ESTRO 37 Abstract book

S1147

ESTRO 37

by creating so-called synthetic computed tomography (synCT) images from the MR image with a wide variety of approaches. Figure 1 shows histograms of the ED and MR signal from corresponding images. It suggests that the dominant peaks in the ED histogram from CT have a correspondence in the MR contrasts. This poses the question which degree of precision in ED information from the CT alone is required for dose planning. To potentially limit the complexity and computational effort of synthetic CT generation, we investigate the impact of simplifying the ED information in the CT on dose planning.

Results SimED using a mean body ED as a single tissue class mostly failed the criterion, whereas water ED passed for four out of six patients. Two tissue classes representing fat and remaining soft tissue improved the result to five out of six patients. Best results were obtained for four tissue classes, adding air and bone, for which all patients passed the criterion. The results for dose volume metrics were in agreement with gamma pass rate analysis. Conclusion For MR-based treatment planning, synthetic CTs with two tissue classes, which could be segmented from an MR image, yield acceptable deviation from the standard CT- based plan, whereas one tissue class might not generally be sufficient. Four classes suggested fully acceptable dose distributions, while the segmentation of air and bone in the MR image is still a major challenge. EP-2086 Synthetic CT generation from standard T2- weighted MR image – lung, vertebrae, and liver targets L. Koivula 1 , F. Guerreiro 2 , B.W. Raaymakers 2 , J. Korhonen 3 1 Helsinki University Hospital, Department of Radiation Oncology, Helsinki, Finland 2 University Medical Center Utrecht, Department of Radiotherapy and Imaging Division, Utrecht, The Netherlands 3 Kymenlaakso Central Hospital, Medical Imaging and Radiation Therapy, Kotka, Finland Purpose or Objective MRI-only based radiotherapy treatment planning (RTP) and MRI-guided online adaptive RTP require the generation of a synthetic CT (sCT) from an MR image. Previously proposed sCT methods focus separately on particular body parts and treatment sites (mainly head & pelvis), which limits the sCT method application. This current work aims to develop a technique enabling the sCT generation for whole body standard T2-weighted MR scan. Material and Methods The proposed sCT method relies on an automatic-atlas- based segmentation followed by MR image intensity conversion to synthetic Hounsfield Units (sHU). Atlas- based segments were used to separate bones, air cavities and soft tissues. The method included separate MR intensity to sHU conversion algorithms for each of the segments. The segmentation and conversion algorithms were scripted into a commercial medical image

Material and Methods Computed tomography (CT) images from six pelvic cancer patients were translated into relative electron density (ED) maps using a clinical Hounsfield lookup table (HLUT). The histograms of the ED maps were used for segmentation of the maps. Seven simplified ED maps (simED) per patient were produced with one, two or four tissue classes. Dose plans were optimized based on the original CT (orgCT) and all simEDs. Uncertainties from patient set-up and HLUT definition were considered during dose calculation to estimate confidence intervals for a set of dose volume metrics (figure 2). The intervals from orgCT were then used as pass or fail (i.e. at least one metric outside the interval) criterion for all dose volume metrics per simED.