S2447
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
established models provide valuable tools for individualized risk assessment in clinical practice. References: 1. Joseph, N., et al. Post-treatment lymphocytopaenia, integral body dose and overall survival in lung cancer patients treated with radical radiotherapy. Radiother Oncol. 2019; 135: 115- 119.http://doi.org/10.1016/j.radonc.2019.03.008.2. Shin, J., et al. HEDOS-a computational tool to assess radiation dose to circulating blood cells during external beam radiotherapy based on whole-body blood flow simulations. Phys Med Biol. 2021; 66(16).http://doi.org/10.1088/1361-6560/ac16ea.3. Palma, G., et al. Normal tissue complication probability (NTCP) models for modern radiation therapy. Semin Oncol. 2019; 46(3): 210- 218.http://doi.org/10.1053/j.seminoncol.2019.07.006. Keywords: NTCP model,lymphopenia,lyman model Digital Poster 2748 Fully automatic selection of vascular input function for dynamic contrast-enhanced (DCE) MRI applied to brain, head and neck, and prostate Teo Asplund Research, RaySearch Laboratories AB, Stockholm, Sweden Purpose/Objective: Computing perfusion-related parameters often requires a vascular input function1 (VIF). Manual selection (e.g., from a point inside a large blood vessel), introduces reproducibility issues. A population-based VIF can be used2, but if it differs significantly from the patient-specific VIF this can negatively affect further analysis3. Here, an automatic approach for finding patient-specific VIFs, applicable to multiple anatomical sites, is presented. Material/Methods: The algorithm follows:Preprocess all DCE-MRI images using a 2D-Gaussian filter ( σ =1.0mm).Let S(x,t) be the signal at x for timepoint t and C(x,t) the concentration converted assuming T1=1440ms for 1.5T or 1930ms for 3T4 using hematocrit5=0.445. Let Cimin=5.0-i mM and Cmax=15.0mM.For each voxel x, such that maxt S(x,t) > 2Iavg, where Iavg is the average intensity of the patient volume at the final timepoint, iterate the following steps until at least 20 curves are found:Discard x if maxt C(x,t) < Cimin or if maxt C(x,t) > Cmax, where i is the current iteration (initially 0).Discard x if Time-To-Peak (TTP = argmaxt S(x,t)) is too early/late (TTP<3 or TTP>15).Finally, find clusters of points belonging to the same structure (e.g., the same blood vessel) using DBSCAN6, a density-based clustering algorithm (with ε =0.75cm and numPts=3).Afterwards, curves are sorted using a
analyses included univariate and multivariate logistic regression to identify SRIL predictors, with Bonferroni correction applied in multivariate analysis. Categorical variables were compared using chi-square or Fisher's exact test, and continuous variables with Mann- Whitney U test. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), Brier score, and stratified 5-fold cross- validation. Calibration was assessed using chi-square goodness-of-fit test and calibration curves. Results: The incidence of SRIL was 61.7% and 32.4% in the IMRT and IMPT cohorts, respectively. Blood generalized equivalent uniform dose was an independent predictor of SRIL in the IMRT cohort (OR=4.682, p=0.002). The NTCP model demonstrated strong predictive power in both cohorts (IMRT: AUC=0.82; IMPT: AUC=0.80). The volume effect parameter (a) was 19.85 for IMRT and 2.35 for IMPT. External validation of the IMRT-derived model in the IMPT cohort revealed suboptimal calibration (calibration slope=0.54), indicating systematic overestimation of risk.
Figure 1 Box plots show the distribution of absolute lymphocyte count (ALC) over time during radiation therapy for patients in the (a) photon and (b) proton treatment groups.
Figure 2 NTCP model parameter fitting for photon and proton therapy cohorts. (a) Parameter estimates for the photon cohort: D ₅₀ = 7.44 Gy (6.53–8.31 Gy), m = 0.42 (0.28–0.63), a = 19.85 (1.14–29.60). (b) Parameter estimates for the proton cohort: D ₅₀ = 3.68 Gy(RBE) (2.74–5.45 Gy(RBE)), m = 0.56 (0.26–1.00), a = 2.35 We developed the first modality-specific NTCP models based on whole-blood DVHs for predicting SRIL in lung cancer patients receiving radiotherapy. Our results (0.61–14.66). Conclusion: demonstrated a strong correlation between hematologic dose parameters and lymphocyte depletion, indicating the potential to estimate SRIL risk using blood dose for both IMRT and IMPT. The
Made with FlippingBook - Share PDF online