ESTRO 2024 - Abstract Book

S3831

Physics - Image acquisition and processing

ESTRO 2024

After approval by the Ethics Committee of the Technische Universität Dresden, dual-energy CT (DECT), T1- and T2 weighted MRI data of 30 prostate cancer patients were retrospectively retrieved. sCTs were generated with (1) syngo.via version VB40 (Siemens Healthineers, Erlangen, Germany), a hybrid method relying on BDA for air, fat, soft tissue, spongious bone, cortical bone assessment, and on the multi-atlas technique for bone localization; (2) syngo.via version VB60 (Siemens Healthineers), which predicts the sCT with a generative adversarial network; (3) MriPlanner v2.2 (Spectronic Medical, Helsingborg, Sweden), using an MRI to CT transfer function based on neural networks. For most comparisons, the MRI were deformably registered to the DECT and the resulting transformations applied to the sCTs, while rigid registration between bony structures, identified in (s)CTs with a threshold of 500HU, was used for detailed comparisons of bone (ANTs v2.4.1). For assessment of image fidelity, the Mean Absolute Error (MAE) within the body and bone regions, as well as the Dice Similarity Coefficient (DSC) in bone, were calculated. For the bone MAE and DSC, the region-of-interest was respectively identified with a mixture of Sobel and morphological operators or with a threshold of 100HU. For dosimetric assessments, CT treatments plans (routine photon therapy, prescribed dose to boost Clinical Target Volume [CTV]: 78Gy) were recomputed on sCT (RayStation v11, Collapsed Cone v5.6, RaySearch, Stockholm, Sweden) using vendor-supplied CT calibration curves. Global 2%/3mm gamma indices with 10% and 90% dose thresholds were obtained with Slicer v5.2.2. Additionally, Dose Volume Histogram (DVH) parameters were calculated (D98% and D2% for target volumes; V70Gy for bladder and rectum, V50Gy for rectum, and D2% for femoral heads). Hypothesis testing consisted of Kruskal-Wallis tests followed by Dunn’s tests (Bonferroni adjusted), with a significance threshold of 5%.

Results:

For one example patient, the imaging data used is presented in Figure 1.