ESTRO 2025 - Abstract Book

S3730

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

884

Digital Poster Using post-radiotherapy MRI T2-maps to automatically detect and localize radiation-induced pneumonitis in lung tumor patients Rabea Klaar 1,2 , Moritz Rabe 3 , Kaltra Begaj 1 , Stefanie Corradini 3 , Chukwuka Eze 3 , Claus Belka 3,4,5 , Guillaume Landry 3 , Christopher Kurz 3 , Julien Dinkel 1,2 1 Department of Radiology, LMU University Hospital, LMU Munich, Munich, Germany. 2 Comprehensive Pneumology Center (CPC-M), Member of the German Center for Lung Research (DZL), Munich, Germany. 3 Department of Radiation Oncology, LMU University Hospital, LMU Munich, Munich, Germany. 4 German Cancer Consortium (DKTK), partner site Munich, a partnership between DKFZ and LMU University Hospital Munich, Munich, Germany. 5 Bavarian Cancer Research Center (BZKF), partner site Munich, Munich, Germany Purpose/Objective: Radiation-induced pneumonitis (RP) is a common side effect of lung radiotherapy (RT) and repeated follow-up CT scans (FuCT) are required for RP assessment and monitoring. T2-mapping has been proposed for the characterization of various lung diseases 1,2,3 . Here, we investigated the potential of T2-maps for automated patient stratification and initial localization of the RP-affected lung area. Material/Methods: 22 patients treated with hypofractionated stereotactic body RT at a 0.35 T MR-Linac (MRIdian, ViewRay Inc.) in 3-10 fractions received FuCT 6-18 weeks after end of RT. In 14 patients radiological findings suspicious for RP were found and segmented (RP mask) by a radiologist. At the same time point, T2-weighted 3D-images were acquired in inspiration breath-hold at a 1.5 T MRI-scanner (MAGNETOM Aera/Sola Fit, Siemens Healthineers) using an echo planar single-shot fast spin echo sequence at five TEs ([18, 36, 61, 100, 131] ms). After inter-T2-image registration (using ANTs) and voxel-wise log-signal fitting to generate T2-maps, the RT planning low-field MRI-scans and the FuCTs were registered (Plastimatch) to the TE=18 ms images to propagate RT planning dose and structures and the RP mask onto the T2-maps. Baseline-corrected T2-values were calculated by voxel-wise subtraction of the mean T2 value in a lung region with maximal distance from the tumor. For patient stratification, mean baseline-corrected T2-values were calculated in the RT planning target volume (PTV) and the volume receiving ≥20 Gy (V20, EQD2) without the gross tumor volume (GTV) and assessed in a univariate receiver operating characteristic curve-area under the curve (ROC-AUC) analysis using 5000 bootstrapping samples. The non-parametric Mann-Whitney U test (a=0.05) was used to probe significance. For the voxel-based RP detection task, the baseline-corrected T2-maps of 10 RP patients (with FuCT and T2-maps at comparable time points) were first masked with the RT planning lung and V20 volume mask and then thresholded using the cut-off at the Youden index of the ROC-AUC analysis. The Dice similarity coefficient, sensitivity, precision, segmentation AUC and 95% Hausdorff distance (HD95) were used to assess the performance using the propagated segmented RP mask as ground-truth. Results: Significant differences between RP and non-RP patients and high median ROC-AUC values were found for PTV (median RP/non-RP=14.2/1.6 ms, p =0.003, AUC=0.88) and V20-GTV (median RP/non-RP=5.25/-2.8 ms, p =0.02, 0.80). T2-based RP mask showed moderate agreement with the CT-based RP mask (Figure 1, Table 1) with median Dice/sensitivity/segmentation AUC=0.32/0.48/0.74.