S608 3rd ESTRO Forum 2015 PO-1121 Model based 2D localization of lung tumors by incorporating the motion of diaphragm and heart H. Teske 1 , P. Mercea 1 , K. Giske 1 , R. Bendl 2 1 German Cancer Research Center (DKFZ), Division of Medical Physics in Radiation Oncology, Heidelberg, Germany 2 Heilbronn University, Faculty of Computer Science, Heilbronn, Germany Purpose/Objective: Management of intra-fractional tumor motion plays an emerging role in modern high-precision radiotherapy. Online localization of tumor positions is an essential prerequisite to enable steep dose falloffs close to the edges of the target volume. Approaches based on 2D images to directly track the target may suffer from varying tumor visibility. As a common remedy, the motion of anatomical surrogates correlating with the motion of the tumor is tracked instead to model the tumor positions. However, most approaches are restricted to merely assess the superior-inferior (SI) motion of the target, disregarding its left-right (LR) direction, which in turn needs to be compensated by increased margins. This shortcoming is addressed by a model-based approach that incorporates the motion of the diaphragm and the heart in order to account for both LR and SI tumor motion directions. Materials and Methods: Fluoroscopic image sets of two patients (187 images each) were analyzed retrospectively. The developed approach to localize the position of tumors located in the left lung is based on a) the motion assessment of surrogate structures and b) an appropriate model to extract the tumor localization according to the combined surrogate displacement. The breathing motion, mainly directed in SI, is incorporated by the use of the diaphragm as a surrogate. However, for the LR motion of lung tumors, the heart can be a substantial motion-inducing source. Pre- segmented upper contours of the diaphragm and lateral contours of the heart are used to assess the motion of both surrogates (see Figure 1). In detail, the motion of a surrogate is represented by a fixed amount of 30 equally distributed corresponding points, between two different time steps. All of the corresponding points of both surrogates are used as an input for a thin-plate-spline interpolation in order to model the tumor motion. For reference, tumor positions in each frame for both patients were manually determined by four experts. The results were compared to the mean of those positions. Results: Using only the motion of the diaphragm as a surrogate for tumor motion, the resulting mean localization accuracy of the developed approach was 1.7 mm in LR and 1.1 mm in SI direction. An improvement of the LR accuracy was achieved by additionally incorporating the heart motion. The combination of both surrogates resulted in an accuracy of 0.8 mm in LR and 1.0 mm in SI direction (see Table 1). The mean inter-observer variability accounts for 0.8 mm in LR and 1.3 mm in SI direction. Conclusions: With the presented approach, the breathing as well as the motion of the heart was taken into account in order to achieve accurate 2D tumor localization without relying on tumor visibility. The ability to tailor the approach to the patient- and situation-specific anatomy allows individual tumor motion modeling.