Abstract Book

S1178

ESTRO 37

Purpose or Objective An inherent problem of magnetic resonance imaging (MRI)-based analyses of morphological tissue changes is the absence of a ground truth. A particular issue in cerebral imaging is the lack of consensus regarding the order and manner, in which registration and segmentation algorithms are employed to evaluate volumes and longitudinal changes of different tissue types, e.g., grey and white matter (GM and WM). Considering shortcomings of a procedure widely applied for global segmentation of the entire brain we developed a modular MRI processing workflow particularly suitable for volumetric analysis of the contralateral hemisphere in The modular and easily extendable scheme contains the most relevant image processing tasks, e.g., tissue segmentation and (non-)linear registration. The registration to the labeled Montreal Neurological Institute (MNI) atlas is performed using the Advanced Normalization Tools (ANTs). Tissue segmentation to GM and WM is conducted with the Statistical Parametric Mapping MATLAB-toolbox (SPM). For initialization of the iterative segmentation, which is based on a Bayesian approach, a tissue probability map (TPM) for each tissue class is required. For our specific requirements, we compared the resulting tissue volumes for the TPM of the full brain (full-TPM) with those generated by the TPM limited to the contralateral hemisphere (cl-TPM). The volumetric comparison was repeated varying the number of Gaussians for modeling the frequencies of the tissue classes. The three methods were compared in a cohort of 17 proton-treated glioblastoma patients having undergone T1-weighted (T1w) MRI before and 3 months after radiochemotherapy, in which we have recently shown a treatment-induced brain tissue volume decrease [1,2]. Results brain tumor patients. Material and Methods

Figure 1 shows the developed image processing pipeline. As can be visually appreciated (Fig. 2), the presence of tumor and/or its resection cavity or edema in the T1w- MRI influences the intensity frequencies in the tumor- bearing hemisphere. The initialization of the segmentation using the full-TPM may hence be biased and result in overestimating longitudinal GM volume decrease or even in reporting apparent WM volume increase [3,4]. This can be overcome by using the cl-TPM, which results in a more reliable automatic segmentation. In our cohort, the full-TPM method resulted in a 3.7% higher GM volume decrease than the cl-TPM method, while the latter reduced the false, longitudinal WM volume increase (0.47% vs 0.11%, respectively). The increment in the number of Gaussians had no significant effect on the tissue segmentation (data not shown). Conclusion The developed workflow with cl-TPM-based segmentation has proven to be robust and reduces the bias related to analyzing longitudinal, volume-based MRI data. References [1] Petr J et al., Radiother Oncol, under review [2] Petr J et al. ISMRM 2017 [3] Prust et al. Neurol, 2015, 85(8) [4] Chamberlain MC et al., Neurol, 2016, 86(10) EP-2138 Contouring of moving targets in maximum and average intensity projection. K.J. Borm 1 , M. Wiegandt 2 , A. Hofmeister 2 , M. Oechsner 1 , S.E. Combs 1 , M.N. Duma 1 1 Technical University of Munich TUM, Department of Radiation Oncology, München, Germany 2 Technical University of Munich TUM, Medical School, Munich, Germany Purpose or Objective The use of four-dimensional computed tomography (4DCT) is currently state of the art for radiotherapy planning for lung cancer. A relevant drawback is the time required to contour a target volume in every single phase of the 4DCT. Target delineation in maximum (MIP) and average intensity projection (AIP) CTs allows rapid definition of internal target volumes in a 4DCT. However the current literature is inconclusive on this topic and lacks clinical evidence. This study was performed to assess the error of MIP and AIP contouring with special emphasis on tumor localization analyzing a large patient collective in combination with simulations on a self- developed lung phantom. Material and Methods 4DCT data from 50 patients with lung tumors (diameter 1.1–7.0 cm) were chosen for this study. The tumors merged to the chestwall (n=20), the diaphragm (n=4), the mediastinum (n=16) or were surrounded entirely by lung tissue (n=10). In the lung phantom two spherical structures (diameter: 1 cm and 2 cm) composed by water equivalent synthetic substance (RW3) were embedded in corkboards to simulate a tumor inside lung tissue. Tumor movement of 10 patients were measured and 4DCTs of the phantom were acquired while simulating the movements. For each motion pattern a theoretical target volume was calculated. Internal target volumes (ITVs) were contoured by a single physician in MIP (ITV MIP ) and AIP CTs (ITV AIP ) and subsequently compared to ITVs contoured in all 10 phases of a 4DCT (ITV 10 ) or the calculated values respectively (Fig. 1). Conformation numbers (CN) between ITV 10 and ITV MIP or ITV AIP were calculated according to van’t Riet et al.