ESTRO 2025 - Abstract Book

S3855

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

4077

Digital Poster Tumor Early Response to Radiotherapy Associates with Peritumoral Microbiota Composition in Oropharyngeal Cancer Patients Benedetta Dionisi Ferrera 1 , Marco Dassie 2 , Eliana Gioscio 1 , Jacopo Iacovacci 1 , Giuseppina Calareso 3 , Nicola Alessandro Jacovelli 4 , Marzia Franceschini 4 , Loris De Cecco 5 , Ester Orlandi 6 , Tiziana Rancati 1 1 Data Science Unit, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 2 Postgraduation School in Radiodiagnostics, Università degli Studi di Milano, Milan, Italy. 3 Radiology Department, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 4 Radiotherapy Department, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 5 Unit of Experimental Oncology, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 6 Radiation Oncology Clinical Department, National Center for Oncological Hadrontherapy, Pavia, Italy Purpose/Objective: Investigate host-related factors that can affect the early response to radiotherapy in oropharyngeal cancer (OC) patients by assessing the potential association between T2-weighted MRI-radiomic features and the salivary microbiota. Material/Methods: Microbiota of head-and-neck-cancer patients (181 patients prospectively collected) from salivary samples was profiled before radiotherapy using 16S rRNA amplicon sequencing . The abundance of bacteria at the genus resolution was quantified using the Metagenomic 16S ThermoFisher pipeline. Genera with relative abundance≥1% in at least 3 patients were considered core salivary genera. We used unsupervised Hierarchical clustering (Euclidean distance, Ward method) on centered log ratio-transformed abundance profiles of core genera to identify clusters of patients with similar microbiota. T2-weighted MRI at diagnosis (T0) and 15 days after the radiotherapy start (T1) were collected and analyzed for the subset of OC patients (n=42). A radiologist segmented tumors manually. We pre-processed MRI images (denoising with a Gaussian filter, bias field correction using N4ITK, and Z-score normalization for standardization), resampled all images to a 2mm isotropic resolution using B-spline interpolation, and discretized MRI intensity values into 32 bins [1]. We performed all preprocessing/feature extraction using Hero imaging with Spaarc-plugin. We removed features with a variance below 0.01 at both T0 and T1. We selected stable features following [2] and removed correlated features (Spearman |ρ|>0.85), keeping the feature with the higher variance. To quantify tumor response, we computed Δ-features as the difference between T1 and T0 values. Further, we considered the Early Regression Index [ERI=(Volume_T0-Volume_T1)/Volume_T0] computed on morphological volumes [3]. We used Kruskal-Wallis and ANOVA tests to assess the association between Δ-features and microbiota-based patient groups. Results: We identified four microbiota clusters, including n=12, n=6, n=6, and n=18 OC patients (Figure-1). The variation of Large-Dependence-High-Gray-Level-Emphasis from the distance-zone matrix (ANOVA p=0.014, Kruskal-Wallis p=0.031) and ERI (ANOVA p=0.036) were significantly associated with the microbiota clustering (Figure-2). Variation in the Large-Dependence-High-Gray-Level-Emphasis highlights variation in large homogeneous regions with high grey-level intensities. Patients in the microbiota-cluster 1 show a more pronounced increase in Large Dependence-High-Gray-Level-Emphasis, possibly due to a more pronounced effect of radiotherapy (inflammation).