ESTRO 2020 Abstract book

S1024 ESTRO 2020

1. Allozi R., Li X.A., White J., Apte A., Tai A., Michalski J.M. Tools for consensus analysis of experts' contours for radiotherapy structure definitions. Rad. Oncol. 2010;97:572–578. PO-1750 Comparison of MRI coils and RT-positioning systems for optimal MRI scanning in SRS cases V. Mengling 1 , R. Perrin 1 , F. Putz 1 , R. Fietkau 1 , C. Bert 1 1 Universitätsklinikum Erlangen, Strahlenklinik, Erlangen, Germany Purpose or Objective The majority of MRI images for treatment planning are acquired in a classical radiological setting. As image registration is a major source of error, positioning in treatment position is desirable. The purpose of this work was to test an MR‐compatible positioning system with novel coil set‐up for stereotactic radiosurgery (SRS) patients that allows MRI in treatment position. Material and Methods SRS‐planning images were acquired on the 1.5 T field strength MAGNETOM Sola equipped with the RT Pro Edition (Siemens Healthineers, Erlangen, Germany). The MRI is equipped with a QFix flat tabletop (Qfix, Avondale, USA) and coil holders to reproducibly position the coils. As the frame and mask fixation device used in our clinic (Brainlab, Munich, Germany) are not MR‐compatible, a wooden replicate of the frame was built. This prototype interfaces with the QFix system and allows patient positioning with exactly the same neck flexion as in treatment position. Two 18‐channel coils (UltraFlex Large) were wrapped around the mask system meeting at the top and bottom for full coverage of the patient head. Additionally, dedicated RT‐planning MRI sequences were implemented, incorporating 3D‐distortion correction, isotropic acquisition and active shimming. To assess the relative image quality of the novel setup, 30 patients were measured prospectively with T1‐MPRAGE sequences in two set‐ups: 1) in the standard radiology setup (BioMatrix Head/Neck 20) and 2) the novel RT‐setup. The potential trade‐off in signal‐to‐noise ratio (SNR) and occurrence of motion artifacts (due to lack of immobilization in the radiology coil) was assessed. The SNR was measured using regions of interest (ROIs), in the anterior, central, and posterior part of the head placed in the white matter and the air at the same distance from the coils. The mean signal in the white matter ROI was divided by the standard deviation of the noise ROI. Results All images in the novel set‐up were suitable for contouring of small metastases as assessed by an experienced radio‐ oncologist. Of the 30 patients measured in the Head/Neck coil, four patients showed severe motion artifacts that were clearly visible inside or around metastases. No motion artifacts were observed in the mask setup. The mask setup presented in this work showed mean SNR of 171±30 in the anterior part, 104±23 in the central and 77±12 in the posterior part of the head. The mean SNR in the radiology setup was found to be 109±21 in the anterior part, 102±20 in the middle and 121±16 in the posterior part of the head.

The mean DSC and MSD between the propagated and clinician‐drawn contours is shown in Figure 1. Many of the propagated contours show reasonable agreement with the clinicians’ contours. The brain, brainstem, CTVs, GTVs, orbits, parotids and spinal cord have a DSC greater than 0.7 and an MSD within 2‐3mm.

Figure 1 Mean dice similarity coefficient and mean surface distance for all contour across over all patients. The distribution of these contours is shown in Figure 2 where it can be seen that results are mostly consistent across patients with a few exceptions which were found to be due to extreme anatomy changes (surgery, swelling etc.). The mandible performed worse than might be expected considering the contrast with surrounding tissue but many of the scans had significant dental artefacts. The other contours with poor scores are small organs similar in size to the voxel resolution which the algorithm may struggle to match.

Figure 2. Dice similarity coefficient for the organs with a mean DSC greater than 0.7 shown for each patient. Conclusion Many of the contours propagated through the deformable registration showed reasonable agreement with the contours created by the clinicians. Velocity deformable image registration is a viable tool for saving time via contour propagation, though post‐propagation editing may be necessary; however, for smaller‐volume organs it is probably quicker to delineate manually.