Abstract

Isak Wahlstedt¹, José David Tascon-Vidarte², Abraham George Smith³, Sune Darkner³, Signe Risum⁴, Mette Pøhl⁴, Kristian Boye⁴, Mirjana Josipovic⁴, Claus Flensted Behrens⁵, Ivan Richter Vogelius⁶

¹Tehnical University of Denmark, Department of Health Technology, Kongens Lyngby, Denmark; ²University of Copenhagen, Denmark, Department of Computer Science, Copenhagen, Denmark; ³University of Copenhagen, Department of Computer Science, Copenhagen, Denmark; ⁴Rigshospitalet, Department of Oncology, Copenhagen, Denmark; ⁵Herlev and Gentofte Hospital, Department of Oncology, Copenhagen, Denmark; ⁶University of Copenhagen, Department of Health and Medical Sciences, Copenhagen, Denmark

Magnetic resonance (MR)-guided online adaptive radiotherapy (MRgoART) allows the dose distribution to be tailored daily. Precise knowledge of the delivered doses to the target and critical organs at risk (OARs) in previous treatment fractions is important to ensure continuous adequate coverage of the target and OAR sparing. Today the task of evaluating dose delivered over multiple treatment fractions is handled by several algorithms performing dose warping based on deformable image registration (DIR). Potentially, this method can improve treatment monitoring and planning decisions made throughout the adaptive workflow and hence increase the therapeutic window. In order to validate dose accumulations conducted for patients treated with MRgoART, we investigated the reproducibility of dose accumulation using four different DIR algorithms on patient MR images at our institution.

The dose accumulations were performed for 5 consecutive patients with liver metastases referred to SBRT of 3 fractions (GTV mean dose ≤ 22.5 Gy, PTV covered by 67% isodose). The fraction doses were calculated on the daily patient anatomy and warped to the pre-treatment MR simulation scan using four DIR algorithms; ANTs, Elastix, MIM and Velocity. The reproducibility of dose accumulated with the different algorithms was compared with respect to GTV, PTV, and liver. All dose accumulations were compared to a dose accumulation performed with only affine transformations in ANTs and evaluated based on dose volume histogram (DVH) analysis.  All DVHs were calculated, based on structure sets and dose files, in dicompyler core. In order to quantify the DIRs behind the dose accumulations, Dice Similarity Coefficient and Hausdorff Distance were evaluated for the liver in 3D Slicer.

Differences in DVHs of the liver were negligible for all patients but patient 1 (Figure 1). For this patient, the entire liver was not encompassed in two of the MRIs, a task handled differently by the four algorithms. However, for some patients the differences in GTV and PTV DVHs were considerable. The differences between the affine algorithm and the DIR algorithms in GTV and PTV coverage was largest for patients 1-3 (Figure 2).Figure 1: GTV, PTV, and liver DVHs for the doses accumulated with the four DIR algorithms as well as with the affine algorithm. The labels are the same for all plots (see bottom right).

Figure 2: In each plot the accumulated doses with all five algorithms are evaluated. The clinical metric assessed for the healthy liver (liver structure with the GTV subtracted) is the volume receiving below 15 Gy. The labels are the same for all plots (see bottom right).

Comparison of accumulated DVHs suggests differences between the algorithms in the preliminary analysis. It may be necessary to use multiple accumulation algorithms to ensure failsafe operation when the treated volume is to be minimized in MRgoART.