Abstract Book

ESTRO 37

S549

Material and Methods Thirty patients undergoing hypofractionated prostate RT, with four implanted cylindrical gold FMs (1 mm diameter, 5 mm length), had repeated MR imaging sessions in a multicenter Medical Ethics board approved study. These sessions included cine-MRI examinations prior to each of five weekly fractions. Patient positioning and immobilization was similar to that during RT. A balanced 3D gradient echo sequence was used, optimised for anatomical and FM contrast. Each cine-MR examination consisted of 55 3D datasets (‘dynamics’) acquired sequentially at the start of each imaging session. Each dynamic was acquired over a 11 second period, with a voxel size 0.96×0.96×2 mm 3 and a 384×384×120 mm 3 field of view, yielding a total acquisition time per examination of 10 minutes, similar to the duration of a RT fraction. A clinician manually identified the top and bottom of each FM on the first dynamic of each cine-MR dataset, to determine the FM centre precisely. FM centres in subsequent dynamics were determined automatically using in-house developed MATLAB code (MathWorks Inc., Natick, MA, USA) by means of the watershed transform applied locally around each FM (radius 7 mm). FM patterns from the first dynamic were correlated by template matching with subsequent dynamics. Rigid transformations of the FM template between first and subsequent dynamics were calculated, yielding the center of mass (COM) translation and template rotations. FM registration errors (FRE) and rotational angle errors, the difference between automatic detected FM and FM positions derived from rigid transformation, were calculated based on the assumption of FM template In total, 141 cine-MR examinations (7,755 dynamics) were acquired. The average absolute fiducial registration error was 0.4 mm whereas the average absolute rotation angle error was 0.7°, confirming both a high rigidity as well as high automatic registration accuracy. Intrafraction motion was strongly patient and session dependent, affected by factors including bladder and rectal filling. Figure 1 is an example of COM translations in a single cine-MR exam. A clear but small (~ 1mm) overall group mean shift towards posterior and inferior was observed (Figure 2). rigidity. Results

Conclusion The present study shows that our method for automatic prostate FM detection in 4D cine MR data is robust and accurate. This dataset was acquired under conditions that are representative for hypofractionated prostate treatment and will help to develop real-time MR- guidance on an MRI linac system. PO-0989 Identification of reproducible and non redundant intra and inter CT scanner Radiomics features. R. Berenguer 1 , V. De la Vara 1 , M.D.R. Pastor-Juan 2 , M.V. Villas 1 , J. Canales-Vazquez 3 , M. Castro-Garcia 3 , S. Sabater 1 1 Hospital General de Albacete, Servicio de Radioterapia, Albacete, Spain 2 Hospital General de Albacete, Servicio de Radiologia, Albacete, Spain 3 University of Castilla-La Mancha, Renewable Energy Research Institute, Albacete, Spain Purpose or Objective Radiomics is a growing field of research but results can be influenced by image acquisition techniques. Our aim was to identify reproducible and non redundant intra and inter CT scanner Radiomics features (RFs). Material and Methods Three different analysis were carried out to evaluate reproducibility. Firstly a phantom was imaged two days apart to carry out the TEST-RETEST analysis. Secondly the same phantom was imaged changing all the CT acquisition parameters one by one in the same scanner to evaluate the reproducibility of the RFs. Finally a second phantom with 10 different materials (Image 1) was imaged by 5 different scanners with exactly the same acquisition parameters to evaluate the inter-CT reproducibility of the RFs. In brief, the common procedure of these 3 analysis was the following: a number of ROIs were segmented on the first or reference