ESTRO 2020 Abstract book

S1000 ESTRO 2020

transformation was applied to the histopathology landmarks. See fig 2.

We have established a relationship between geometric contour errors and dosimetric impact, fitting a cohort- valid model to predict changes in clinical DVH statistics for bladder and rectum in prostate VMAT, based on contour error magnitude and location relative to PTV. Hence, we have demonstrated the possibility of dosimetric impact analysis prior to dose planning. Such analysis will reduce resources necessary for automated workflows in which contours are automatically generated or deformed for adaptive RT, by focusing human intervention only where dosimetrically important. PO-1716 Histology correlation of in-vivo [68Ga]PSMA- PET/MRI: A method description and uncertainty estimation K. Sandgren 1 , J. Jonsson 1 , A. Keeratijarut Lindberg 1 , T. Näsmark 1 , S. Said 1 , M. Staneva 1 , A. Bergh 2 , K. Riklund 1 , A. Widmark 1 , T. Nyholm 1 1 University Umea Norrlands Universitetssjukhus, Radiation Sciences, Umea, Sweden ; 2 University Umea Norrlands Universitetssjukhus, Medical Biosciences, Umea, Sweden Purpose or Objective In radiotherapy PET and MRI are two important imaging methods, providing both morphological and functional information. By correlating histopathology to in-vivo images we can learn what is actual visible in the images. Here we present a method, and an estimation of its uncertainties, that can be used to correlate histopathology to in-vivo PET/MRI data of prostate cancer. Material and Methods Full diagnostic PSMA-PET/MRI (3T SIGNA GE) was performed on 50 patients prior to radical prostatectomy. Three-plane T2w MRIs were used to delineate prostate volumes. Individualized 3D-printed molds were created using a pre-created CAD model designed with 5 mm slits. The specimens were placed in their mold directly after surgery, to preserve the in-vivo shape of the prostate. High resolution ex-vivo T2w MRI of the specimen was performed before formalin-fixation (within 90 minutes after the specimen left the patient´s body). The position of the MRI slices was localized according to the slits in the mold. The formalin-fixed specimen was sliced from apex to base in 5 mm thick slices by the pathologist using the built-in slits in the mold. The standard procedure of PAD analysis was performed including paraffin embedding and microtoming in 5 µm thick slices. The slices were digitalized, and all tumors annotated. A 2D affine registration was used to register the histopathology to the ex-vivo images on a slice by slice basis. The affine transform was used to account for specimen shrinkage during pathology preparations. The registered histopathology was then rigidly registered to the in-vivo MRI using the ex-vivo MRI as an intermediate. Both transforms were applied to the annotations. See fig 1.

Fig 2. a) Landmarks identified in in-vivo image (blue), b) corresponding landmark identified in the deformed histology (pink) and c) transformed landmark (pink) and in-vivo landmark (blue). Results This is an ongoing study and our experience is that the procedure provides accurate mapping of pathological findings to in-vivo PET/MRI data. The uncertainty estimation was done in 16 patients. 122 landmarks were identified. The average magnitude of the offset in the transverse plane were: Dx = 1.4(1.8) mm and Dy = 1.6(1.9) mm. Twenty-five landmarks were multiply observed, the offset of these were Dx histo = 0.13(0.25) mm, Dy histo = 0.24(0.55) mm and in-vivo Dx invivo = 0.27(0.72) mm, Dy invivo = 0.29(0.79) mm. Conclusion We present a time-efficient, non-expensive method that can be used to correlate histopathology to in-vivo images. PO-1717 4D-segmentation-based anatomy-restricted motion-compensated reconstruction of on-board 4D CBCT scans M. Susenburger 1 , P. Paysan 2 , I. Peterlik 2 , A. Strzelecki 2 , D. Seghers 2 , M. Kachelrieß 1 1 German Cancer Research Center, X-Ray Imaging and Computed Tomography, Heidelberg, Germany ; 2 Varian Medical Systems, Imaging Laboratory, Baden-Dättwil, Switzerland Purpose or Objective 4D CBCT is primarily used to aid in the setup of lung patients, especially those being treated using SBRT. Current commercially available 4D reconstruction methods are based on binning of the acquired projection data into respiratory-correlated bins such as the phase correlated reconstruction [Medical Physics, 32(4), 1176ff] or the McKinnon-Bates (MKB) algorithm [Medical Physics, 45(8), 3783ff]. In contrast, motion-compensated (MoCo) 4D CBCT image reconstruction performs a deformable motion estimation followed by a reconstruction that accounts for that motion (by letting the voxels move). MoCo outperforms the conventional static-voxel approaches [MedPhys 46(9):3799ff, 2019] but suffers from unrealistic motion patterns, such as a “breathing spine” artifact. This artifact is typically induced by the required regularization (smoothness) of the estimated motion vector fields (MVFs). Our MoCo approach is the artifact- specific cyclic motion compensation (acMoCo) [MedPhys 39(12):7603ff, 2012] [MedPhys 40(10):101913, 2013]. To overcome the problems of this unrealistic cross-organ motion transfer we propose a method based on a 4D segmentation to transfer anatomical knowledge into the motion estimation, e.g. by applying guided bilateral filters and enforcing zero motion on stationary structures. Material and Methods Our thorax CBCT scans have been acquired with a Varian TrueBeam radiotherapy system using a full arc with shifted detector. The projections have been sorted into 20 overlapping respiratory bins (5% step size with 10% width) followed by FDK reconstruction. Then the acMoCo algorithm estimates the motion and generates the MVFs. One of the 20 reconstructed phases is segmented into three anatomical regions (lungs, spine and ribs) and

Fig 1. a) A model of a prostate-mold, b) ex-vivo MRI, c) histopathology with annotations and d) in-vivo PET/MRI with co-registered annotations. To evaluate the uncertainties of the method, three medical physicist student identified landmarks in the in- vivo image and the corresponding landmark in the transformed histopathology. All landmarks were delineated and compared in position, after the rigid