S2478
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
to identify sub-regions at higher risk of relapse on pre- RT MRI and validate population-based thresholds across independent external cohorts, compared with conventional relative Cerebral Blood Volume (rCBV). Material/Methods: We performed a retrospective multicentre analysis including adult patients with HGG, available pre-RT MRI with T2*-Dynamic susceptibility contrast (DSC) perfusion, and MRI at relapse. Clinical and radiotherapy data were collected.At both time points, automatic deep-learning autosegmentation was performed on anatomical MRI (T1, T1-Gd, T2-FLAIR), followed by manual refinement into five tumour sub- volumes: necrosis, FLAIR hyperintensity, contrast- enhancing tumour, surgical cavity, and post-surgical modifications.Relapse images were rigidly and elastically co-registered to pre-RT MRI to ensure voxel- wise correspondence.Perfusion biomarkers (rCBV, CTH, OEF) were reconstructed using Cercare Medical Neurosuite (v17.1, Aarhus, Denmark). Bias-field and leakage correction were applied before map generation; images were resampled to isotropic 1 mm ³ voxels after reconstruction.The relapse volume, defined as the new contrast enhancement appearing after RT, was co-registered to pre-RT space to test biomarker predictive power.Voxel-wise ROC analyses were conducted to establish patient-specific optimal cut-offs, while a population-level threshold was identified using the Youden Index calculated from institutional data (Cohort 1). These thresholds were then applied to two independent external cohorts to assess reproducibility and predictive performance (Cohorts 2 and 3). Results: A total of 186 patients were included in the reference cohort, and 10 and 13 patients were included from the two external validation cohorts.In Cohort 1, voxel-wise ROC analyses demonstrated that advanced perfusion biomarkers outperformed conventional rCBV in predicting future relapse voxels (Figure 1).
unchanged definition) and supports its potential role in guiding RT optimization for HNC tumors. The observed stability of results under contour perturbation further strengthens the reliability of the biomarker. References: Funding: Hi-ROC, EP_PerMed_JTC2024, FRRB_EP_PERMED2024-548Reference:[1] Winter et al. “Clinical validation of a prognostic preclinical magnetic resonance imaging biomarker for radiotherapy outcome in head-and-neck cancer” 2025 Mar:204:110702. doi: 10.1016/j.radonc.2024.110702. Epub 2024 Dec 27.[2] Seetha et al. “Stability of Multi- Parametric Prostate MRI Radiomic Features to Variations in Segmentation” J Pers Med. 2023 Jul
22;13(7):1172. doi: 10.3390/jpm13071172. Keywords: HNC,IMAGING BIOMARKERS,ADC
Digital Poster Highlight 3975
Multicentre validation confirms the superiority of advanced perfusion biomarkers over rCBV for relapse prediction in high-grade glioma Camilla Satragno 1 , Mohammed El Aichi 1 , Ana Stan 1 , Linda Mrissa 1 , Cedric Yuste 1 , Killian Sambourg 1 , Marco Banini 2 , Jade Briend-Diop 1 , Taha Hachemi 3 , Valentina Pinzi 1 , Cristina Veres 1 , Fadwa BOUGUERRA 1 , Anna GUEIDERIKH 1 , Francisco Ioan Velichi 1 , Gabriel Garcia 4 , Samy Ammari 5 , Noura Sellami 1 , Liliana Belgioia 6,7 , Franscesco Lanfranchi 8 , Andrea Bianconi 9 , Julian Jacob 3 , Sophie Bockel 1 , Frederic Dhermain 1 , Eric Deutsch 1 , Charlotte Robert 1 1 Inserm U1030, Radiothérapie Moléculaire et Innovations thérapeutiques, Gustave Roussy, Université Paris-Saclay, Villejuif, France. 2 Radiation Oncology Unit, Careggi University Hospital, Florence, Italy. 3 Radiation Oncology Department, Pitié- Salpêtrière Hospital, Sorbonne University, AP-HP, Paris, France. 4 Imaging Department, Gustave Roussy, Université Paris -Saclay, Villejuif, France. 5 Imaging Department, Gustave Roussy, Université Paris-Saclay, Villejuif, France. 6 Department of health sciences, University of Genova, Genoa, Italy. 7 Dipartimento di Radioterapia, Ospedale Policlinico San Martino, Genova, Italy. 8 Dipartimento di Medicina Nucleare, Ospedale Policlinico San Martino, Genova, Italy. 9 Dipartimento di Neurochirurgia, Ospedale Policlinico San Martino, Genova, Italy Purpose/Objective: Local relapse remains the predominant pattern of failure in high-grade glioma (HGG) despite highly conformal radiotherapy (RT).By characterizing intra- tumour heterogeneity through advanced perfusion biomarkers, Capillary Transit Time Heterogeneity (CTH) and Oxygen Extraction Fraction (OEF), we aimed
Using population-derived thresholds (CTH = 3.79, OEF = 0.44, rCBV = 2.48), CTH showed the highest discriminative power (AUC = 0.71, sensitivity = 0.64, specificity = 0.67), followed by OEF (AUC = 0.62) and
Made with FlippingBook - Share PDF online