ESTRO 2020 Abstract Book

S397 ESTRO 2020

plans in each CT series. The plot of all evaluation scenarios, each with its corresponding DVH, results in DVH- bands (Figure 1). Local dose differences between restored/distorted and initial dose distributions were also analysed by reporting absolute errors in smalls volumes DE(vol=2%).

important anatomical changes, as it is already the case in X-ray radiotherapy.

OC-0706 First MRI-guided MLC-tracking using a deformable motion phantom P. Borman 1 , P. Woodhead 2 , M. Perrin 3 , E. Barberi 3 , J. Lagendijk 1 , B. Raaymakers 1 , M. Fast 1 1 UMC Utrecht, Radiotherapy, Utrecht, The Netherlands ; 2 Elekta AB, Elekta AB, Stockholm, Sweden ; 3 Modus QA, Modus QA, London, Canada Purpose or Objective With the introduction of hybrid MRI-linac systems, it has become possible to visualize moving anatomy using MRI during radiotherapy. MRI can provide high enough imaging frequencies, enabling real-time feedback control of the delivery beam for motion-compensated radiotherapy treatments. Recently, the technical feasibility of MRI- guided MLC-tracking on a clinical 1.5T MRI-linac was demonstrated, using a rigid phantom. There are some tumor types (e.g. lung tumors) that can show significant deformations due to intra-fraction motion. Here, we present first results of MLC-tracking of a deformable target using a prototype of a deformable motion phantom. Material and Methods All experiments were performed on the Unity MR-linac (Elekta AB, SWE), featuring 1.5T MR imaging and a 7MV linac. The radiation beam was dynamically shaped using the integrated 160-leaf MLC, moving in the cranial-caudal (CC) direction. The phantom (ModusQA, CAN) consisted of a cylindrical insert filled with a contrast solution and an open cell foam. Inside this foam was an ellipsoidal 30x40mm liquid-filled clearing representing the tumor. The foam was attached to a piston which was driven with a sinusoidal pattern (f = 0.17Hz, A = 10mm) in CC direction. 5mm hollow fiducials were present at the edges of the principal axes of the ellipsoid, which were visible on the MR and EPID images as dark spots (fig. 1). Online MRI was performed using a balanced 2D cine sequence (3x3x15mm, 4Hz), and the images were continuously forwarded to in-house software where the ellipsoidal foam target was automatically segmented using the floodfill algorithm of OpenCV v2.4. Subsequently, the initial contour was isotopically expanded by 8mm to contain the neighboring fiducials. This contour was then used to calculate new leaf positions (fig. 1), which were updated every 80ms. On continuously acquired EPID images (4Hz), the MLC aperture and fiducials were visualized, such that the phantom deformation could be correlated to the aperture deformation. Results From the EPID images, the amount of deformation (m1- m2, fig. 1) was estimated by comparing the expanded and compressed states: 9.4±1.9mm. The corresponding change in MLC aperture was estimated from the long axis change of a fitted ellipse: 10.0±1.4mm. The length change of the long axis of the MLC aperture was also compared to that of the target contour from the MR (fig. 2), resulting in a Pearson correlation of 0.73 and a RMSE of 2.1mm.

Results The evaluation of initial plans on repeated CTs showed large dose distortions, substantially reduced after restoration. In the analysis of local dose differences, median DE(vol=2%) were lowered from 9.2 Gy (1rCT) and 8.2 Gy (2rCT) in distorted plans to 3.1 Gy(1rCT) and 2.8 Gy (2rCT) in restored plans. Deviations respect to the reference dose in dose-volume parameters extracted from DVH-bands (nominal and worst-case values) shown general improvement of DVH metrics in restored plans (Figure 2). After DR, median D95 was increased by 1 Gy, keeping all restored plans within clinical limits, except one patient that would require off-line adaptation including contour deformation. According to CTV coverage criteria (D95%>95%Dprescription), 36% (5/14) and 23% (3/13) of the cases would have needed full adaptation in 1rCT and 2rCT, respectively, without DR being performed.

Conclusion Accurate and fast dose restoration on repeated CTs does not require new ROI segmentation nor deformation and is compatible with an online adaptive workflow. DR methods improve DVH-based scores in target volumes and organs at risk. The robustness of the initial plans is also propagated to restored plans. Off-line adaptation rate was drastically reduced. However, full adaptation will be required for