ESTRO 37 Abstract book

S99

ESTRO 37

Purpose or Objective MRI-Linacs enable real-time imaging for advanced in- room motion management in radiotherapy. However, the constraint of acquisition time and image quality suggests 2D cine-MRI centered in the tumor as state-of-the-art imaging for motion detection. Bridging the gap between 2D and 3D images is therefore required, with patient- specific motion models being the most viable solution. Standard global motion models are built using a single surrogate, thus assuming a linear correlation between surrogate and changing anatomy, independently from the anatomical location. To date, no global motion model based on 4DCT and cine-MRI images able to provide regional adaptation has been reported. In this work, we present a novel 4DCT global motion model based on Regions Of Interest (ROIs), aiming at accurately compensating for changes during treatment. Material and Methods 4DCT, 3DT1-weighted MRI and 2D sagittal cine-MRI data were generated with a digital CT/MRI phantom (resolution 1×1×1mm3) animated by patient-derived signals. In the simulated planning phase (9mm motion), a motion model was built on 3D motion fields derived from a 3D deformable image registration (DIR) between breathing phases in the 4DCT dataset. In this step, three regions of interest (ROIs) were defined (upper, middle and lower lung) and correlated with surrogate motion fields in correspondent ROIs. Surrogate motion fields were derived by a 2D DIR between sagittal CT slices centered in the tumor (reference 0% exhale: CTslice0%). In the simulated treatment phase (16mm motion), the model was corrected by 3DMRI to compensate for inter- fraction motion and applied to in-room cine-MRI data to compensate for intra-fraction motion. The application was performed in each ROI by deriving a pseudo cine-CT during treatment (i.e. the in-room surrogate motion fields were obtained by 2D DIR between CTslice0% and cine-MRI). FigureA shows the workflow of the ROI-based motion model. Analysis of the method was carried out by considering geometrical differences of target and diaphragm between the estimated 3DCT and the ground truth provided by the phantom. A comparison with a conventional motion model based on single surrogate (tumor or diaphragm) was also performed. Results For the tumor, the proposed method resulted in a difference (median-IQR) with respect to the ground truth of 0.87-0.59mm, against 0.97-0.48mm and 1.15-1.10mm when using tumor and diaphragm alone as surrogates. For the diaphragm, the ROI-based method resulted in an error of 0.69-0.93mm, whereas 2.82-2.09mm and 1.66- 2.11mm were measured for tumor and diaphragm, respectively. As a result, the ROI-based motion model was able to compensate on both tumor and diaphragm together (FigureB). Conclusion A novel ROI-based motion model was proposed with improved image guidance results with respect to conventional strategies. Future studies will rely on the application of the method to patient data and on a dosimetric evaluation to enable closed-loop adaptive radiotherapy.

0.17Hz and 15mm amplitude. The absolute position was recorded on a workstation with an accuracy of approximately 10ms and served as the gold standard. After real-time reconstruction, the MR images were streamed directly to the same workstation and given a timestamp. After acquisition the center of mass of the high-contrast object was estimated on each image frame and fit to a sinusoidal model. The same model was fit to the gold standard. From the phase difference between the two fits the apparent latency was calculated and compared to simulations.

Results Partial Fourier only has an influence on the latency when a ‘reverse linear’ sampling pattern is used, showing a linear relationship (fig. 2). The latency can be reduced further by a ‘high-low’ sampling pattern, in which case it is consistently 0.17s. A ‘linear’ sampling pattern results in a consistently high latency of 0.55s. These results are in agreement with simulations (not shown) and the notion that most image content is contained in the center of k- space. Therefore the time from sampling k=0 until the end of the acquisition is the dominant factor for the overall imaging latency (fig. 1).

Conclusion The MR acquisition can be a major contributor to the latency of the feedback chain, especially compared to the mechanical latency of the MLC. It is however controllable by choosing the sampling pattern in such a way that the central part of k-space is sampled latest. Although PF always increases the temporal resolution, it does not always decrease the imaging latency, as was observed for the ‘linear’ and ‘high-low’ sampling patterns. OC-0188 A ROI-based global motion model for MRI- guidance in radiation therapy: a phantom study N. Garau 1 , C. Paganelli 1 , G. Meschini 1 , R. Via 1 , M. Riboldi 2 , G. Baroni 1 1 Politecnico di Milano, Dipartimento di Elettronica- Informazione e Bioingegneria, Milano, Italy 2 Ludwig-Maximilians-Universität Münche n, Department of Medical Physics, Munchen, Germany

Made with FlippingBook - Online magazine maker