of Cancer. 2012;48(4):441-446. doi:10.1016/j.ejca.2011.11.036. Keywords: Spheroids, Machine Learning, OCT
Digital Poster 5144
Unsupervised clustering of histopathological images from xenograft tumors links keratin to radiosensitivity of head and neck squamous cell carcinoma Cylia Ouadah 1,2 , Alex Zwanenburg 1,3 , Safayat M. Khan 4,5 , Carina Wenzel 3,6 , Lydia Koi 2,7 , Artur Yakimovich 8,9 , Annett Linge 7,10 , Michael Baumann 4,11 , Maria J. Besso 4,5 , Ina Kurth 4,11 , Antje Dietrich 1,10 , Mechthild Krause 7,10 , Steffen Löck 1,2 , So ň a Michlíková 1,2 1 OncoRay - National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Helmholtz-Zentrum Dresden- Rossendorf, Dresden, Germany. 2 Institute of Radiooncology - OncoRay, Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany. 3 National Center for Tumor Diseases (NCT), NCT/UCC Dresden, a partnership between DKFZ, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, and Helmholtz-Zentrum Dresden-Rossendorf (HZDR), Dresden, Germany. 4 Division of Radiooncology/Radiobiology, German Cancer Research Center (DKFZ), Heidelberg, Germany. 5 Heidelberg Institute for Radiation Oncology (HIRO), and National Center for Radiation Research in Oncology (NCRO), Heidelberg, Germany. 6 Institute of Pathology, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany. 7 Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany.
Made with FlippingBook - Share PDF online