S1461
Interdisciplinary - Other
ESTRO 2026
stakeholder inclusion and explicit articulation of underlying values, promoting transparency, equity, and sustainability in complex care decisions. References: Funtowicz S., Ravetz J.R. Science for the post-normal age. Futures, 1993/2020. 25(7): 739–755. https://doi.org/10.21428/6ffd8432.8a99dd09 (Ripubblicato nel 2020 https://commonplace.knowledgefutures.org/pub/6qqf gms5/release/1)Waltner-Toews D, Biggeri A, De Marchi B, Funtowicz S, Giampietro M, O’Connor M, Ravetz JR, Saltelli A, van der Sluijs JP. Pandemie post-normali. Perché CoViD-19 richiede un nuovo approccio alla scienza. Recenti Prog Med 2020;111(4):202-204. doi 10.1701/3347.33181Cristina Manga, Alba L’Astorino. Cos’è e cosa non è la scienza post- normale, Intervista a Silvio Funtowicz in Scienza Politica e Società l’approccio post-normale in teoria e nelle pratiche. Edizioni CNR, 2023, 43-47, doi: 10.26324/SIA1.PNS3 Keywords: post-normal science, decision making, uncertainty Molecular and Microenvironmental Profiling from Histopathology for Precision Radiation Oncology Xiaohan Xing, Lei Xing Radiation Oncology, Stanford University, Palo Alto, USA Purpose/Objective: The tumor microenvironment (TME) strongly shapes radiation sensitivity, contributing to variability in treatment response, resistance, and recurrence. Although TME features can be reflected by spatial transcriptomics (ST) and histopathology, ST is impractical for routine use. In contrast, histopathology is widely available but lacks explicit molecular readouts. To address this limitation, we propose a multimodal AI framework that transfers spatial molecular information from ST into histopathology- derived representations, enabling non-destructive inference of radiosensitivity-associated molecular states. This approach may support biologically informed treatment stratification and personalized radiotherapy using standard pathology alone. Material/Methods: We introduce a multi-scale multimodal learning strategy that aligns two large-scale foundation models: Digital Poster 4832 UNI (trained on 100,000 pathology slides) [1] and Visiumformer (trained on 3.94 million ST profiles) [2]. Rather than training a unified model from scratch, we integrate the two modalities through multi-scale contrastive alignment. We curated 355 samples from the HEST-1K dataset [3], comprising 801,157 paired H&E patches and ST spots across 16 tissue types. At
the patch level, we enforce consistency between paired histology and ST embeddings. At the region level—where each region is defined as a cluster of nine neighboring patches—we further constrain cross- modal agreement. To maintain hierarchical coherence, we additionally encourage alignment between each patch and its corresponding parent region. This multi- scale contrastive alignment effectively transfers spatial molecular knowledge from ST into histopathology- based representations, enabling downstream tasks such as gene expression prediction, gene mutation status prediction, and spatial spot classification. Results: We evaluated our framework on two downstream tasks. (1) Gene expression status prediction: Using slide-level features from the BCNB dataset (n=1,058 WSIs) [4], our method showed improved ER/PR/HER2 status prediction. Compared with the pretrained UNI model, the aligned model achieved higher AUC and balanced accuracy (Table1). (2) Spatial spot classification: On the DLPFC dataset (n=12 WSIs) [5], linear probing on patch-level features classified spots into seven cortical layers. Our model substantially outperformed the pretrained UNI baseline, yielding 16.56% and 14.54% improvement for the balanced accuracy and weighted F1 score, respectively (Table 2).
Conclusion: Our proposed multi-scale contrastive alignment method effectively transfers spatial molecular knowledge from ST to histopathology, improving downstream prediction and enabling spatially informed TME characterization. By enhancing molecular inference from routine H&E slides, our framework provides access to radiation-relevant biomarkers and intrinsic molecular heterogeneity without sequencing, supporting biologically informed risk stratification and precision radiotherapy. References: [1] Chen et al. "Towards a general-purpose foundation model for computational pathology." Nature medicine 2024.[2] Han et al. "Towards unified molecule- enhanced pathology image representation learning via integrating spatial transcriptomics." Pattern Recognition 2025.[3] Jaume et al. "Hest-1k: A dataset for spatial transcriptomics and histology image
Made with FlippingBook - Share PDF online