S2467
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
Conclusion: That STP-only model outperformed the STP&MTP model, despite the latter containing richer information from individual growth trajectories. However, substructure growth can be impacted by parameters beyond age, e.g., demographics, which were not included and could improve model flexibility. A shared mean trajectory is useful for defining typical population growth, but modelling approaches that account for individual variation (e.g., additional parameters and matched case-control design) may allow for better prediction of development in children’s brains. References: [1] Jernigan, T.L., et al, 2016. The pediatric imaging, neurocognition, and genetics (PING) data repository. Neuroimage, 124, pp.1149-1154.[2] Casey, B.J., et al., 2018. The adolescent brain cognitive development (ABCD) study: imaging acquisition across 21 sites. Developmental cognitive neuroscience, 32, pp.43- 54.[3] Henschel, L., et al. 2020. Fastsurfer-a fast and accurate deep learning based neuroimaging pipeline. NeuroImage, 219, p.117012.[4] Leroy, A.,et al., 2022. MAGMA: inference and prediction using multi-task Gaussian processes with common mean. Machine Learning, 111(5), pp.1821-1849. Keywords: growth modelling, paediatric, brain Digital Poster Highlight 3567 An MRI-based radiomic biomarker of radiosensitivity for individualized radiotherapy dose prescription in cervical cancer Kohei Oguma 1,2 , Yutaka Shiraishi 1 , Masafumi Sawada 1 , Atsuya Takeda 1 1 Department of Radiology, Keio University School of Medicine, Shinjuku-ku, Japan. 2 Cancer Center, Keio University School of Medicine, Shinjuku-ku, Japan Purpose/Objective: Radiotherapy for cervical cancer is typically prescribed uniformly across patients, without fully considering tumor radiosensitivity, such as the relative radioresistance observed in non-squamous histology. This study aimed to identify an MRI-based radiomic biomarker associated with radiosensitivity and to evaluate its feasibility for guiding individualized dose prescription using dose–response simulation. Material/Methods: A total of 176 patients with cervical cancer treated with definitive radiotherapy were retrospectively analyzed and divided by treatment period into a development cohort (n = 122, 2015–2020) and a test cohort (n = 54, 2020–2022) (Table 1). Radiomic features were extracted from the gross tumor volume and high-risk clinical target volume (HR-CTV) on pretreatment T2-
weighted MRI after standardized preprocessing. In total, 2,161 reproducible radiomic features verified for inter-observer stability, 30 clinical factors, and a dosimetric parameter defined as the minimum equivalent dose in 2 Gy per fraction (EQD2) covering 90% of the HR-CTV (HR-CTV D90) were comprehensively analyzed to identify radiosensitivity- related radiomic biomarkers. A survival model for local control (LC) was developed based on the Cox proportional hazards model to simulate dose– response relationships and estimate the HR-CTV D90 required to achieve a 90% 2-year LC for individual patients and subgroups stratified by radiosensitivity- related factors, such as the radiomic biomarker and histology.
Results: The radiomic feature Long Run High Gray Level Emphasis (LRHGLE) was significantly associated with LC and effectively characterized the relationship between delivered HR-CTV D90 and the 2-year LC rate. The survival model demonstrated high predictive performance (concordance index = 0.905) on the test cohort, outperforming all comparator models (P < .001). The dose–response simulation agreed with observed outcomes and revealed substantial interpatient variability in the dose required to achieve LC. The required dose was considerably higher in radioresistant subgroups (P < .05): 66.4 Gy (95% CI, 56.3–69.1 Gy) vs 91.3 Gy (95% CI, 76.3–122.1 Gy) for LRHGLE (low vs high), and 67.1 Gy (95% CI, 64.1–71.6 Gy) vs 99.5 Gy (95% CI, 78.5–122.5 Gy) for histologic subtype (squamous vs non-squamous) (Figure 1). These findings suggest that a uniform prescription of HR-CTV D90 = 70–85 Gy may result in under- or overtreatment depending on tumor radiosensitivity.
Made with FlippingBook - Share PDF online