Abstract Book

ESTRO 37

S538

ADC map. These data were also acquired for 9 distinct patients on a conventional diagnostic 3T Achieva scanner (Philips Healthcare, Best, The Netherlands) for comparison. Test-retest data was registered using deformable registration. The within-subject voxel-by- voxel coefficient of variation (wCV) was calculated for the set of T 2 - and ADC maps for each patient. A log transform was performed when the variation between the two maps depended on the mean value. The images were smoothed with an increasing kernel size and the wCV was calculated to assess the dependence of the wCV on ROI size. At voxel level, the wCV is mainly influenced by imaging noise, which is reduced by smoothing, leaving the effect of daily variations in the patient and machine. Results The quality of MR-Linac images is represented in Figure 1, showing test and re-test data of both the T 2 - weighted, the T 2 - and the ADC map of the prostate. The wCV values of the T 2 and ADC map are shown in Figure 2 for the MR- Linac and 3T scanner. The wCV at the MR-Linac at voxel- level are 17.5% for the patient and 14.5% for the volunteer for the T 2 maps, and 23.2% and 13.4% for the ADC maps. A reduction in this variation is found with an increase in the smoothing kernel size. With a kernel size of 0.5 mm 3 , the wCV for the T 2 maps are 4.6% and 4.5%, and 7.5% and 4.9% for the ADC maps. The repeatability of the T 2 map is similar in the patients scanned at the 3T scanner, with a median wCV of 18.9% at voxel level and 3.9% for a kernel size of 0.5 mm 3 . For ADC the repeatability at 3T is better with 10.3% at voxel level and 3.5% for 0.5 mm 3 . Conclusion In this study, we showed the feasibility of quantitative imaging on the MR-Linac in prostate cancer patients. When compared to the conventional diagnostic scanner, the MR-Linac produces comparable results for T 2 mapping. For ADC mapping, the MR-Linac shows a greater variability between the two sessions compared to the conventional diagnostic scanner. The voxel-level differences can be explained by the fact that the protocols were not matched to have exactly the same SNR.

PO-0975 Relationship of dose, FDG PET, CT lung response imaging, and radiation pneumonitis in NSCLC patients M. La Fontaine 1 , G. Defraene 2 , J. Van Diessen 1 , S. Van Kranen 1 , B. Reymen 3 , D. De Ruysscher 3 , J. Belderbos 1 , J.J. Sonke 1 1 Netherlands Cancer Institute, Radiotherapy, Amsterdam, The Netherlands 3 MAASTRO, Radiation Oncology, Maastricht, The Netherlands Purpose or Objective Radiation pneumonitis (RP) is an important dose limiting toxicity in NSCLC patients undergoing (chemo- )radiotherapy. Imaging modalities such as PET and CT have shown high inter-patient variability in lung tissue response to dose, yet its association is still unclear with clinically scored RP. This study evaluated imaging metrics in predicting RP, and their effectiveness in comparison to dose metrics. In addition, pre-treatment patient characteristics were investigated to find associations to inter-patient dose sensitivity derived from imaging. Material and Methods 65 patients with inoperable, stage II-III NSCLC were treated with (chemo)radiotherapy as part of the (NCT01024829) PET-boost trial. All patients received an FDG PET/CT scan at 3 months post-treatment, which was deformably registered to the planning CT. For lungs minus GTV, a linear region was found between planned dose and 1) post-treatment FDG PET scan, and 2) pre- to post-treatment CT HU change (delta CT). Slopes of the linear regions for post-treatment FDG PET (PET-slope) and delta CT (CT-slope) were investigated in individual patients for RP prediction (Grade ≥2 according to CTCAE). Due to a low number of RP events in this cohort (6 patients), the F1 score, a harmonic mean of sensitivity and precision, was used. The F1 score (ranging from 0 to 1) was compared between PET- and CT- slopes, as well as for dose (EQD2: α/β =3 Gy) metrics of lung V20, lung V5, mean lung dose (MLD), and mean heart dose (MHD). Univariate analysis was applied (alpha = 0.10) to search for pre-treatment patient characteristics associated with patient sensitivity for dose in CT-slope and PET-slope. Results Post-treatment FDG PET SUV displayed higher linearity (median r: 0.87 [IQR:0.64-0.94], p=0.002) to dose than delta CT (0.67 [IQR:0.28-0.84]) for individual patients. In addition, PET-slope had a higher F1 score than CT-slope with values of 0.59 and 0.40, respectively. In comparison, dose metrics displayed F1 scores of 0.50 (lung V20), 0.42 (lung V5), 0.36 (MLD), and 0.12 (MHD) (Figure 1a). The best identifier for RP was a combination of MLD and PET- slope with a F1 score of 0.83 (sensitivity: 0.83, precision: of 0.83) (Figure 1 b-c). Univariate analysis of patient