Abstract Book

ESTRO 37

S508

2 AOU Careggi, Medical Physics Unit, Florence, Italy 3 Skåne University Hospital, Department of Hematology- Oncology and Radiation Physics, Lund, Sweden 4 Lund University, Medical Radiation Physics- Department of Clinical Sciences, Lund, Sweden 5 AOU Careggi, Radiotherapy Unit, Florence, Italy Purpose or Objective The use of optical surface (OS) scanning for patient setup verification is increasing. To manage deformations of patient’s anatomy the Catalyst™ (C-RAD Positioning AB, Uppsala, Sweden) OS scanning system uses a deformable image registration (DIR) algorithm. In this study we have investigated the accuracy of Catalyst™, in respect to the registration algorithm and the reference surface, using a deformable phantom and The Catalyst™ projects light patterns onto the object and the reflected light is acquired by a CCD camera to create a 3D surface model. Patients set-up is verified by matching acquired images with a reference image extracted from CT or acquired with the OS system. An in-house build deformable phantom called "Sliced Mary" (Fig.1a) was used. It was capable of realistic body movements and deformations, such as head and arm rotations, body torsion and moderate breast/abdomen swelling. It contained anatomical details (Fig 1.b) and 10 inner targets at different depths. The phantom was deliberately manually deformed and positioned at the iso-center with the Catalyst™ guidance. 8 deformations were tested (torsions around x, y and z axes; breast and abdomen enlargement) on 4 targets (Fig. 1c). For each deformation a CBCT was acquired. For comparison purposes, the same procedure was repeated using also 3 rigid displacements (rotations around x, y and z axes). For CBCT matching a local approach, focused on the target instead on the surrounding anatomical structures, was used. To evaluate the Catalyst™ accuracy, the difference in means of absolute values between CBCT and Catalyst™ registration (x, y, z, rot, pitch and roll) were used. Two reference images were evaluated, CT and OS, also, a rigid surface registration algorithm was compared to the DIR algorithm. A two tailed Student t-test was carried out (p<0.05). CBCT data as reference. Material and Methods

4DCT for every degree of gantry rotation. The template associated with the gantry angle closest to the projection image was matched to the band-pass filtered kV image using the normalized cross-correlation as a measure of similarity. This resulted in 2D PBT positions in rotating imaging axis coordinates. Triangulation with multiple other registrations was used to determine the 3D position. The range of motion for each resulting motion trajectory was compared to the motion observed on the planning 4DCT. To evaluate the accuracy of PBT motion monitoring, CBCT projections of a 3D printed thorax phantom, consisting of soft tissue, bony structures, airways, and lungs with blood vessels > 1 mm and 3 lung tumors, were acquired. Breathing motion (irregular, with drift, peak-to-peak amplitude ~2.5 mm in lateral, ~20 mm in longitudinal, and ~7 mm in the vertical direction) were simulated by automatically moving the treatment couch in developer mode (TrueBeam, Varian Medical Systems) during image acquisition. Results For the phantom dataset, the 2D position was determined in 89% of the images. In the remaining frames, the bronchus was not visible on the kV images. The 3D root- mean-square error was 2.6 mm (Figure A). For the clinical data, the percentage of frames for which 2D position could be determined ranged from 71 – 91% for the full-fan CBCT scans. The longitudinal motion in the 3D trajectories ranged from 5 – 16 mm, while on the planning 4DCT the motion ranged from 4 – 8 mm. For 1 patient, the motion was within the range of the 4DCT for both CBCT scans, for the other 3 patients the motion exceeded the motion of the planning 4DCT for both CBCT scans (range discrepancy: 1 - 9 mm). Figure B shows a motion trajectory of the patient with the largest discrepancy between motion on the 4DCT and CBCT projections.

Conclusion Markerless template matching + triangulation using kV projection images acquired during gantry rotation can be used to monitor the position of the PBT on a standard LINAC. The motion of the PBT can differ between fractions and may show substantial differences compared to the motion on the 4D planning CT. PO-0937 Rigid and deformable registration in patient set-up verification with optical surface scanning. S. Pallotta 1,2 , M. Kugele 3,4 , L. Redapi 1 , L. Marrazzo 2 , C. Talamonti 2 , L. Livi 5 , S. Ceberg 4 , G. Simontacchi 5 1 University of Florence, Department of Biomedical Experimental and Clinical Sciences "Mario Serio", Florence, Italy

Results The mean absolute differences between Catalyst™ and CBCT registration, using OS as reference (Fig.2 a), were less than 1.8mm 1.1mm 1.3mm in lat long and vrt directions respectively and 0.6° 0.8° 0.6° in rot, roll and pitch, respectively. Small differences between DIR and the rigid algorithm were found and statistically significant only for y (p=0.02) and pitch (p=0.04). The mean absolute differences between Catalyst™ and CBCT increased when using the surface extracted from CT as a reference (Fig.2 b).